Time Evolution Analysis of Riders’ Preference Attention and Satisfaction on Real-Time Crowdsourcing Logistics Platform

Abstract

This study examines the evolutionary trends in preference focus and satisfaction among real-time crowdsourced logistics platform riders (hereafter referred to as riders) in China since the outbreak of the COVID-19 pandemic. By analyzing real-time comments from 5,652 riders and applying the Latent Dirichlet Allocation (LDA) model to identify riders’ preferences and calculate their attention levels, combined with riders’ actual ratings to infer satisfaction levels (ranging from very dissatisfied to satisfied), this research for the first time explores the changing patterns of riders’ preferences and their attention and satisfaction trends in the pandemic context from a long-term dynamic perspective. The findings reveal that riders prioritize interaction quality; the range of fluctuation in average attention levels is large when high and small when low; attention to platform management, system utility, and rider relationships is on the rise, whereas interest in other preferences is declining; there is a significant correlation between rider attention and platform policies; and social media and related public opinion influence riders’ preferences. These insights are instrumental for the platform to adjust the policy direction duly and meet the core demands of riders according to the limited priority.

Plain language summary

How Riders’ Preferences and Satisfaction Evolve on Crowdsourcing Delivery Platforms

This study examines the evolutionary trends in preference focus and satisfaction among real-time crowdsourced logistics platform riders (hereafter referred to as riders) in China since the outbreak of the COVID-19 pandemic. By analyzing real-time comments from 5,652 riders and applying the Latent Dirichlet Allocation (LDA) model to identify riders’ preferences and calculate their attention levels, combined with riders’ actual ratings to infer satisfaction levels (ranging from very dissatisfied to satisfied), this research for the first time explores the changing patterns of riders’ preferences and their attention and satisfaction trends in the pandemic context from a long-term dynamic perspective. There is a significant correlation between rider attention and platform policies; and social media and related public opinion influence riders’ preferences. These insights are instrumental for the platform to adjust the policy direction duly and meet the core demands of riders according to the limited priority.

Keywords

time evolution riders’ preference attention riders’ preference satisfaction real-time crowdsourcing logistics platform text mining

Introduction

In 2021, Chinese ministry of social security announced that the number of employees in the gig economy has reached about 200 million. Among these, takeaway riders have emerged as a pivotal workforce within urban ecosystems. Yet, their development in Chinese cities encounters multifaceted constraints: absence of institutional safeguards, employment volatility, and heightened risks of COVID-19 infection. As pointed out in Roland Berger: The impact of the pandemic on China’s Transportation and Logistics Industry, the explosion of orders during the pandemic fully exposes problems like the lack of belonging of riders and the low adaptability of platform management.¹ Unreasonable platform measures, the discrimination from some consumers and businesses can lead to riders’ striking, conflict between riders and consumers, serious and frequent traffic accidents because of overspeed of riders. To satisfy riders’ preferences is an urgent problem to be solved.

But what are the preferences of Chinese riders? What is the long-term evolutionary trend of riders’ preferences attention? How does the satisfaction change about preferences? What are the causes of evolutionary fluctuations? How does the pandemic affect riders’ preferences? All these have not been received full attention.

Our study distinguishes itself from prior research by a dynamic, longitudinal approach. We employ text mining and evolutionary analysis to track shifts in riders’ preferences and satisfaction levels on China’s real-time crowdsourcing logistics platforms during the COVID-19 pandemic. Utilizing the Latent Dirichlet Allocation (LDA) model, we identify changes in riders’ preferences, revealing insights into their working conditions, adaptability to policy shifts, and social identities. This not only provides empirical backing for platform management strategies and policy development but also advances the static analytical framework of previous studies into a dynamic realm. Our enhancements to the LDA model account for both the frequency and significance of keywords, ensuring vital terms are not overlooked due to their commonality. By analyzing online review data from rider app stores, we categorize various preferences and quantify their attention and satisfaction levels. This enables a detailed examination of the evolution, variance, and underlying reasons behind different preference trends over time, offering a more accurate and comprehensive understanding of gig workers’ shifting landscapes.

Organization is as follows: Section “Literature Review”: reviews on the relevant literature if the crowdsourcing logistics, real-time crowdsourcing logistics platforms, the importance of preferences and satisfaction, LDA preference modeling, and evolutionary analysis. Section “Methodology”: empirical analysis methodology of riders’ preferences attention on real-time crowdsourcing logistics platform. Section “Evolution Analysis”: evolutionary analysis of riders’ satisfaction and attention on real-time crowdsourcing logistics. Section “Discussion” and Section “Conclusion”: Theoretical and practical significance.

Literature Review

The Transformative Role of Crowdsourcing in Logistics: Insights and Strategies

Crowdsourcing has emerged as a vital employment avenue for a significant portion of the workforce. To understand crowdsourcing logistics, it’s essential to recognize its role in engaging a broad base of individuals for task completion, facilitated by the internet’s connectivity (Coleman et al., 2017). Crowdsourcing has gained prominence in various sectors, including transportation, where it contributes to service improvement through widespread community involvement (Agboola & Tunay, 2023). In logistics, crowdsourcing is revolutionizing the industry by enhancing efficiency and customer satisfaction. Research by Y. Zhang et al. (2022) on riders’ satisfaction in real-time logistics platforms highlights the importance of aligning services with customer feedback. Further, Samad et al. (2023) and Nguyen et al. (2023) explore engagement incentives and data-driven compensation schemes, respectively, underscoring the significance of incentives and technology in crowdsourced delivery models. These studies illustrate crowdsourcing logistics’ potential to transform supply chain management by focusing on efficiency, sustainability, and innovative operational strategies. Future research should seek new advancements.

Real-Time Crowdsourced Logistics Platform Riders and the Determinants of Satisfaction

Research on real-time crowdsourcing logistics platform riders, although nascent, underscores the significant influence of riders’ satisfaction on platform management and competitive edge (Y. Zhang et al., 2022). As principal service providers using personal vehicles, riders’ sensitivity to wage levels plays a pivotal role in their platform allegiance, driven by wage comparisons and order volume (Castillo et al., 2017). This sensitivity spotlights the criticality of devising an effective pricing strategy to foster continued rider engagement and maintain the supply-demand equilibrium. Beyond monetary incentives, psychological and non-monetary factors—ranging from the pursuit of new experiences to the enjoyment found in task completion—also contribute to riders’ sustained participation (Mladenow et al., 2016; Samad et al., 2023). Emerging studies highlight the integral role of service quality in shaping user satisfaction on such platforms. For instance, investigations into JD Logistics’ service quality reveal the direct impact of response times and package handling on customer satisfaction (Yang, 2021). Additionally, the sustainability of crowdsourced logistics in SMEs exemplifies the model’s adaptability and effectiveness, affirming its viability (Agyemang, 2022). Methodological insights from data-driven decision-making and decision support systems elucidate the multifaceted considerations crucial in understanding and enhancing rider satisfaction (Baki, 2022; D. Liu et al., 2020). Moreover, the agility of data science in responding to global crises like COVID-19 offers fresh perspectives on real-time preference monitoring and analysis (Yu & Singh, 2022). Hence, this study ventures into the temporal dynamics of these preferences, aiming to provide both theoretical insight and practical guidelines to optimize platform operations and rider experiences.

LDA Topic Modeling and Evolutionary Analysis

The LDA is a probabilistic topic model that generates documents that are a mixture of many underlying topics, each topic is characterized by the probability distribution of the words (Blei et al., 2003). LDA can be effectively used to discover potential topics from a large collection of documents and output keyword combination. These keywords were collected naturally for a related topic (Guo et al., 2022). In previous studies, the potential Dirichlet distribution (LDA) topic model is used for product feature mining for online reviews (K. Chen et al., 2015; Wan et al., 2020), to make sure consumers’ interests (Sim et al., 2021), and employee job satisfaction (Jung & Suh, 2019).

Blei et al. (2003) used the DTM method to divide the text by time to process the corpus data for each time window one by one to complete an analysis of the evolution of preferences over time. AlSumait et al. (2008) used the Online LDA method to update the current model according to the new data streams, without requiring access to the previous data (AlSumait et al., 2008). This method supports detecting emerging preferences in real time. Wang and Xie (2021), combining LDA and sentiment to analyze the public response to COVID-19 in Weibo (W. Wang & Xie, 2021). The study found that people showed support for frontline doctors and nurses and care about financial life during the COVID-19.

Existing studies have provided an important reference for this paper. But the evolutionary trend of riders’ preferences attention and satisfaction of real-time crowdsourcing logistics platform has not been quantitatively investigated from a dynamic perspective. This study improves the previous LDA models’ attention to frequency and uses the weight and frequency of keywords to avoid important keywords being filtered due to high frequency and make the classification results more accurate. Based on the online comment data of the rider app Store, the LDA model text is used to classify different riders’ preferences and calculate their attention; riders’ preferences satisfaction was calculated using the rider’s score to the platform App in the App market. Ranking riders’ preferences by satisfaction and attention, and analyze the evolutionary characteristics, fluctuation amplitude, causes and trends of different riders’ preferences in different periods. Technical routes are shown in Figure 1.

Figure 1.

Research framework.

Methodology

Data Collection

Social media platform enables riders to express their opinions quickly with its open and immediate features, and textual data produced by this becomes an important source of information for the platform and relevant government departments to understand the needs of riders. At present, Chinese riders lack a dedicated and unobstructed complaint channel to express their real experience, and users’ comments and ratings on the app in the app store have become the main channel for their communication and complaints. Therefore, this article chooses app store comments to be more reliable (Y. Zhang et al., 2022). This study has selected the Apple App Store, Huawei App Store, OPPO App Store, and 360 Software Manager platform with comments and scoring function, and collected comments from riders on common real-time crowdsourcing logistics platforms such as UU errand Running Man-end, Rookie Points me, Dada Knight Edition, Hummingbird Crowdsourcing, Meituan Crowdsourcing, and SF City Knight APP (Other real-time crowdsourcing logistics platforms have found no comments in the app store from January 2020 to March 2022, with missing data). Among them, Takeout and Meituan accounted for 90% of the market share, their real-time crowdsourcing logistics platform Hummingbird crowdsourcing APP and Meituan crowdsourcing APP have covered most parts of China.

Finally, 5,878 comments data were obtained and collected. Each data includes comment date, comment, and rating, as shown in Table 1. Simple data period is from January 2020 to March 2022.

Table 1.

Format Example of Online Comment Data.

No.	Time	Contents	Rating
1	2020-01-14	Please ask the technician to make more efforts, the navigation distance of platform is really inaccurate! Can the location information be accurate to the number of buildings? It is more convenient to find the customer’s address in a new community than in an old one.	4
2	2020-03-31	My fine appeal was not approved by the platform. And I was also fined when encountered objective situations such as road closure and motorcycle failure.	1
3	2020-04-13	I clicked the “order grabbing” button in the lower right corner of the iPhone’s screen when the platform sent orders. Then there would be a prompt interface to confirm in the middle, so I could not grab the order! Android phones are different! Can’t the developers set it to be the same as the Android phone?	1

Data Preprocessing

Because the original rider comment data has problems like sparsity and more noise words, preprocessing of original data is needed to improve the credibility of the data analysis. (1) Delete or summarize the repeated evaluation, access only one of the data, and eliminate the symbols, facial expressions, letters, and random codes in the text. (2) Word segmentation and remove stop words, Jieba in Python Software, analysis of Chinese Library, is used for word segmentation and part-of-speech tagging to perform partiactions on each annotated data and convert it into a word list, combined with the use of “HIT Discontinuation Thesaurus Expansion Table” to filter stop words, and delete some words that does not have practical significance for the comment data analysis but appeared frequently such as “have or no,”“can only,”“influence,”“see” and so on. (3) By reviewing the initial findings, a manual exclusion list was created to identify more relevant preferences in the post-processing step. (4) Eliminate comments that are irrelevant to the rider and without analytical significance. It eventually received a total of 5,652 comments after noise reduction, delete meaningless words, eliminate irrelevant comments and other processing. The word frequency distribution after word segmentation is shown in Figure 2. It can be seen, the long tail effect (Y. Zhang et al., 2022) is evident in the rider comment data, which becomes the reason for choosing the improvement model in the next section.

Figure 2.

Word distribution of riders’ online comments.²

Topic Model

Feature Word Extraction

Calculating word frequency after rider comment, filtering the topic-irrelevant words, screening high-frequency words that represent topic of each comment (Nie et al., 2020), providing a corpus for the topic partitioning of the LDA models, with some high-frequency words are shown in Table 2, and using Python’s Wordcloud library to draw the word cloud. The color and size of words are correlated to the frequency of words. As shown in Figure 3.

Table 2.

Part of High Frequency Words of the Riders’ Comments.

Keywords	Frequency	Keywords	Frequency	Keywords	Frequency
Platform	685	Software	374	Apple	198
Fine	443	The order	363	Prompt	194
Consumers	433	Time	352	Address	185
List	413	The unit price	300	Equipment	184
System	398	Mobile phone	288	Voice	170
Customer service	386	The map	243	Deposit	154
Merchants	377	Positioning	241	Distance	154

Figure 3.

Wordcloud map of the high frequency.

The LDA Topic Model

LDA model selection to determine the optimal number of topics is the most challenging task, as it ultimately affects the annotations and results of the supervised classifier to some extent. (Wahid et al., 2022). This paper provides topic modeling of user-generated content of riders based on the Python toolkit SKlearn library, and adopts Gibbs sample, the number of sampling iterations was set to 300, and measures the change degree of confusion index by comparing the number of different topics to understand the change situation of degree of confusion index and coherence score for Figure 4.

Figure 4.

The trend of topics’ perplexity and coherence score of LDA model.

This article refers to Y. Zhang et al. (2022) and D. Li et al. (2020), the lower the degree of confusion, the better the clustering effect. The higher the coherence score, chances are that the more advantageous the words in each topic and better classification. As the chart shows, although the overall trend of degree of confusion is increasing, there are two obvious local optima, the topic numbers are 13 and 15. What is found by comparing different degree of topic coherence under different number of topics, like Figure 5, there is a clear downward trend in the scores when the topic number changes from 12 to 13, there is a clear upward trend in scores when the number of topics changes from 13 to 14, the coherence score of topics 13 has become the lowest (worst classification). Overall, when the number of topics is 15, there is a low degree of confusion and a high degree of coherence. It is more reasonable that the topic of the real-time crowdsourcing logistics platform rider topic is set to 15. The classification results are presented in Table 3. Due to the long-tail effect of keywords, this study will mainly analyze the large weight (more than 0.01) keywords.

Figure 5.

Part nodes of hierarchically cluster.

Table 3.

LDA Topic Model Classification Results.

No.	Topics	Keywords and weight
1	Training system	Equipment 184.01, deposit 154.01, training 57.01, activities 56.01, clothes 38.01, reward 38.01, offline 19.01, members 13.01, team 13.01, pay 9.01
2	Call center service	Customer service 386.01, telephone 82.68, background 68.31, labor 61.01, responsibility 22.01 overbearing 21.01, clause 19.01, reason 13.76, resolve problem 13.01
3	Delivery during pandemic period	Map 243.01, work 72.01, crowdsourcing 53.01, pandemic 40.01, mode 28.01, industry 12.01
4	Riders’ relationship	Life 27.01, brother 23.01, permission 15.01, order 11.01, place 10.01
5	Quality of interaction	Consumer 433.01, merchant 377.01, address 185.01, community 50.01, evaluation 9.82, content 9.53, pick-up 7.33
6	Usefulness of system	Version 145.01, route 104.01, setting 83.01, data 20.01, font 11.48, page 5.17, window pops 4.32
7	Punishment system	Fine 443.01, disorderly fine 40.82, reason 39.01, distribution 15.01, random delivery of orders 11.01, send the order 5.15
8	System ease of use	Prompt tone 194.01, message 25.01, headset 24.01, photo 24.01, Face recognition 23.01, volume 11.01
9	Platform management	System 398.01, price reduction 21.01, professional 15.01, detection 14.70, appraise 3.44
10	Features of order	Order 363.01, unit price 300.01, subsidy 49.01, weather 24.03, by-way 14.56, destination 13.78, distance 7.55
11	Information quality	Location 241.01, situation 50.01, opportunity 40.01, ID 37.01, comment 33.01, error 27.01
12	Prospective earnings	Deposit 94.01, salary 52.01, commission 26.01, dataflow 20.01, hard-earned money 19.01, finance 16.01
13	Equipment difference	Mobile 288.01, iPhone 198.01, voice 170.01, intelligent delivery of orders 22.01, fair 10.48
14	Delivery time	Time 352.01, advertising 37.01, cycling 14.36, adjustment 12.01, red light running 11.01, efficiency 1.01
15	Experience sharing	Distance 154.01, distribution fee 83.01, seal the account 79.01, suggestion 73.33, interface 40.01, management 33.0, rules 21.01, direction 19.71, region 18.01, speed 16.01, commission 16.01, priority 14.70, gap 6.01

Keyword information in the topics extracted from the dataset is relevant, the keywords shown in Figure 4 indicate that each topic contains specific information about different aspects. For instance, Topic 1, characterized by keywords such as “equipment,”“deposit,”“training,” and “activity” aligns with existing research on the relationship between training systems and efficiency in crowdsourced logistics platforms (Y. Zhang et al., 2022). Topic 2, highlighted by “customer service,”“phone,”“background,” and “manual work” reflects content related to customer service, resonating with literature on the impact of customer satisfaction on business performance (Yang, 2021). Topic 3, encapsulating “map,”“work,”“crowdsourcing,” and “pandemic” unveils delivery matters during the pandemic, illustrating the impact of the pandemic on the logistics industry (D. Li et al., 2020). Furthermore, Topic 4’s focus on the relationships between riders and Topic 5’s depiction of interactions between riders and businesses/consumers during the pandemic provide insights into platform management strategies, consistent with studies on the effect of social interaction on job satisfaction (Ullah et al., 2023). Topic 6, “ Usefulness of system” reveals operational insights with terms such as “version,”“settings,”“font,” and “popup” underscoring the critical role of system design and user interface in facilitating rider engagement and efficiency, aligning with studies on technology acceptance and user experience (Agyemang, 2022). Topic 7, “Punishment System,” includes keywords like “fine,”“disorderly fine,”“reason,” and “blind faction” highlighting the governance mechanisms within the platform. This aspect ties back to literature on organizational behavior and control systems, suggesting a direct impact on rider compliance and platform integrity (Khan et al., 2022). Topic 8, also categorized under “System ease of use” is marked by “cue tones,”“message,”“headphones,” etc. further demonstrating the importance of system alerts and communication tools in maintaining high operational standards and rider awareness, reflecting principles from human-computer interaction studies (Xia et al., 2018). Topic 9, “Platform Management” incorporates “system,”“price reduction,”“profession,” and “detection” illustrating the complex facets of platform operations management, resonating with supply chain management literature on the efficacy of strategic decisions in market positioning and competitive advantage (Shi et al., 2021). Topics 10 to 15 delve into “Features of order,”“Information quality,”“Prospective earnings,”“Equipment difference,”“Delivery time,”“Experience sharing,” each offering unique insights into logistics platform dynamics. These topics substantiate the multifaceted nature of platform ecosystems, as explored in logistics and supply chain transparency research (Khan et al., 2022).

Hierarchical Clustering Verification

Hierarchical clustering is a simple and effective unsupervised algorithm that can divide “n” data samples into “L” clusters automatically (Ding et al., 2023; Qi et al., 2021). Inputting the similar matrices into SPSS software for clustering analysis. In practice, the classification method is chosen for systematic classification, the clustered objects are variables. By using the cosine distance amount, work with a specified desired number of clusters (i.e., L) as the threshold and end the clustering when L clusters are obtained. At last, the classification results can be observed by deriving a spectral plot.

Figure 5 shows some nodes of the genealogy diagram, where tasks with similar attributes (i.e., keywords) are divided into different groups by hierarchical clustering repeatedly and also hierarchical cluster forms a tree structure on the right side of the diagram (X. Li et al., 2022). From Figure 6, it can easily be seen those keywords like “customer service, telephone, background, artificial, responsibility, force, clause, reason” all reflect the contradiction between the rider and the platform customer service, which is consistent with topic 1 (i.e., customer service). Also, the keywords “consumer, merchant, address, community, evaluation, content, pick-up and so on” express the interaction between riders and consumers, which is consistent with topic 5 of the LDA model (i.e., interactive quality content). Therefore, the hierarchical clustering model is basically consistent with the classification results of the LDA model, indicating that the LDA method is relatively accurate.

Figure 6.

Cumulative attention trends in riders’ preferences.

Calculate the Level of Concern

In this section, the paper draws on Y. Wang et al. (2022) approach to analyze the evolutionary characteristics of these 15 topics in order to capture changes in rider concerns over a three-month time interval (Y. Wang et al., 2022). First, the review data after Section “The LDA Topic Model” are divided by time interval. Then, equation (1) is used to calculate the rider’s attention to each topic at different times.

A D_{z t} = \sum_{i = 1}^{m_{z}} f_{z i t} \times w_{k i z} z = 1, 2, \dots, 15

(1)

In equation (1), $A D_{z t}$ is the consumer’s attention to topic $z$ in period $t$ , $f_{z i t}$ is the frequency of occurrence of keyword $i$ in the topic $z$ in period $t$ , $w_{z i}$ is the weight of keyword $i$ in the topic $k$ obtained from LDA topic modelling. The details of the table of word frequency are shown in Supplemental Appendix 1. The results of the calculation are shown in Table 4.

Table 4.

The Attention (Ranking) of Riders’ Preferences from 2020.01 to 2022.03.

Topic	2020.03	2020.06	2020.09	2020.12	2021.03	2021.06	2021.09	2021.12	2022.03
Quality of interaction	25,852.29 (1)	32,752.77 (1)	29,890.68 (1)	28,819.16 (1)	33,830.96 (1)	26,882.13 (1)	29,904.17 (1)	21,832.3 (1)	16,487.6 (1)
Punishment system	5,798.14 (6)	22,104.81 (2)	28,343.93 (2)	17,078.07 (4)	15,583.37 (2)	15,207.2 (2)	11,705.74 (4)	17,839.06 (2)	8,135.64 (3)
Features of order	16,194.46 (2)	11,262.22 (5)	17,264.86 (3)	14,948.31 (6)	6,307.79 (8)	14,526.67 (3)	13,140 (2)	15,340.81 (3)	11,762.09 (2)
Call center service	9,607.97 (4)	17,758.52 (4)	10,962.66 (7)	21,257.1 (2)	12,195.17 (3)	6,680.96 (7)	12,842.23 (3)	10,363.19 (5)	4,855.55 (6)
Equipment difference	11,379.42 (3)	20,972.97 (3)	11,982.3 (5)	15,744.49 (5)	8,278.37 (6)	8,136.86 (5)	11,000.48 (5)	14,420.39 (4)	4,367.55 (8)
Platform management	4,001.11 (8)	2,490.82 (13)	12,426.82 (4)	17,554.46 (3)	11,281.74 (4)	7,974.9 (6)	8,759.66 (6)	9,637.17 (7)	5,210.15 (5)
Delivery time	4,339.87 (7)	10,682.34 (6)	11,425.74 (6)	6,955.29 (8)	9,285.31 (5)	9,238.29 (4)	6,058.19 (7)	9,953.32 (6)	8,110.59 (4)
Training system	6,240.52 (5)	8,678.68 (7)	8,189.68 (8)	5,268.43 (10)	4,002.4 (11)	5,986.52 (8)	5,441.55 (8)	2,676.22 (11)	4,419.34 (7)
Delivery during pandemic period	3,438.22 (9)	5,156.34 (9)	6,103.35 (9)	6,098.33 (9)	7,206.43 (7)	4,029.22 (9)	4,455.27 (10)	6,073.35 (9)	3,251.21 (9)
Information quality	3,035.27 (11)	7,731.49 (8)	3,057.28 (11)	2,272.22 (13)	5,630.31 (9)	1,633.11 (13)	4,695.27 (9)	6,229.3 (8)	2,627.15 (11)
System ease of use	2,151.19 (12)	4,107.3 (10)	4,078.28 (10)	7,496.52 (7)	4,762.29 (10)	2,423.16 (11)	1,307.12 (12)	1,088.11 (13)	617.05 (13)
Experience sharing	3,392.69 (10)	3,629.5 (11)	2,389.92 (12)	2,903.91 (12)	3,230.09 (12)	2,780.7 (10)	1,012.45 (13)	3,057.9 (10)	2,395.92 (12)
Usefulness of system	1,309.09 (13)	2,845.22 (12)	2,325.53 (13)	3,928.68 (11)	1,980.67 (13)	1,907.19 (12)	2,466.19 (11)	2,183.36 (12)	2,683.69 (10)
Prospective earnings	1,247.18 (14)	1,178.19 (14)	1,542.21 (14)	1,397.17 (14)	204.07 (14)	454.06 (14)	812.14 (14)	737.14 (14)	588.13 (14)
Riders’ relationship	53.03 (15)	77.03 (15)	0 (15)	50.02 (15)	137.07 (15)	81.03 (15)	119.05 (15)	102.05 (15)	131.05 (15)

Evolution Analysis

Evolution Characteristics of Riders’ Attention

The paper calculates the cumulative preference of each rider from January 2020 to March 2022 and make a statistical chart to analyze the time evolution characteristics of riders’ preferences. Table 5 is converted into Figure 7 for ease of understanding. The results are shown in Table 6 and Figure 6. It is easy to find that the most concerned preferences of riders are quality of interaction, punishment system, features of order, call center service and equipment difference. According to text analysis, their basic meanings are followed by the punishment system of the platform, interaction with consumers, whether the order is the type they are familiar with, whether the communication of customer service staff is effective and the impact of differences in communication equipment on order grabbing. However, riders often pay less attention to their own income which indicates that during the pandemic period, riders, as front-line employees, are willing to bear greater risks to maintain the normal operation of social logistics (Friedland & Balkin, 2023).

Table 5.

Descriptive Statistical Characteristics of the Attention of Riders’ Preferences.

No.	Riders’ preferences	Sum	Max	Min	Average	Range	SD
1	Punishment system	141,795.82	28,343.93	5,798.00	15,755.09	22,545.93	6,487.75
2	Quality of interaction	246,252.06	33,830.96	16,487.60	27,361.34	17,343.36	5,134.07
3	Call center service	106,523.38	21,257.10	4,855.55	11,835.93	16,401.55	4,809.39
4	Equipment difference	106,282.41	20,972.97	4,367.55	11,809.16	16,605.42	4,576.26
5	Platform management	79,336.72	17,554.46	2,490.82	8,815.19	15,063.64	4,385.05
6	Features of order	120,746.75	17,264.86	6,307.79	13,416.31	10,957.07	3,123.29
7	Delivery time	76,049.07	11,425.74	4,340.00	8,449.90	7,085.74	2,169.99
8	System ease of use	28,030.83	7,496.52	617.05	3,114.54	6,879.47	2,077.61
9	Information quality	36,911.13	7,731.49	1,633.11	4,101.24	6,098.38	1,950.35
10	Training system	50,903.82	8,678.68	2,676.22	5,655.98	6,002.46	1,805.75
11	Delivery during pandemic period	45,811.50	7,206.43	3,251.21	5,090.17	3,955.22	1,296.69
12	Experience sharing	24,793.39	3,629.50	1,012.45	2,754.82	2,617.05	731.12
13	Usefulness of system	21,629.53	3,928.68	1,309.00	2,403.28	2,619.68	689.71
14	Prospective earnings	8,160.11	1,542.21	204.07	906.68	1,338.14	430.98
15	Riders’ relationship	750.30	137.07	0.00	83.37	137.07	41.91

Figure 7.

Fluctuation of riders’ preferences.

Table 6.

The Trend Characteristics of the Attention of Riders’ Preferences.

Trend	Preferences	Terminal growth amount
Upward	Platform management	1,209.04
	Delivery time	3,770.71
	Usefulness of system	1,374.61
	Riders’ relationship	78.02
Downward	Equipment difference	−4,752.43
	Call center service	−7,011.87
	Punishment system	2,337.50
	Quality of interaction	−9,364.69
	Delivery during pandemic period	−187.01
	Training system	−1,821.18
	System ease of use	−1,534.14
	Features of order	−4,432.37
	Experience sharing	−996.77
	Prospective earnings	−659.05
	Information quality	−408.12

In the three months from January 2020 to March 2022, the cumulative value, maximum value, minimum value, mean value and standard deviation of the 15 riders’ preferences were obtained. What’s more, as shown in Table 6, riders’ preferences are ranked by standard deviation. In this paper, the fluctuation of attention is represented by the standard deviation. It is found for the first time that riders with higher attention tend to have higher fluctuation value. For example, the average value of attention of punishment system is 15,755.11, the standard deviation is 6,487.72, the average value of attention of interaction quality is 27,361.34 and the standard deviation is 5,134.07. However, the preference with low attention generally has lower fluctuation value, the average value of attention on expected earnings is 8,160.11, the standard deviation is 430.98, the average value of attention on riders’ relationship is 750.33 and the standard deviation is 41.91. The following will specifically analyze riders’ preferences of the top three in terms of fluctuation size (i.e., punishment system, quality of interaction, call center service) and the five preferences with the lowest fluctuation size (i.e., delivery during pandemic period, experience sharing, usefulness of system, prospective earnings, and riders’ relationship).

Highly Volatile Preferences

In the graph, horizontal coordinates 1 to 9 indicate the time intervals 1-2020.03, 2-2020.06, 3-2020.09, 4-2020.12, 5-2021.03, 6-2021.06, 7-2021.09, 8-2021.12, and 9-2022.03. The same applies below. According to the fluctuation of riders’ preferences concern, the preference can be divided into two-peaked and multi-peaked types. The analysis in this paper follows the following three steps: first, describe the fluctuation and trend of riders’ preferences. Then focus on inflection points for preference concern fluctuations.

Punishment System

It can be seen from Figure 7 that the punishment system belongs to the two-peaked type, with an overall downward trend. As can be seen from Table 6 that riders’ concern about the “punishment system” ranks second, with an average value of 15,755.09, the standard deviation is 6,487.75, with the largest fluctuation, the range is 22,545.93, ranking the first. Therefore, it can be deduced from the survey data that the maximum peak of “punishment system” is 2020.09 and the trough is 2021.09.

In September 2020, the peak of “punishment system” attention was typical. In 2020, most of the instant crowdsourcing logistics platforms are still in the stage of development. To obtain more benefits, the instant delivery time is reduced constantly. In this regard, the government departments of various regions in the country have made corresponding laws to regulate riders’ safe driving behaviour.³ If the rider works overtime, it means bad comments, fines and even dismissal. Riders must drive at excessive speeds to meet the time, which put their lives at risk. This behavior has caused heated debate in the public.⁴ Under the pressure of public, the platform proposed measures to increase customers’ waiting time for riders, with the platform “Meituan” proposing to “give riders 8 min of extra time to deliver”.⁵ Also, the new function of “I’m willing to wait for another 5 min or 10 min” was launched on the platform “Elema”.⁶ On April 18, 2021, the platform “Meituan” introduced the responsible return rules, which stipulated that riders would only be fined if they refused the responsible order more than 12 times a day.⁷ If the rider refuses the order for the first time, the first refusal will be penalty-free.⁸ The responsible return rules provide more opportunities for riders to choose orders and make riders pay less attention to the “punishment system.” On July 8, 2021, a rider posted a video on TikTok, BiliBili and some other platforms about the unreasonable judgment of the instant crowdsourcing logistics platform’s unreasonable penalties for overtime, meal damage and complaints against riders.^9,10 This video received 3,512 likes and 1,383 comments, which increased the attention of the “punishment system.” The number of high likes also shows that these phenomena are common among the rider groups. Therefore, it is not difficult to find that social media and public opinion have an impact on riders’ preferences.

Quality of Interaction

It can be seen from Table 5 that the “quality of interaction” has the highest ranking of attention per period, with a standard deviation of 5,134.07, ranking the second in fluctuation and the second highest extreme deviation of 17,343.36. The most obvious peak of attention to “quality of interaction” appears in September 2021.

The “quality of interaction” is reflected in consumers expressing their feelings about products or services by transmitting information. The main keywords are “consumer,”“merchant,”“address,”“community,”“telephone,”“evaluation,”“content,” and “pick up.” On June 25, 2021, the public opinion analysis report released by the China Consumer Association on 20 days (from June 1 to 20) of tracking consumer rights protection information pointed out that there were 63,043 negative messages about instant logistics such as home delivery and overtime.¹¹ These consumer complaints led to high fines for riders directly.¹² In July of the same year, for the safety of riders, many consumers called for less takeout orders and rush orders on Typhoon days. Also, the platform suspended the delivery in risk areas such as Shanghai, Zhoushan and Ningbo.¹³ This paper speculates that there may have a relationship between the increase of riders’ attention to “quality of interaction” and public opinions online. The author further analyzes that any public opinion on social media may have a butterfly effect. Also, relevant authorities and news media have their own purposes and intentions for the news and short videos they publish.

Call Center Service

Table 6 shows that the average value of riders’ attention to “call center service” is 11,835.93, ranking the fourth among the 15 preferences, with a standard deviation of 4,809.39, ranking the third in the degree of fluctuation, and a range of 16,401.55, ranking the fourth.

Customer service is the way for riders to contact the platform. The main keywords of “call center service” include “customer service,”“telephone,” and “problem solving.” It can be found that the problems caused by “call center service” are mainly whether the customer service personnel can help the rider solve the emergency and the attribution of the complaint responsibility. In July 2020, a rider released a communication recording with Meituan customer service on the TikTok platform, in which he was fined for exceeding the time limit by 25 s due to bad weather. The customer service informed him that he could not appeal because the fact that he had exceeded the time limit, resulting in bad consequences.¹⁴ This kind of unfair event quickly generated empathy among the rider groups and caused the increase of attention. In real life, accidents in the delivery process and fines at the end of delivery are the main contents of customer service.^9–15 The Supervision and Administration of Online Catering Services announced by the State Administration of Market Supervision and Administration¹⁶ specifies the obligations of platform service riders and the delivery requirements for riders. It indicates that the rider can contact the customer service for emergency assistance and the customer service should respond and aid in a timely manner.

Analysis of Rising and Falling Trends

This study measures the magnitude of trend change in terms of the difference in attention at the end of the increase, that is, 2022.03 versus 2020.03. As shown in Table 6.

As can be seen from Table 6, riders’ preferences for platform management, delivery time, system usefulness, and rider relationship showed an overall upward trend, with a change in concern between 2022.03 and 2020.03 of 1,209.04, 1,374.61, and 78.02. Riders’ preferences for equipment differences, customer service and so on showed a decreasing trend overall.

Evolutionary Analysis of Riders’ Preferences Satisfaction

There are six rider ratings, that is, 0, 1, 2, 3, 4, and 5, which indicates rider satisfaction according to the actual ratings, that is, [0, 1) means very dissatisfied, [1, 2) means relatively dissatisfied, [2, 3) means neutral, [3, 4) means relatively satisfied and [4, 5] means very satisfied, and the satisfaction of each preference rider was calculated for each time interval during 2020.01 to 2022.03 The arithmetic mean, overall satisfaction mean and standard deviation, ranked the preferences according to satisfaction criteria as shown in Table 8. The distribution of riders’ preferences satisfaction (hereinafter: preference satisfaction) is all in (1, 4), while the overall satisfaction arithmetic mean is distributed in (1, 3).

As can be seen from the thermodynamic diagrams of satisfaction in Figure 8, blue indicates a positive attitude of neutral to satisfied, and the darker the blue, the higher the satisfaction; red indicates a negative attitude of dissatisfaction, and the darker the red, the higher the dissatisfaction. The thermodynamic diagrams can clearly see the trend of attitude evolution of riders’ preferences.

Figure 8.

Thermodynamic diagrams of preferences’ satisfaction.

During the period from January 2020 to March 2022 with a time interval of 3 months, the mean value of 15 riders’ preferences satisfaction was calculated and ranked to obtain Table 7, and the mean value, minimum, maximum, and standard deviation were calculated for 2 years, and riders’ preferences were ranked by the size of standard deviation as shown in Table 8. In line with the idea of 4.1, this paper indicates the fluctuation of satisfaction by standard deviation. From Table 8, the preferences with large fluctuation in specific analysis are platform management, rider relationship, system ease of use, information quality, delivery time and the preferences with moderate fluctuation are order grabbing attributes, penalty system, equipment difference, expected revenue and customer service. This study analyzes the evolutionary trends from both transversal and longitudinal perspectives. Transversal is the change of satisfaction for the same preference over time. Longitudinal is the change of satisfaction of different preferences in the same period.

Table 7.

Average of Preference’ Satisfaction (Ranking) from 2020.01 to 2022.03.

No.	Preference	1	2	3	4	5	6	7	8	9
1	Delivery during pandemic period	3.23 (2)	2.125 (2)	2.379 (1)	3 (1)	3.105 (1)	3.296 (1)	2.556 (8)	3.028 (3)	2.583 (3)
2	Delivery time	3.07 (3)	2.036 (5)	1.886 (5)	2 (9)	2.818 (3)	3.25 (2)	2.7 (6)	3.227 (2)	2.5 (5)
3	Training system	2.56 (8)	1.887 (7)	1.84 (7)	2.509 (2)	3.102 (2)	2.696 (3)	2.949 (3)	2.764 (5)	2.541 (4)
4	Riders’ relationship	1.26 (13)	1.868 (8)	1.769 (10)	2.375 (3)	2.744 (4)	1.933 (11)	3.467 (1)	2.897 (4)	2.083 (9)
5	Information quality	2.57 (6)	1.543 (12)	1.409 (14)	2.174 (5)	2.586 (6)	2.417 (5)	3.12 (2)	2.593 (6)	3.188 (2)
6	Equipment difference	2.58 (5)	2.236 (1)	1.622 (11)	2.063 (8)	2.308 (10)	2.692 (4)	2.333 (11)	2.425 (7)	2.056 (11)
7	System ease of use	2.64 (4)	2.022 (6)	2.364 (2)	1.667 (12)	2.414 (8)	1.375 (15)	2.84 (4)	1.6 (14)	3.417 (1)
8	Usefulness of system	3.31 (1)	2.074 (3)	2.095 (4)	2.138 (6)	2.154 (12)	2.056 (8)	2.714 (5)	1.739 (11)	1.722 (13)
9	Quality of interaction	1.63(11)	2.05 (4)	1.865 (6)	1.796 (10)	2.62 (5)	1.935 (10)	2.514 (10)	2.218 (8)	2.118 (8)
10	Platform management	1.5 (12)	1.608 (10)	1.415 (13)	2.191 (4)	2.514 (7)	2.158 (7)	2.541 (9)	3.967 (1)	2.421 (6)
11	Features of order	2.12 (9)	1.7 (9)	1.83 (8)	1.64 (13)	2.412 (9)	1.947 (9)	2.641 (7)	2.118 (9)	1.913 (12)
12	Experience sharing	2.57 (7)	1.59 (11)	2.136 (3)	1.708 (11)	1.63 (14)	2.192 (6)	1.462 (15)	1.682 (13)	2.381 (7)
13	Call center service	2.08 (10)	1.531 (13)	1.818 (9)	1.489 (15)	1.75 (13)	1.389 (14)	1.833 (13)	1.704 (12)	2.077 (10)
14	Prospective earnings	1.25 (14)	1.469 (14)	1.469 (12)	2.08 (7)	1.529 (15)	1.684 (12)	1.897 (12)	1.407 (15)	1.133 (14)
15	Punishment system	1.23 (15)	1.286 (15)	1.36 (15)	1.5 (14)	2.179 (11)	1.44 (13)	1.656 (14)	1.765 (10)	1.111 (15)

Note. The first line 1 to 9 represents the time, that is, 1-2020.03, 2-2020.06, 3-2020.09, 4-2020.12, 5-2021.03, 6-2021.06, 7-2021.09, 8-2021.12, and 9-2022.03.

Table 8.

Descriptive Statistical Characteristics of Average of Preferences’ Satisfaction.

Preference	Average	Max	Min	SD
Platform management	2.036	3.967	1.415	0.731
Riders’ relationship	2.33	3.467	1.261	0.636
System ease of use	2.199	3.417	1.375	0.618
Information quality	2.303	3.188	1.409	0.578
Delivery time	2.524	3.25	1.886	0.504
Usefulness of system	2.192	3.308	1.722	0.469
Training system	2.522	3.102	1.84	0.405
Delivery during pandemic period	2.808	3.296	2.125	0.387
Experience sharing	1.917	2.565	1.462	0.373
Quality of interaction	2.084	2.62	1.629	0.308
Features of order	1.95	2.641	1.64	0.308
Punishment system	1.518	2.179	1.111	0.307
Equipment difference	2.233	2.692	1.622	0.3
Prospective earnings	1.569	2.08	1.133	0.282
Call center service	1.701	2.08	1.389	0.23

Analysis of Satisfaction Fluctuations

Platform Management

The mean value of riders’ satisfaction with “platform management” is 2.036, ranking 10th and in a neutral attitude; it fluctuates the most, with a standard deviation of 0.731. As can be seen from the thermodynamic diagrams in Figure 8, the satisfaction with “platform management” changes from red to blue. The overall satisfaction trend is rising. The fluctuation trend of “platform management” belongs to single-peak type, the highest peak is 2021.12, which is also the highest point of satisfaction, reaching 3.967, and belongs to the quite satisfied attitude. In December 2021, the Ministry of Human Resources and Social Security issued a standard for rider skills, so that change the rider from “odd job” to “professional.”“This is not only the need for policy protection, but also the need for the platform to gradually improve the system, the Meituan and Ele. me platform are to provide relevant courses and training to ensure that riders meet the knowledge and skills required.^17,18

Longitudinally, from 2020.03 to 2020.09, the satisfaction ranking of “platform management” is in the bottom five, with the lowest ranking in 2020.09, in 13th place, only above “penalty system” and “Information quality.” From 2020.09 to 2020.12, the satisfaction ranking of “platform management” jumped from the 13th to the 4th place, only lower than “pandemic distribution,”“training system,” and “rider relationship,”“Rider relationship.” Then the ranking of “platform management” dropped, and from 2020.12 to 2022.12, the ranking of “platform management” rose from ninth to first, the highest ranking. By 2022.06, the ranking of “platform management” dropped again to sixth, which was lower than “ease of use of system,”“quality of information” and so on.

Rider Relationship

The mean value of riders’ satisfaction with “rider relationship” is 2.33, ranking the fourth, which is a neutral attitude; the standard deviation of satisfaction within 2 years is 0.636, ranking the second in terms of fluctuation; Figure 8 thermodynamic diagrams shows that the satisfaction with “rider relationship” changes from the overall trend of satisfaction is upwards. “The fluctuation trend of “rider relationship” is multi-peaked, and the most obvious wave trough is 2021.06, which is also the turning point of satisfaction, that is, from relatively unsatisfied to neutral. In June 2021, it was reported that young people were more willing to choose “deliver the take-out” than “enter the factory.”¹⁹ The rider industry attracts employment, and the increase in the number of employees makes the competition for riders more intense, and satisfaction with the “rider relationship” decreases. Before September 2020, information disclosure platforms emphasize the need to maintain fair competition for platform riders, and to hold riders who use cheating programs legally responsible.¹³ Government measures to create a fair environment for riders to grab orders. The satisfaction of riders increased.

Longitudinally, 2020.03 the satisfaction ranking of “rider relationship” was the lowest, at 13th place, just above “expected benefits” and “penalty system.” From 2021.06 to 2021.09, the satisfaction ranking of “rider relationship” jumped from the 11th to the 1st place, with the highest satisfaction ranking. From 2021.09 to 2022.03, the satisfaction ranking of “rider relationship” dropped from the first place to the ninth place, and the satisfaction ranking was lower than “System ease of use,”“quality of information,” and “pandemic delivery,”“training system,”“delivery time,”“platform management,”“experience sharing,” and “experience sharing.”

System Ease of Use

The “system ease of use” is based on Xia et al. (2018) study, which concluded that “perceived ease of use” reflects the user’s belief that using the system will be effortless, including easy to operate the system, clear pages, and easy troubleshooting. The mean value of riders’ satisfaction with “system ease of use” is 2.199, ranking seventh and in a neutral attitude; the standard deviation of satisfaction is 0.618, ranking third in terms of fluctuation; the thermodynamic diagrams in Figure 8 shows that the satisfaction with “system ease of use” changes from the overall satisfaction trend is rising. The fluctuation trend of “system ease of use” belongs to multi-peak type, and the time interval when the rider’s attitude changes more obviously is 2021.06, and the satisfaction level changes from neutral to relatively unsatisfied. Instant crowdsourcing logistics platform ease of use for riders to save delivery time, before June 2021 riders in the TikTok platform summary Meituan and Ele. me platform use in the display interface, system dispatch and other operational problems, 591 likes.²⁰ Such problems platform cannot be resolved in a timely manner will also affect rider satisfaction. Longitudinally, from 2020.03 to 2021.06 “system ease of use” satisfaction ranking from 8th to 15th, that is, the last, lower than other preferences. From 2021.12 to 2022.03, the satisfaction ranking of “ease of use of the system” increased from 14th to 1st, which is higher than other preferences.

Preference Satisfaction Trend Analysis

The riders with decreased satisfaction have fewer preferences, a total of 3, namely experience sharing, expected benefits, and system availability, and the riders with decreased satisfaction have more preferences, a total of 12. Among them, the riders whose satisfaction decreased are relatively dissatisfied. Riders’ preferences with increased satisfaction were 8 neutral and 4 negatives.

Table 9 shows that riders with a neutral attitude prefer pandemic distribution, distribution time, training system, rider relationship, information quality, equipment differences, system ease of use, interaction quality, and platform management. It can be seen that the satisfaction of platform management, rider relationship and system ease of use has increased significantly. Riders with a negative attitude prefer the attributes of order grabbing, experience sharing, expected income, and information quality on the whole showing an upward trend. Downtrend riders’ preferences are all negative attitudes. The largest change was system availability.

Table 9.

The Trend Characteristics of the Average of Preference’ Satisfaction.

Trend	Preference	Average	Terminal growth amount
Upward	Delivery during pandemic period	2.808	−0.643
	Delivery time	2.524	−0.567
	Training system	2.522	−0.021
	Riders’ relationship	2.33	0.822
	Information quality	2.303	0.617
	Equipment difference	2.233	−0.521
	System ease of use	2.199	0.781
	Quality of interaction	2.084	0.489
	Platform management	2.036	0.921
	Features of order	1.95	−0.205
	Experience sharing	1.917	−0.184
	Call center service	1.701	−0.003
Downward	Usefulness of system	1.701	−1.586
	Prospective earnings	1.569	−0.117
	Punishment system	1.518	−0.12

Attention-Satisfaction Fluctuation Analysis

This study will combine fluctuations in attention (standard deviation of attention) and fluctuations in satisfaction (standard deviation of satisfaction), while analyzing the fluctuations in riders’ preferences. For the convenience of observation, in this study, the abscissa represents the fluctuation of attention, and it is placed in the position where the fluctuation of satisfaction is 0.4. The ordinate represents the fluctuation of satisfaction, and it is placed at a position where the fluctuation of attention is 3,000. The larger the ordinate, the greater the satisfaction, and the more easily the rider’s preference satisfaction is affected. Establish a Cartesian coordinate system, as shown in Figure 9. This study refers to the research of M.-C. Chen et al. (2021) and divides riders’ preferences into necessary, one-dimensional, temporarily irrelevant and attractive riders’ preferences according to quadrants (M.-C. Chen et al., 2021).

(1) The first quadrant: The first quadrant is a preference with large fluctuations in attention and satisfaction, and only includes platform management. The influence of platform measures, government policies and public opinion, such preferences are necessary riders’ preferences and should be satisfied first. Slow system review, ineffective plug-in detection, and lack of professional guidance are the daily concerns of riders. Timely improvement of these problems will help improve the satisfaction of “platform management” and reduce attention.

(2) The second quadrant: The second quadrant is the preference with small fluctuation in attention and large fluctuation in satisfaction, including riders’ preferences including delivery time, system usefulness, system ease of use, information quality, training system, Rider relationship. In this quadrant, the attention of riders’ preference is not easily affected, and their satisfaction is easily affected, sorting should be placed after the first quadrant. Such preferences are one-dimensional (satisfaction) riders’ preferences, and the absence of these preferences can lead to rider dissatisfaction. Too little delivery time, whether the system functions are needed, whether the system functions are labor-saving, whether the system information is understandable, specific and reliable, whether the training content is detailed, and the competitive relationship with other riders will seriously affect rider satisfaction. For most riders, the demands on the system capabilities, training and competitive environment provided by the platform need to be met. The second quadrant ordering should be placed after the first quadrant.

(3) The third quadrant: The third quadrant is the preference with less fluctuation in attention and satisfaction, including riders’ preferences including pandemic distribution, experience sharing, and expected income. In this quadrant, both attention and satisfaction of riders’ preference are less susceptible. The satisfaction and lack of these rider preferences will not cause much change in rider satisfaction and attention in a short period of time and belong to temporarily irrelevant rider preferences. Therefore, from a long-term perspective, the distribution situation during the pandemic, the sharing of distribution experience, and the expected benefits may have a greater chance of affecting riders’ attention and satisfaction. Platforms should strategically integrate such preferences with other preferences to enhance platform attractiveness.

(4) The fourth quadrant: The third quadrant is a preference with large fluctuations in attention and small fluctuation in satisfaction, including riders’ preferences including punishment system, customer service, equipment differences, order grabbing attributes, and interaction quality. In this quadrant, the satisfaction of riders’ preference is not easily affected, and the degree of attention is easily affected. Therefore, the platform punishment system, customer service response services, differences in rider equipment, order grabbing attributes, and interactions with consumers will affect the attention of riders’ preferences.

Figure 9.

Rectangular coordinate system of attention-satisfaction.

Discussion

This study attempts to track the evolution trend of the attention and satisfaction of riders’ preferences on an instant crowdsourcing logistics platform over a long period of time.

Theoretical Contributions

Dynamics of Attention and Satisfaction

The long-term data analysis reveals that gig workers’ preference attention and satisfaction are not static but evolve dynamically over time in response to changing market conditions, platform policies, and public opinion. This finding offering a new perspective for understanding the behaviors of gig workers in the crowdsourced logistics sector.

Refined Classification of Preferences

Punishment system, customer service, equipment differences, platform management, and preference for grabbing orders are the preferences that riders of Chinese timely crowdsourcing logistics platforms are most concerned about. The preference of a rider with a high average attention degree generally has a larger fluctuation range; the preference of a rider with a low average attention degree generally has a small fluctuation range. There is an exception preference, the average value of interaction quality attention is high, and the fluctuation range is small. The attention of platform management, system usefulness and rider relationship are on the rise, and the attention of the rest of the preferences is a downward trend. At present, it seems that the news related to the rider’s preference with high attention is negative. On the other hand, the platform should strengthen its improvement.

Constructing a Theoretical Framework for the Evolution of Crowdsourced Logistics Gig Worker Preferences

This framework not only reveals the temporal trends of preference changes but also highlights the interconnectedness between different categories of preferences.

The study uncovers the diversity and complexity of gig worker preferences. This refined classification promotes a deeper understanding of the factors influencing gig worker satisfaction. Furthermore, the analysis of correlations between different categories of preferences provides a basis for platforms to optimize gig worker management strategies and enhance their efficiency and satisfaction.

Practical Implications

This paper divides riders’ preference into necessary riders’ preference (platform management); one-dimensional riders’ preference: delivery time, system usefulness, system ease of use, information quality, training system, rider relationship; temporarily irrelevant The preferred riders’ preferences are pandemic distribution, experience sharing, and expected benefits; attractive riders’ preferences include punishment system, customer service, equipment differences, order grabbing attributes, and interaction quality. Necessary riders’ preferences should be met first. One Dimensional riders’ preferences the absence of these preferences can cause rider dissatisfaction and should be ranked second. Attractive riders’ preferences can help identify potential rider needs, create a more positive experience for riders and increase rider trust in platform services. Platforms should strategically integrate temporarily unrelated preferences with other preferences to enhance platform appeal. Based on this, the research proposes some recommendations for the crowdsourcing logistics participants:

Consumers: (1) Promote Awareness and Support: Recognize riders’ challenges and choose platforms known for fair practices and positive contributions to the gig economy. (2) Use Feedback Constructively: Engage with platform feedback mechanisms thoughtfully to aid in service enhancement and rider performance.

Riders: (1) Advocacy for Rights: Band together to advocate for rights and engage with platforms and authorities for fair policy development. (2) Pursue Continuous Improvement: Seize training opportunities provided by platforms and share experiences to build a supportive rider community.

Merchants: (1) Ensure Collaborative Operations: Collaborate with logistics platforms to align delivery operations with rider welfare and customer satisfaction, adopting rider-friendly practices for improved delivery efficiency. (2) Provide Constructive Feedback: Actively offer feedback on the delivery process to help refine rider preferences and boost service quality.

Impact of Policies

The research emphasizes the practical significant impact of changes in platform policies and public opinion on social media on gig workers’ preference attention and satisfaction. Public opinion can rapidly spread and affect gig workers’ acceptance of platform policies and their overall job satisfaction. This necessitates that platforms not only focus on internal policy formulation and implementation but also actively manage external public relations. Additionally, this paper underscores the importance of gig workers’ technological skills for crowdsourced logistics platforms, noting that these skills are directly linked to riders’ motivation to participate and the enhancement of platform loyalty. Based on the analysis above, the study proposes tailored policy recommendations as follows:

Platforms: (1) Focus on Essential Preferences: Enhance platform management through equitable compensation, clear policies, and robust communication channels. This underpins workforce stability and operational efficiency. (2) Enhance Key Preferences: Implement features and policies that improve delivery timing, system functionality, and user-friendliness. Invest in technological advancements to boost information accuracy, provide comprehensive training, and nurture rider relationships. (3) Capitalize on Emerging Preferences: Employ data analytics to unearth and cater to evolving rider needs with innovative services, such as adaptable order options and superior customer support, distinguishing the platform and boosting rider contentment. (4) Adapt to Dynamic Preferences: Stay alert to trends and feedback regarding pandemic-related distribution, experience sharing, and expected rewards, swiftly integrating these insights as they gain relevance.

Government: (1) Regulatory Framework: Create regulations ensuring fair treatment and proper compensation for gig workers, promoting transparency, and protecting riders’ rights. (2) Support Safety and Training: Support the development of safety measures and training programs, enhancing riders’ skills for their professional growth.

Limitations

Further exploration into how platforms can leverage digital marketing strategies to enhance gig workers’ participation and satisfaction is warranted, given the crucial role of technological skills in the adoption and effective use of digital marketing tools within the gig economy.

Conclusion

This investigation delves into the evolving dynamics of gig workers’ preferences within the crowdsourced logistics sector, employing an in-depth LDA topic model analysis of rider reviews to reveal intricate patterns of attention and satisfaction changes over time. The study highlights the fluidity of gig workers’ preferences, impacted by shifts in platform participants, platform policies, and social media discourse. Significant contributions of this research include the development of a refined classification system for gig worker preferences, pinpointing key areas such as platform management, delivery efficiency, and technology enhancements crucial for boosting worker satisfaction and operational performance. Furthermore, the research underscores the pivotal influence of policy adaptations and public sentiment in molding these preferences, indicating the necessity for platforms to proactively manage both internal strategies and external public relations to cultivate a supportive and content workforce. By illustrating the interconnectivity among various preference categories and the influence of external factors, this research opens avenues for future inquiries into technological and strategic interventions aimed at enriching the gig work experience.

Supplemental Material

sj-docx-1-sgo-10.1177_21582440241271145 – Supplemental material for Time Evolution Analysis of Riders’ Preference Attention and Satisfaction on Real-Time Crowdsourcing Logistics Platform

Supplemental material, sj-docx-1-sgo-10.1177_21582440241271145 for Time Evolution Analysis of Riders’ Preference Attention and Satisfaction on Real-Time Crowdsourcing Logistics Platform by Yi Zhang, Dan Li and Shengren Liu in SAGE Open

Footnotes

Acknowledgements

The authors gratefully acknowledge the support of The National Social Science Fund of China. The support of Prof. Benoit from Georgia Institute of Technology during the initial stages of this research is also appreciated.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was funded by The National Social Science Fund of China (Contact Number: 22BGL124).

Ethical Approval

All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.

Informed Consent

Before starting the Photovoice activity, the participants received information on the use of the images and gave written informed consent to voluntary participation. Informed consent was obtained from all individual participants.

ORCID iDs

Yi Zhang

Dan Li

Data Availability Statement

The datasets generated during the current study are available from the corresponding author on reasonable request. The data that support the findings of this study are available from the corresponding author, upon reasonable request.

Supplemental Material

Supplemental material for this article is available online.

Notes

References

Agboola

O. P.

Tunay

(2023). Urban resilience in the digital age: The influence of information-communication technology for sustainability. Journal of Cleaner Production, 428, 139304. https://doi.org/10.1016/j.jclepro.2023.139304

Agyemang

K. O.

(2022). A Bayesian inference model for sustainable crowd source logistics for small and medium scale enterprises (SME) in Africa. American Journal of Industrial and Business Management, 12(4), 750–773. https://doi.org/10.4236/ajibm.2022.124038

AlSumait

Barbará

Domeniconi

(2008). On-line LDA: Adaptive topic models for mining text streams with applications to topic detection and tracking [Conference session]. 2008 Eighth IEEE International Conference on Data Mining. 3-12.10.1109/icdm.2008.140

Baki

(2022). An integrated multi-criteria structural equation model for green supplier selection. International Journal of Precision Engineering and Manufacturing-Green Technology, 9(4), 1063–1076. https://doi.org/10.1007/s40684-021-00415-7

Blei

D. M.

A. Y.

Jordan

M. I.

(2003). Latent Dirichlet allocation. Journal of Machine Learning Research, 3, 993–1022.

Castillo

V. E.

Bell

J. E.

Rose

W. J.

Rodrigues

A. M.

(2017). Crowdsourcing last mile delivery: Strategic implications and future research directions. Journal of Business Logistics, 39(1), 7–25. https://doi.org/10.1111/jbl.12173

Chen

Kou

Shang

Chen

(2015). Visualizing market structure through online product reviews: Integrate topic modeling, TOPSIS, and multi-dimensional scaling approaches. Electronic Commerce Research and Applications, 14(1), 58–74. https://doi.org/10.1016/j.elerap.2014.11.004

Chen

M.-C.

Hsu

C.-L.

Huang

C.-H.

(2021). Applying the Kano model to investigate the quality of transportation services at mega events. Journal of Retailing and Consumer Services, 60, 102442. https://doi.org/10.1016/j.jretconser.2021.102442

Coleman

Hurley

Koliba

Zia

(2017). Crowdsourced Delphis: Designing solutions to complex environmental problems with broad stakeholder participation. Global Environmental Change, 45, 111–123. https://doi.org/10.1016/j.gloenvcha.2017.05.005

10.

Ding

Cao

Chen

Sun

Zhang

Tao

(2023). Large-scale multimodal multiobjective evolutionary optimization based on hybrid hierarchical clustering. Knowledge-Based Systems, 266, 110398. https://doi.org/10.1016/j.knosys.2023.110398

11.

Friedland

Balkin

D. B.

(2023). When gig workers become essential: Leveraging customer moral self-awareness beyond COVID-19. Business Horizons, 66(2), 181–190. https://doi.org/10.1016/j.bushor.2022.05.003

12.

Guo

Wang

Xing

(2022). Mining multi-brand characteristics from online reviews for competitive analysis: A brand joint model using latent Dirichlet allocation. Electronic Commerce Research and Applications, 53, 1–11. https://doi.org/10.1016/j.elerap.2022.101141

13.

Jung

Suh

(2019). Mining the voice of employees: A text mining approach to identifying and analyzing job satisfaction factors from online employee reviews. Decision Support Systems,123, 1–12, https://doi.org/10.1016/j.dss.2019.113074

14.

Khan

Parvaiz

G. S.

Dedahanov

A. T.

Abdurazzakov

O. S.

Rakhmonov

D. A.

(2022). The impact of technologies of traceability and transparency in supply chains. Sustainability, 14(24), 16336. https://doi.org/10.3390/su142416336

15.

Chaudhary

Zhang

(2020). Modeling spatiotemporal pattern of depressive symptoms caused by COVID-19 using social media data mining. International Journal of Environmental Research and Public Health, 17(14), 1–22. https://doi.org/10.3390/ijerph17144988

16.

Xue

Yang

Lou

(2022). Ontology-based mapping approach for automatic work packaging in modular construction. Automation in Construction, 134, 1–34. https://doi.org/10.1016/j.autcon.2021.104083

17.

Liu

Mao

Liao

Guo

Zhou

Jin

(2020). Adaptive loading plan decision based upon limited transport capacity [Conference session]. Database Systems for Advanced Applications, 25th International Conference, DASFAA 2020, Jeju, South Korea, September 24–27, 2020, Proceedings, Part III (pp. 685–697).

18.

Mladenow

Bauer

Strauss

(2016). “Crowd logistics”: The contribution of social crowds in logistics activities. International Journal of Web Information Systems, 12(3), 379–396. https://doi.org/10.1108/ijwis-04-2016-0020

19.

Nguyen

C. T.

Yuen

K. F.

Kim

T. Y.

Wang

(2023). An empirical study of the determinants of driver-partner engagement behavior in crowd logistics: A motivational theory perspective. International Journal of Logistics Management, 35(2), 601–622 https://doi.org/10.1108/ijlm-12-2022-0496.

20.

Nie

R.-X.

Tian

Z.-P.

Wang

J.-Q.

Chin

K. S.

(2020). Hotel selection driven by online textual reviews: Applying a semantic partitioned sentiment dictionary and evidence theory. International Journal of Hospitality Management, 88, 1–16. https://doi.org/10.1016/j.ijhm.2020.102495

21.

Wang

Chen

Liao

Zhang

Benito

R. M.

(2021). Culture under complex perspective: A classification for traditional Chinese cultural elements based on NLP and complex networks. Complexity, 2021, 1–15. https://doi.org/10.1155/2021/6693753

22.

Samad

T. A.

Ganguly

K. K.

Das

(2023). Towards a framework for development of crowd logistics: Paving the way for sustainable logistics. Computers & Industrial Engineering, 176, 109008. https://doi.org/10.1016/j.cie.2023.109008

23.

Shi

Yao

Luo

(2021). Innovative platform operations with the use of technologies in the blockchain era. International Journal of Production Research, 61(11), 3651–3669. https://doi.org/10.1080/00207543.2021.1953182

24.

Sim

Lee

S. K.

Sutherland

(2021). The impact of latent topic valence of online reviews on purchase intention for the accommodation industry. Tourism Management Perspectives, 40, 100903. https://doi.org/10.1016/j.tmp.2021.100903

25.

Ullah

Khan

Rakhmonov

D. A.

Bakhritdinovich

K. M.

Jacquemod

Bae

(2023). Factors affecting digital marketing adoption in Pakistani small and medium enterprises. Logistics, 7(3). https://doi.org/10.3390/logistics7030041

26.

Wahid

J. A.

Shi

Gao

Yang

Wei

Tao

Hussain

Ayoub

Yagoub

(2022). Topic2Labels: A framework to annotate and classify the social media data through LDA topics and deep learning models for crisis response. Expert Systems with Applications, 195, 1–15. https://doi.org/10.1016/j.eswa.2022.116562

27.

Wan

Peng

Xiao

Liu

Jiang

Liu

(2020). An association-constrained LDA model for joint extraction of product aspects and opinions. Information Sciences, 519, 243–259. https://doi.org/10.1016/j.ins.2020.01.036

28.

Wang

Xie

(2021). Optimal pricing of crowdsourcing logistics services with social delivery capacity. Journal of Combinatorial Optimization, 43(5), 1447–1469. https://doi.org/10.1007/s10878-020-00693-y

29.

Wang

Chi

J.-H.

Yuan

(2022). Consumers’ attitudes and their effects on electric vehicle sales and charging infrastructure construction: An empirical study in China. Energy Policy, 165, 1–14. https://doi.org/10.1016/j.enpol.2022.112983

30.

Xia

Zhang

(2018). A TAM-based approach to explore the effect of online experience on destination image: A smartphone user’s perspective. Journal of Destination Marketing & Management, 8, 259–270. https://doi.org/10.1016/j.jdmm.2017.05.002

31.

Yang

(2021). Research on logistics service quality of JD. Journal of Management and Humanity Research, 5, 55–68. https://doi.org/10.22457/jmhr.v05a052131

32.

Singh

(2022). Seven principles for rapid-response data science: Lessons learned from Covid-19 forecasting. Statistical Science, 37(2), 266–269. https://doi.org/10.1214/22-sts855

33.

Zhang

Shi

Abdul-Hamid

Zhang

Shen

(2022). Factors influencing crowdsourcing riders’ satisfaction based on online comments on real-time logistics platform. Transportation Letters, 15(5), 363–374. https://doi.org/10.1080/19427867.2022.2052643

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.04 MB