Examining the impact of social media’s educational content on adolescent social value formation

Abstract

The integration of social media into the daily lives of adolescents has sparked considerable interest in its influence on their social value formation. Social media, serving as a repository of varied information and knowledge, plays a pivotal role in subtly shaping the value orientations and behavioral propensities of adolescents. This research delves into the dynamics between educational content on social media platforms and the evolution of adolescents’ social values, addressing a notable gap in quantitative analysis and model development within this discourse. Employing the social media educational content-emotion calculation (SMEC-EC) method, the study quantitatively analyzes the social values of adolescents influenced by social media, thereby establishing a benchmark for evaluation. Further, it introduces a novel computational model for compensating feature extraction specifically designed for adolescent social values, enhancing the analytical perspective in feature analysis. Lastly, the research adopts a stepwise regression ordinary least squares (OLS) model to dissect the correlation between educational content on social media and adolescent social values, uncovering the underlying mechanisms by which social media influences these values. This research provides empirical evidence and theoretical guidance for educators and policymakers on how to effectively guide adolescents in forming positive and healthy social values within the social media environment, contributing to the promotion of social media as a platform for positive socialization among young people.

Keywords

social media educational content-emotion calculation (SMEC-EC) method feature extraction correlation analysis ordinary least squares (OLS) model

Introduction

Social media platforms have become an important channel for adolescents to access information, learn new knowledge, and interact with others. Due to the high popularity and frequent use of these platforms, their impact on adolescents is becoming increasingly significant, especially in the formation of values and social concepts. As the application of social media in the educational field continues to expand, educational content has also begun to be widely disseminated through these platforms, including formal education courses, informal learning resources, skill training videos, heuristic discussions, and more. In the contemporary digital era, social media has become ubiquitously intertwined with daily life, significantly impacting adolescents.^1–4 This demographic’s increasing engagement in the digital sphere has been observed to mold their cognition, emotions, and social values, predominantly through interactions on various platforms.³ These platforms, abundant with diverse informational and educational content, play a subtle yet pivotal role in shaping minors’ values, especially during their critical developmental stages.⁴ Despite its prevalence, the dynamics between values propagated through social media and conventional societal values, particularly their effects on adolescent development, are not thoroughly elucidated, thus forming the basis of this study.

Adolescents’ exposure to educational content on social media has become a common phenomenon and has a profound impact on their cognitive development and the formation of their values. In this era of information explosion, social media is not only the main platform for adolescent interaction but also an important channel for knowledge acquisition and learning. The educational content on these platforms varies greatly, covering everything from academic knowledge and skill development to life wisdom and moral concepts, all of which have the potential to positively affect adolescents’ knowledge, critical thinking abilities, creativity, and sense of social responsibility.^5,6 At the same time, due to the strong interactivity and fast information dissemination of social media, educational content can spread quickly and widely, often in a form that is more vivid and closer to life, making it easier for adolescents to accept and absorb. However, the accuracy and quality of this content vary, presenting challenges in information filtering and authenticity discernment.⁷ Therefore, understanding and guiding adolescents’ exposure to educational content on social media is not only crucial for their personal growth but also significant for the long-term development of society and cultural construction.

The growing omnipresence of social media and its effects on adolescent socialization have attracted considerable attention from academics and educators alike.^8,9 An exploration into how social media content influences the development of adolescent social values is imperative for comprehending their behavioral patterns, emotional inclinations, and decision-making processes.^10,11 Insights from this investigation are anticipated to offer strategic guidance for parents, educators, and policymakers, aiming to promote healthy social media engagement among adolescents and cultivate positive value systems.

Predominant research in the field has largely been centered on descriptive analysis and case studies, with a limited focus on quantitatively examining the relationship between social media content and adolescent values.^12–14 A notable oversight in existing methodologies is the underutilization of emotion calculation for the analysis of social media texts, coupled with a lack of specialized computational models for feature extraction. This gap has implications for the explanatory capacity and practical applicability of research outcomes.^14,15

Addressing these deficiencies, this thesis is structured into three distinct parts. Initially, it employs the SMEC-EC method for a quantitative examination of the influence of social media on adolescents’ social values, thereby laying a scientifically rigorous data foundation for the study. Following this, the development of a compensation computation method for the feature extraction of adolescent social values represents a significant innovation in this field. Conclusively, the implementation of a stepwise regression OLS model to meticulously analyze the correlation between educational content on social media platforms and adolescent social values not only broadens the horizons of social science research methodologies but also provides theoretical underpinning and empirical evidence to support educational practices in the realm of social values. The study aims to offer novel perspectives and methodologies for comprehending and influencing the formation of social values among adolescents in the context of social media.

In this study, the feature extraction technique of the SMEC-EC emotion computation method, a notable innovation in the field of social media sentiment analysis, not only fills the gap in traditional text analysis in capturing the informal and implicit emotional expressions unique to social media but also specifically enhances the accuracy and detail of recognizing complex emotions among adolescents. By precisely mapping the intricate relationship between adolescents’ emotional expressions and their value formation, this innovative technology of the study offers a new analytical tool for understanding the psychological and socialization processes of adolescents in the age of social media, providing significant theoretical and practical significance for guiding educators and policymakers on how to better utilize social media for education and value shaping.

Quantifying the impact of social media on adolescent social values

This paper is based on reasonable assumptions, beginning with the premise that educational content on social media has the potential to shape adolescents’ social values, with its contained information and interaction modes capable of influencing their cognition and behavior. Secondly, it assumes that the emotional expressions and information exchanges on social media platforms can be quantified; further, it posits that in the process of extracting features of adolescents’ social values, data missing can occur, which can be effectively compensated and corrected through a constructed missing data compensation calculation model. Lastly, it assumes that there is a statistically significant correlation, which can be revealed through a stepwise regression OLS model, showing the specific connection between educational content and adolescents’ values, providing strategies for guiding the values of young people. These assumptions form the foundation of the study, aiming to comprehensively understand and reveal the complex mechanisms behind the formation of social values among adolescents in the social media environment.

The SMEC-EC method is an advanced text analysis technique specifically designed for the informal and unstructured language features prevalent on social media. It takes into account various factors such as emoticons, emotional punctuation, and rhetorical devices to capture and analyze the emotional tendency and intensity of texts. This method initially identifies and evaluates direct emotional expressions in the text through natural language processing (NLP) technology, where symbols like smiley faces or exclamation marks may directly express happiness or strong emotions. Simultaneously, SMEC-EC can identify and correct emotional reversals caused by rhetorical devices such as irony or exaggeration through contextual analysis algorithms, ensuring the accuracy of emotion recognition. Additionally, this method may utilize machine learning models to understand and adapt to the constantly evolving language habits and expression styles on social media. In this study, the SMEC-EC method is applied to analyze how adolescents receive and express educational content on social media, correlating the results of emotion computation with changes in adolescents’ social values, thereby providing researchers with a highly precise and sensitive tool to quantify and understand the emotional dynamics and value formation processes of adolescents in the complex social media environment. Figure 1 illustrates the comprehensive process employed in this research, depicting the flowchart for the quantification and feature extraction of adolescent social values.

Figure 1.

Flowchart for quantification and feature extraction of adolescent social values.

In the application of the SMEC-EC method, the importance of a representative and relevant corpus is emphasized. This corpus must encompass vocabulary and expressions pertinent to the research topic to ensure the validity of the model’s training and application. For instance, the emotional intensity of a social media evaluation text (df) within a specific value category (cz) is indicated by SV(f, z), with the text’s inherent gain intensity represented by ζ(f, z) and its emotional value under the value category denoted by TV(f, z). The specific formula for the SMEC-EC computation is articulated as follows:

S V (f, z) = ζ (f, z) + T V (f, z)

(1)

In the realm of social media, textual communication frequently incorporates informal and unstructured language elements, such as emojis and exclamation marks, collectively termed emotional punctuation. Additionally, rhetorical devices like irony and hyperbole create emotional reversals, intricately layering the text with nuanced emotional content. These elements are indispensable for discerning the genuine emotional intent behind the text. Emotional punctuation typically amplifies or attenuates the emotive tone of sentences, while emotional reversals can subvert the initial emotional expressions. In this research, the inclusion of these aspects into the SMEC-EC method significantly enhances the accuracy of capturing and interpreting adolescents’ complex emotional responses to educational content on social media. This refined approach aligns the quantification results more closely with adolescents’ authentic social values and emotional experiences.

The methodological framework posits that the number of negation words preceding an emotional word, q, is denoted by NE_q, and the emotional value of q by ω_q. The series of emotional punctuations following q is indicated by O_q, with their associated emotional values represented by x_w. The set of degree adverbs preceding q is marked as F_q, with the intensity of these adverbs symbolized by α_f. The occurrence of emotional reversal in word q within the value category z is signaled by ψ_q,z. The emotional value of a social media evaluation text, f, under the value category z is notated as TV(f, z), with the precise calculation method outlined as follows:

T V (f, z) = \sum_{q = 1}^{| Q_{z} |} {(- 1)}^{N E_{q} + ψ_{q, z}} ω_{q} \sum_{o = 1}^{| O_{q} |} β_{o} \sum_{f = 1}^{| F_{q} |} α_{f}

(2)

On social media platforms, emotional reversal is defined as expressing an emotional stance or evaluation that contradicts the literal meaning of the language used. This rhetorical strategy, contextually dependent and prevalent in social media discourse, manifests in forms such as sarcasm, exaggeration, and mockery. Users often employ emotional reversal to express complex emotions or articulate profound criticism and dissatisfaction toward specific events or topics. Recognizing emotional reversals is pivotal in the quantification of adolescent social values. It enables the revelation of genuine emotions and attitudes not overtly expressed in the text, thus allowing a more accurate assessment and understanding of the values and emotional inclinations of adolescents as influenced by social media. The specific calculation for this aspect is delineated as follows:

ψ_{q, z} = {\begin{cases} 1, z \in {H A, L I} A N D ? \in O_{q} \\ 0, otherwise \end{cases}

(3)

In the multi-modal environment of social media, communication extends beyond text, with visual elements playing a crucial role. Hence, integrating these elements into the assessment of social value gain intensity is imperative. Such elements enhance linguistic expressiveness, rendering the portrayal of adolescents’ values more vivid and multifaceted. Adolescents utilize visual elements, such as font color and size, to imbue their communications with additional emotional depth or emphasis. These elements, functioning as non-verbal communicative tools, are imbued with metaphorical and symbolic significance, thereby influencing the reception and interpretation of messages. For example, larger fonts might be employed to express intense emotions or to highlight specific points, while bright colors could be used to draw attention or convey particular moods. Incorporating these elements into the SMEC-EC method refines the accuracy in capturing and quantifying the emotional weight and intensity of expression of social values conveyed through these non-verbal cues. These cues add further dimensions to adolescents’ expression and emotional nuances on social media.

It is posited that the set of emotional words within a social media evaluation text, categorized under z, is denoted as Q_z. The font size of the text is indicated by FS(f), the presence of black font color by φ(f), and the occurrence of special text features like flashing or rotating text by λ(f). The gain intensity of the social media evaluation text, represented by ζ(f, z), is calculated as follows:

ζ (f, z) = {\begin{cases} 0, T V (f, z) = 0 \\ {| Q_{z} |}^{F S (f) + φ (f) + λ (f)}, o t h e r w i s e \end{cases}

(4)

Feature extraction of adolescent social values under the influence of social media

In the current social media landscape, the social values of adolescents constitute a dynamic belief and attitude system. These values, traditionally influenced by societal institutions, are now significantly shaped through online interactions and information exchanges on digital platforms. The understanding of these values is pivotal in analyzing adolescent behavior and decision-making, as they manifest and evolve through social media interactions. Consequently, the process of feature extraction in this study involves a comprehensive analysis of linguistic, visual, and behavioral data on social media to accurately capture and quantify the evolving social values of adolescents. This analytical approach is critical for studying adolescent behavior and informing relevant educational policy development. In this research, adolescent social values are categorized into positive and negative social group values. It is postulated that the vector representing adolescent social values is indicated by a, while the set of social value vectors for category z within time slice s is denoted by F_s,z, and the quantity of these vectors is expressed as |F_s,z|. The formula for calculating the social values of the adolescent group is as follows:

l_{G R} (s, z) = \frac{\sum_{a \in F_{s, z}} a}{F_{s, z}}

(5)

Owing to the challenges in obtaining social media interaction data, this study approximates value changes by analyzing the overall emotional shifts in comments made by adolescents engaging with SMEC, thereby facilitating cluster analysis. The text of SMEC is symbolized by N, segmented into equal-length text fragments represented by N = {n₁, n₂, …, n_S}, with the number of these segments denoted by S. The value feature sequence of adolescents in response to social media evaluation text is represented by i_N = {T₁, T₂, …, T_S}, and their emotional feature sequence under N by i_N.

The social value orientation within adolescents reflects their core beliefs and attitudes, and the analysis of emotional clustering in comments elucidates these values’ actual expressions in specific contexts. A comparison for consistency is conducted to ascertain whether the features of social values are accurately mirrored in emotional expressions and to determine if these expressions genuinely represent the adolescents’ values. It is acknowledged that inconsistencies may emerge when expressions on social media are swayed by immediate emotions, contextual factors, group dynamics, or specific topical discussions, thereby diverging from adolescents’ general values. Additionally, these inconsistencies might stem from the extraction algorithm’s failure to capture subtle variances in values accurately. Addressing these discrepancies is instrumental in refining the analysis’s precision, ensuring that the study’s results more accurately mirror the values and emotional states of adolescents. It is posited that the vector representing the social values of the adolescent group within a particular time slice is denoted by l_GR, and the adjusted social value vector is indicated by l_CO. The specific formula for correction is as follows:

l_{C O} = η μ + (1 - η) l_{G R}

(6)

In the context of this research, it is assumed that the vector corresponding to the social values associated with the uth social media evaluation text by an adolescent within a given time slice is represented by l_u. The total number of social media evaluation texts disseminated within the same time slice is represented by v. Subsequently, an averaging process is applied to the aggregate social value vectors of adolescents within that time slice, culminating in the following calculation formula:

l_{A V} = \frac{l_{1} + l_{2} + \dots + l_{l}}{v}

(7)

In addressing the underrepresentation of positive social values in adolescents’ expressions on social media, the study introduces specific supplementary and intervention measures. These measures, termed “positive deficiency compensation,” are implemented in scenarios where adolescents’ expressions on social media lack representations of socially endorsed positive values. The intervention strategies may encompass educational initiatives, showcasing of positive online role models, or providing opportunities for adolescents to exhibit positive social behaviors. These strategies are designed to foster the frequent embodiment and emphasis of universally recognized positive values such as honesty, empathy, and responsibility, in adolescents’ social media interactions. The temporal distance from the last non-deficient social value instance is quantified by T_PR, with the corresponding calculation formula delineated as follows:

ψ_{PR} = \tanh (\frac{1}{T_{PR}})

(8)

Conversely, “negative deficiency compensation” in adolescent social values pertains to corrective or intervention measures applied to adolescents expressing socially disapproved negative values on social media. This encompasses establishing platform rules to curb the proliferation of negative discourse, offering mental health support, facilitating values education, and imparting critical thinking skills. These measures aim to mitigate the adverse effects of social media on adolescents, such as the propagation of hate speech, inclination towards violence, or dissemination of misinformation. The negative compensation coefficient is formulated to reflect the extent of emotional impact in subsequent time slices following a current emotional deficiency. The number of steps leading to the next instance without a deficient social value is symbolized by T_NE, with the specific calculation formula presented as follows:

ψ_{N E} = \frac{1}{\tanh (\frac{1}{T_{N E}})}

(9)

Acknowledging the multifaceted influences on adolescents’ behaviors on social media, such as variations in psychological states, social environment dynamics, and impacts of specific events, this study recognizes potential discontinuities or gaps in their value expression over time. To provide a more precise and continuous portrayal of changes in adolescent social values, the adolescent social value deficiency compensation method is proposed. This method aims to enhance the understanding of the development of adolescents’ values and the degree of influence exerted by social media.

The core principle of this method is predicated on the assumption that the missing social values at a given time point can be inferred by considering the adjacent non-missing moments in the time series and the prevailing social values of the group. Rooted in the notions of continuity and group influence, this approach suggests that an individual’s values do not undergo substantial changes in a brief period and are shaped by group trends. By integrating the values from preceding and subsequent moments with the overall trends of the group’s values, the method enables a reasoned estimation and compensation for the missing values. It is posited that the vector representing the compensated social value at the current moment is denoted by l_LO, with the personalized factor symbolized by ϕ, where a lower ϕ value implies greater influence from the group’s social values. The coefficients for positive and negative compensation are indicated by ψ_PR and ψ_NE, respectively. The social value vector corresponding to the last non-missing moment is represented by l_PR, and the vector depicting the current moment’s social values of the adolescent group by l_GR. The specific formula for this calculation is outlined as follows:

l_{L O} = φ (\frac{ψ_{PR} l_{PR} + ψ_{N E} l_{N E}}{ψ_{PR} + ψ_{N E}}) + (1 - φ) l_{G R}

(10)

Correlation analysis of educational content on social media platforms and adolescent social values

In this segment of the study, a stepwise regression OLS model is utilized to examine the correlation between educational content on social media platforms and adolescent social values. Figure 2 provides a visual representation of the correlation analysis model.

Figure 2.

Diagram of the correlation analysis model.

The model’s dependent variable is the indicator of adolescent social values, which is quantified using the algorithm output derived from the performance of social media, as elaborated in the preceding sections. The primary explanatory variable in this model is adolescents’ degree of exposure to educational content on social media, gauged through the frequency, depth, and quality of their interactions with such content. Additional control variables considered include the misleading nature of the educational content on social media platforms, its relevance to interests, and the influence of peers. The random error term in the model accounts for the impact of all variables that are either unobservable or not included in the model, thus encompassing potential model errors. The regression coefficients are composed of the intercept term and the partial regression coefficients for each explanatory variable. These coefficients indicate the expected value of the dependent variable when all explanatory variables are zero, as well as the expected magnitude of change in the dependent variable for each unit change in an explanatory variable. The dependent variable is represented by A, the core explanatory variable by B, control variables by M, the random error term by λ, the constant term by γ₀, and the regression coefficients by γ₁ and γ₂. The specific regression model, encompassing individuals denoted by i and platforms by j, is formulated as follows:

B_{i j} = γ_{0} + γ_{1} A_{i j} + γ_{2} M_{i j} + λ_{i j}

(11)

To ascertain the influence of various factors on the correlation between educational content on social media platforms and adolescent social values, this study introduces three hypotheses:

(a) Hypothesis concerning the impact of misleading educational content (H1):

Null Hypothesis (H1_0): The misleading nature of educational content on social media platforms exerts no significant influence on adolescents’ social values.

Alternative Hypothesis (H1_a): The misleading nature of educational content on social media platforms significantly impacts adolescents’ social values, with this impact being negative.

(b) Hypothesis pertaining to the impact of educational content’s interest relevance (H2):

Null Hypothesis (H2_0): The relevance of educational content to adolescents’ interests on social media platforms does not significantly influence their social values.

Alternative Hypothesis (H2_a): The relevance of educational content to adolescents’ interests on social media platforms significantly influences their social values, with this impact being positive.

Null Hypothesis (H3_0): Peer influence does not significantly affect the social values formed by adolescents through educational content on social media.

Alternative Hypothesis (H3_a): Peer influence significantly affects the social values formed by adolescents through educational content on social media.

In these hypotheses, the dependent variable is denoted as HA, misleading content as K, content relevant to interests as D, peer influence as Z, control variables as M, the random error term as λ, the constant term as γ₀, and regression coefficients as γ₁ and γ₂. The individuals and platforms are represented by i and j, respectively. Regression models are employed in this study to test the aforementioned hypotheses:

H A_{i j} = γ_{0} + γ_{1} K_{i j} + γ_{2} M_{i j} + λ_{i j}

(12)

H A_{i j} = γ_{0} + γ_{1} D_{i j} + γ_{2} M_{i j} + λ_{i j}

(13)

H A_{i j} = γ_{0} + γ_{1} Z_{i j} + γ_{2} M_{i j} + λ_{i j}

(14)

The features extracted by the method proposed in this paper are directly related to adolescents’ social values because these features are meticulously selected and quantified from social media content. They represent the value concepts and behavior patterns that adolescents frequently encounter in the social network environment. These features can reveal the value orientations that adolescents may adopt or resist during their social interactions and information consumption processes, while also reflecting the potential influence and imitation effect within the social media environment. Through comprehensive analysis of these features, researchers can understand and predict how adolescents’ values are shaped by social media content, as well as how these values manifest and evolve in real life, thus directly linking to the formation of their social values.

Experimental results and analysis

Conducting experiments and data preprocessing is a crucial first step. This process begins with the collection of social media text data, followed by cleaning to remove irrelevant information such as URLs, user tags, and special formatting codes. Next, tokenization is performed to divide continuous text into meaningful units, while also identifying and labeling non-textual emotional carriers such as emoticons and emotional punctuation. Subsequently, the text undergoes part-of-speech tagging, syntactic parsing, and the removal of stop words to refine the data and retain key components of emotional expression. Through this detailed series of preprocessing steps, it is ensured that the data input into the model is clean, structured, and rich in emotional information. Table 1 displays a comparative evaluation of emotion lexicons, categorizing them into negative and positive value categories. The SMEC-EC method, employing the lexicon adopted in this study, is observed to yield higher metrics of accuracy, precision, recall, and F1-score across both categories. For the negative category, accuracy is recorded at 0.936, precision at 0.925, recall at 0.945, and F1-score at 0.936; for the positive category, these figures are 0.954, 0.936, and 0.945, respectively. When juxtaposed with SentiWordNet and General Inquirer, the lexicon utilized in this research demonstrates superior performance in all measured metrics. This substantiates the high efficacy of the lexicon selected for this study in analyzing adolescent social values, thereby offering significant data to elucidate the interplay between educational content on social media and adolescent values.

Table 1.

Comparative results of emotion lexicons.

Emotion lexicon	Value category	Accuracy	Precision	Recall	F1-score
SentiWordNet	Negative	0.825	0.814	0.814	0.824
SentiWordNet	Positive	0.825	0.778	0.916	0.847
General Inquirer	Negative	0.889	0.879	0.923	0.889
General Inquirer	Positive	0.889	0.923	0.889	0.915
Lexicon used in this study	Negative	0.936	0.925	0.945	0.936
Lexicon used in this study	Positive	0.936	0.954	0.936	0.945

Table 2 of the study presents a performance evaluation of various feature extraction methods employed for analyzing adolescent social values. The methods compared include Latent Semantic Analysis (LSA), Latent Dirichlet Allocation (LDA), Valence Aware Dictionary and sEntiment Reasoner (VADER), Recursive Neural Tensor Network (RNTN), and the novel method proposed in this research. Upon consideration of all performance metrics, the method developed in this study exhibits superior results in terms of accuracy, recall, and F1-score, indicating its remarkable efficacy in precisely identifying and categorizing adolescent social values. Notably, the high recall rate of this method underscores its proficiency in capturing essential information, a critical aspect in social value analysis where overlooking key data could lead to analytical biases. Therefore, it is concluded that the feature extraction method introduced in this research is highly effective, characterized by both high accuracy and completeness, providing substantial data support for the study of adolescent social values.

Table 2.

Comparative results of different adolescent social value feature extraction methods.

Method name	Value category	Accuracy	Precision	Recall	F1-score
LSA	Negative	0.778	0.887	0.736	0.812
LSA	Positive	0.778	0.678	0.887	0.773
LDA	Negative	0.923	0.952	0.925	0.926
LDA	Positive	0.923	0.831	0.915	0.874
VADER	Negative	0.934	0.925	0.956	0.935
VADER	Positive	0.934	0.936	0.897	0.926
RNTN	Negative	0.924	0.907	0.921	0.915
RNTN	Positive	0.924	0.934	0.887	0.923
Method used in this study	Negative	0.938	0.914	0.958	0.928
Method used in this study	Positive	0.938	0.947	0.935	0.954

Table 3 offers a comparative analysis of execution times for the aforementioned feature extraction methods, assessing both training and testing durations in milliseconds (ms). LSA is noted for having the shortest training time (12564.54 ms), whereas RNTN registers the longest (24587.23 ms). The training time for the method employed in this study is measured at 22365.89 ms, which, while not the shortest, is reasonably within an acceptable range. In terms of testing time, VADER records the briefest duration (256.31 ms), in contrast to the prolonged duration observed for RNTN (912.47 ms). The testing time for the method utilized in this study stands at a moderate 345.26 ms, slightly exceeding that of LSA and LDA but significantly shorter than RNTN. This finding highlights the necessity of balancing efficiency and performance in practical applications. Although the training and testing times of the method utilized in this study are not the shortest, its combination of performance advantages, as demonstrated in Table 2, offers an optimal compromise between efficiency and accuracy.

Table 3.

Comparative results of execution time for different adolescent social value feature extraction methods.

Method name	Training time (ms)	Testing time (ms)
LSA	12564.54	364.25
LDA	22145.25	287.25
VADER	18965.37	256.31
RNTN	24587.23	912.47
Method used in this study	22365.89	345.26

Figure 3 in this study presents the empirical analysis results of the feature extraction process for adolescent social values across various types of educational content on social media platforms. The analysis involved the collection of different categories of educational content from these platforms, categorized as knowledge-based, skill enhancement, and heuristic content.

Figure 3.

Empirical analysis results of feature extraction for adolescent social values.

The study conducted feature extraction experiments on these content types and compared the performance of the method developed in this research with several established methods, including LSA, LDA, VADER, and RNTN. The metrics used for comparison were accuracy, precision, recall, and F1-score. The findings, as depicted in the figure, reveal that in processing knowledge-based content, the method employed in this study, while slightly lower in recall than RNTN by 1.4%, surpasses RNTN and other models in terms of precision and other evaluative metrics. This suggests a more accurate differentiation between positive and negative samples, indicating a more balanced overall performance. In the realm of skill enhancement content, characterized by a plethora of novel terms, all models generally exhibit lower performance. Nevertheless, the method proposed in this study records the highest accuracy, reaching 73%, and outshines other models in precision, recall, and F1-score. This implies a relatively stronger capability of the proposed method to process and adapt to the linguistic nuances of skill enhancement content. Regarding heuristic content, the method used in this study exceeds the performance of other models across all metrics, highlighting its effectiveness in capturing the characteristics of heuristic content and conducting efficient feature extraction.

Figure 4 in the study provides an analytical overview of the trends in positive and negative social values among adolescents in relation to various types and forms of educational content on social media platforms. The analysis is based on the data encapsulated in the figure, which enables an examination of how different educational content types and presentation forms influence adolescents’ social value orientations. In terms of positive social values, blog posts demonstrate the highest association with knowledge-based content, with a 74% correlation, suggesting their effectiveness in positively influencing adolescents through the dissemination of knowledge. Conversely, instructional videos, interactive Q&A sessions, and infographics show comparatively weaker associations in this regard, indicating less effectiveness in knowledge transmission. For skill enhancement content, interactive Q&A sessions and infographics emerge as the most impactful, with 62% and 64% effectiveness, respectively, implying their significant role in positively influencing adolescents’ skill development. Although instructional videos also exhibit a positive influence, blog posts appear to have the least impact in this domain. Regarding heuristic content, blog posts and instructional videos hold relatively higher positions, suggesting their effectiveness in stimulating positive social values among adolescents. On the other hand, infographics and interactive Q&A sessions demonstrate a less substantial impact, highlighting a potential area for improvement.

Figure 4.

Trends of positive and negative social values among adolescents under different types of educational content on social media platforms.

The analysis of negative social value trends reveals that instructional videos (62%), interactive Q&A sessions (68%), and infographics (62%) exhibit higher negative trends in the context of knowledge-based content, pointing to possible deficiencies in positive guidance during knowledge dissemination. Blog posts, in contrast, show a considerably lower negative trend of 4%, indicating more effective positive guidance. In the realm of skill enhancement content, blog posts exhibit the highest negative trend, while infographics and interactive Q&A sessions demonstrate lower negative trends, suggesting their relative effectiveness and positive influence in this area. Among heuristic content, interactive Q&A sessions show the highest negative trend, indicating a need for enhanced positive guidance and monitoring in such formats.

The trend analysis provided in Figure 4 reflects the high value placed on instant gratification and visual stimulation in contemporary society. Concurrently, the persistently negative trend in blog posts regarding skill enhancement content is closely tied to the strong projection of bloggers’ personal experiences and subjective attitudes, demonstrating a societal and cultural inclination towards individualism and self-expression. In terms of heuristic content, the significant negative trend in interactive Q&A exposes a lack of regulation in societal discourse and a disregard for constructiveness and positive orientation in open interactions. In contrast, the relatively lower negative trend in blog posts and infographics on heuristic content may reflect a more cautious and thoughtful socio-cultural background of the authors and designers during creation. These trends reveal the profound impact of socio-cultural factors on the modes of educational content delivery, including technological advancements, a preference for instant communication, the rise of individualism, and changes in the public discussion space.

Table 4 in the study delineates the results derived from employing a stepwise regression OLS model to analyze the correlation between SMEC and adolescent social values. The table elucidates that the coefficient associated with the misleading nature of educational content is consistently negative across the entire sample and all subgroups, reaching statistical significance at the 1% level (denoted by three asterisks). This finding indicates an inverse relationship between the misleading nature of educational content and the positivity of adolescents’ social values, suggesting that an increase in misleading content correlates with a decrease in positive social values. Notably, this negative impact is more pronounced in subgroups focused on exploration and difference, implying a heightened adverse effect on adolescents engaged in self-identification and appreciating diversity. The relevance of educational content to interests, while not statistically significant in the full sample and most subgroups, shows significance in the consistency class at the 5% level (marked by one asterisk). The negative coefficient of this variable suggests that even when content is interest-aligned, it may not always correlate positively with social value positivity, indicating that content attractiveness does not necessarily equate to its educational merit. In contrast, the peer influence variable, with consistently positive coefficients across all samples and statistically significant at the 1% level, implies a positive correlation with adolescents’ social values. This finding suggests that positive peer behaviors and attitudes can foster the development of positive social values in adolescents. The sample size and F-value provide sufficient observational data for the correlation analysis, with the full sample comprising 10,652 observations. The F-value, being moderate, indicates a well-fitting model. The Adj_R² values, although not high, are deemed acceptable given the inherent complexity of social science data. These values, consistent across all subgroups, demonstrate the model’s stable explanatory power across diverse sample types.

Table 4.

Correlation analysis results between educational content on social media platforms and adolescent social values.

Variable	Full sample	Classification by identification degree			Classification by expression diversity
	Full sample	Identification class	Exploration class	Diversity class	Consistency class	Difference class
	Coefficient	Coefficient	Coefficient	Coefficient	Coefficient	Coefficient
Misleading nature of educational content	−0.047***	−0.0458***	−0.0635***	−0.0337**	−0.0426***	−0.0614***
Misleading nature of educational content	(−6.5215)	(−3.1254)	(−5.2487)	(−2.3154)	(−4.4875)	(−4.2315)
Relevance to interest in educational content	−0.023	−0.0412	−0.0224	−0.0425	−0.0358*	−0.0124
Relevance to interest in educational content	(−1.4952)	(−1.5698)	(−0.8125)	(−1.2145)	(−1.6875)	(−0.4231)
Peer influence	0.0245***	0.0158***	0.0248***	0.0234***	0.0268***	0.0247***
Peer influence	(9.6587)	(4.2365)	(5.1247)	(3.8745)	(9.1246)	(3.8752)
Sample size	10652	4215	3689	2546	6238	4328
F-value	57.89	15.62	21.46	8.35	41.78	16.89
Adj_R²	0.0156	0.0114	0.0158	0.0083	0.0178	0.0124

Note: *, **, *** represent statistical significance at the 10%, 5%, and 1% levels, respectively.

The proposed method for compensating missing data in the extraction of adolescent social values’ features stands out for its innovativeness and compensatory nature compared to other methods. By capturing the characteristics of adolescent social values more accurately, this method may more effectively identify and compensate for the subtle differences and complexities of social values that traditional methods might overlook. This approach specifically addresses the gaps that may arise in the process of value formation among adolescents and attempts to quantify these elusive variables, which is uncommon in previous research. However, a potential drawback of this method is its complexity and the difficulty of implementation, requiring advanced algorithms and substantial data support, which may limit its widespread use and application. Moreover, due to its novelty, the lack of extensive validation and peer review might necessitate further testing and refinement by the industry. The interpretation and application of data through this method may require interdisciplinary expertise, including psychology, sociology, and data science, potentially adding to the complexity of the research.

The potential practical significance of the above experimental results lies in providing educators, policymakers, and social media platform developers with deep insights into how social media content influences the formation of adolescents’ social values. By ensuring the accuracy of the SMEC-EC method, the research offers a reliable tool to help stakeholders understand the role of different emotional expressions in value transmission. Comparative studies of various feature extraction methods, especially in terms of accuracy and execution time, reveal the most effective analytical means, providing a basis for designing more precise interventions and educational content. Analyzing the impact trends of educational content on values across different platforms can help create a more positive and constructive online learning environment. Additionally, robustness checks enhance the credibility of the research findings, meaning these insights can be widely applied in practical strategies and educational practices to promote healthy value formation and socialization processes among adolescents. Overall, the results of this study may have significant practical implications for optimizing the educational potential of social media, guiding adolescents in safe internet use, and nurturing them into responsible digital citizens.

Conclusion

The research presented in this paper centers on the influence of social media on adolescents’ social values, endeavoring to bridge existing gaps in the field. Initially, the SMEC-EC method was employed for a quantitative assessment of social media content. This approach facilitated a scientific and rigorous evaluation of the interplay between social media content and adolescents’ social values from an emotional perspective. Subsequent to this, the study introduced a novel compensation computation method for the extraction of features associated with adolescent social values. This advancement potentially aids researchers in more precisely capturing the nuances of adolescent social values and addressing shortcomings in current research methodologies. Furthermore, the stepwise regression OLS model, developed within this study, provided a comprehensive analysis of the relationship between educational content on social media platforms and adolescent social values. This model not only elucidated the statistical correlation between these two aspects but also augmented the reliability of the research findings through robustness verification.

In this research, feature extraction based on the SMEC-EC method represents a critical point of innovation. This method allows researchers to delve into the subtle emotional nuances within social media texts and convert these nuances into quantifiable data points. Using the SMEC-EC method, researchers can accurately capture complex emotional expressions, such as those conveyed through emoticons, emotional punctuation, and rhetorical devices, including irony or humor expressions that traditional text analysis methods might overlook. This highly refined feature extraction not only enables a more comprehensive understanding of how adolescents express and receive emotions on social media but also allows for a more precise mapping of the relationship between these emotional expressions and the formation of social values. This innovation provides new perspectives and technical means for assessing and understanding the emotional and value dynamics of contemporary adolescents amidst their increasing use of social media, significantly advancing the depth and breadth of research into the impact of social media.

The core findings of this paper reveal a significant correlation between SMEC and the formation of adolescent social values. Firstly, the research quantitatively demonstrates the association between emotional expressions on social media and value shaping, finding that positive emotional messages are more likely to promote the formation of positive values. Secondly, it effectively addresses the issue of incomplete data in the analysis of social value features, enhancing the completeness of understanding the factors influencing adolescent values. Lastly, it further confirms the positive correlation between specific aspects of educational content (e.g., moral education messages and messages reinforcing a sense of social responsibility) and positive changes in adolescent social values. Experimentally, the study’s application of the SMEC-EC method was validated through the comparison of various emotion lexicons, ensuring the method’s accuracy and appropriateness. The performance of different adolescent social value feature extraction methods was rigorously evaluated, focusing on accuracy and execution time. Empirical analysis shed light on the influence of diverse educational content on social media platforms upon adolescent social values. This research delved into the trends of both positive and negative social values among adolescents in relation to different types of educational content on social media platforms. The correlation between educational content and adolescent social values was meticulously analyzed, with robustness checks fortifying the credibility of the findings.

Statements and declarations

Footnotes

Conflicting interest

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

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

References

Charoensukmongkol

. The impact of social media on social comparison and envy in teenagers: the moderating role of the parent comparing children and in-group competition among friends. J Child Fam Stud 2018; 27(1): 69–79. DOI: 10.1007/s10826-017-0872-8.

Kline

Metcalf

Patel

, et al. Adolescent experiences with social media and suicidality. Acad Pediatr 2023; 23(4): 755–761. DOI: 10.1016/j.acap.2022.09.020.

Zhang

. Trust in social media brands and perceived media values: a survey study in China. Comput Hum Behav 2022; 127: 107024. DOI: 10.1016/j.chb.2021.107024.

Gan

, et al. Disentangling the factors driving users' continuance intention towards social media: a configurational perspective. Comput Hum Behav 2018; 85: 175–182. DOI: 10.1016/j.chb.2018.03.048.

Peker

Yildiz

. Examining the relationships between adolescents’ emotion regulation levels and social media addiction. Clinical and Experimental Health Sciences 2022; 12(3): 564–569. DOI: 10.33808/clinexphealthsci.869465.

Noon

Maes

Karsay

, et al. Making the good better? Investigating the long-term associations between capitalization on social media and adolescents’ life satisfaction. Media Psychol 2023; 27(2): 161–185. DOI: 10.1080/15213269.2023.2227941.

Schreurs

Meier

Vandenbosch

. Exposure to the positivity bias and adolescents’ differential longitudinal links with social comparison, inspiration and envy depending on social media literacy. Curr Psychol 2023; 42(32): 28221–28241. DOI: 10.1007/s12144-022-03893-3.

Atkin

Yoo

White

, et al. Validation of the multiracial youth socialization (MY-Soc) Scale among racially diverse multiracial emerging adults. J Fam Psychol 2022; 36(1): 13–22. DOI: 10.1037/fam0000879.

Bañales

Hope

Rowley

, et al. Raising justice‐minded youth: parental ethnic‐racial and political socialization and Black youth’s critical consciousness. J Soc Issues 2021; 77(4): 964–986. DOI: 10.1111/josi.12486.

10.

Marttila-Tornio

Ruotsalainen

Miettunen

, et al. Association between psychosocial problems and unhealthy health behavior patterns among Finnish adolescents. Child Psychiatr Hum Dev 2020; 51: 699–708. DOI: 10.1007/s10578-020-00967-w.

11.

Páez Gallego

De-Juanas Oliva

García Castilla

, et al. A study about social values and psychological well-being in young people: implications for educational social work. Soc Work Educ 2020; 39(6): 721–736. DOI: 10.1080/02615479.2020.1793935.

12.

Hoyos-Cifuentes

Fernández-Otoya

Rodríguez-Gómez

, et al. Emotional intelligence, human values, creation and dissemination of content on social networks by girls in countries with emerging economies. Int J Adolesc Youth 2024; 29(1): 2306886. DOI: 10.1080/02673843.2024.2306886.

13.

Caner

Efe

Başdaş

. The contribution of social media addiction to adolescent LIFE: social appearance anxiety. Curr Psychol 2022; 41(12): 8424–8433. DOI: 10.1007/s12144-022-03280-y.

14.

Hamid

Singh

. Emotion quantification techniques for cognitive reappraisal: a systematic review and scientometric analysis. Artif Intell Rev 2023; 56(S3): 3363–3416. DOI: 10.1007/s10462-023-10606-0.

15.

Yukalov

. Quantification of emotions in decision making. Soft Comput 2022; 26(5): 2419–2436. DOI: 10.1007/s00500-021-06442-5.