Application of clustering algorithm in social network user scenario prediction

Abstract

In order to improve the ability of social network user behavior analysis and scenario pattern prediction, optimize social network construction, combine data mining and behavior analysis methods to perform social network user characteristic analysis and user scenario pattern optimization mining, and discover social network user behavior characteristics. Design multimedia content recommendation algorithms in multimedia social networks based on user behavior patterns. The current existing recommendation systems do not know how much the user likes the currently viewed content before the user scores the content or performs other operations, and the user’s preference may change at any time according to the user’s environment and the user’s identity, Usually in multimedia social networks, users have their own grading habits, or users’ ratings may be casual. Cluster-based algorithm, as an application of cluster analysis, based on clustering, the algorithm can predict the next position of the user. Because the algorithm has a “cold start”, it is suitable for new users without trajectories. You can also make predictions. In addition, the algorithm also considers the user’s feedback information, and constructs a scoring system, which can optimize the results of location prediction through iteration. The simulation results show that the accuracy of social network user scenario prediction using this method is higher, the accuracy of feature registration of social network user scenario mode is improved, and the real-time performance of algorithm processing is better.

Keywords

Data clustering social network user context behavior analysis cold start

1 Introduction

The service functions of social networks, combined with data mining and behavior analysis methods for user characteristics analysis and scenario prediction of social networks, found the regularity of social network users [1 –3]. Research on social network user scenario prediction methods is of great significance in the construction of social networks and the analysis of behavior characteristics.

At present, common clustering methods include K-means algorithm, fuzzy C-means algorithm and hierarchical segmentation clustering mining methods [4 –7]. The fuzzy C-means clustering method is liable to fall into a local optimal solution in the clustering process of social network user scenario data mining, while the hierarchical segmentation clustering method for big data association mining is affected by the segmentation threshold [8, 9], It has greater sensitivity to the initial clustering center. With the advent of the era of big data, especially for the field of user behavior recommendation, the efficiency and accuracy of data are more demanded. Existing research has not fully considered the clustering algorithm in Importance in these two areas. In terms of efficiency, we can use the existing distributed platform Hadoop MapReduce computing framework [10 –12] to transplant the clustering algorithm. In terms of accuracy, we selected K-means algorithm as our core algorithm and used K-means++algorithm to optimize candidate points and algorithm optimization in distributed environment [13, 14]. And compared with the traditional density-based DBSCAN algorithm, the DBSCAN algorithm is more sensitive to parameter changes than the K-Means algorithm, and the optimized K-Means algorithm shows good efficiency and stability [15, 16]. In the existing research, the service recommendation for users is to use the user’s current GPS point and historical data to perform service recommendation for the user’s current status. When the user’s next possible location point is predicted, a service recommendation is made to the user [17]. Based on predictions, service recommendations are made to users through textual information (such as time and temperature). Before the user reaches the next location point, based on the user feedback information for a period of time, make accurate service recommendations for the user [18 –20]. And the user behavior changes, the system will promptly make feedback, modify user prediction parameters, and make further recommendations for users. However, the results of sociological research and our life experience show that similar behaviors and tastes are often exhibited in the same social circle. In order to make full use of auxiliary information to make trust predictions, [21] aggregates social networks through heterogeneity and proposes a new joint social network mining (JSNM method to solve this problem. In addition, this presents a unique challenge because the available data is often passively observed. Literature [22, 23] uses social science theory to propose a methodology to support the study of online trust evolution. Literature [24 –26] pointed out that due to personal characteristics Diversity. The previous recommendation algorithm will change with the current user and the recommendation performance is unstable. A dynamic competitive recommendation algorithm is proposed. This algorithm relies on a multi-component competition algorithm to provide continuous and stable recommendations in social networks. Real Twitter data evaluates the algorithm [27, 28]. Literature [29] points out that traditional content recommendation methods cannot be applied in the Web 3.0 environment, so by analyzing social networks, a more appropriate and reliable recommendation is proposed. Literature [30, 31] combined social network and analysis of semantic concepts to propose a personalized researcher Recommended method to researchers recommend appropriate research partners to promote the exchange of knowledge and discovery.

Due to the advent of the Web 2.0 era, users and digital content on multimedia social networks have exploded. The emergence of mobile terminal devices has also made it possible for users to access multimedia social networks at any time, so the interaction between users and the system has become increasingly complex. Design a multimedia content recommendation method based on user context-aware analysis. The proposed recommendation algorithm can predict the intent of a new user based on the user’s behavior sequence in the social network, and analyze the user’s preferences based on the intent. Therefore, the algorithm has no data sparse And cold start issues. Based on the social network user scenario prediction method based on clustering, the association topology model of social network user scenario distribution is constructed. The segmented feature extraction method is used to extract the associated feature of the social network user scenario pattern. The fuzzy C-means clustering method is used. Data mining is performed to realize social network user scenario prediction. Finally, performance tests were performed through simulation experiments to show the superior performance of the method in this paper in improving the accuracy of user scenario prediction. Finally, the GSRM model and algorithm proposed in the paper are implemented, and the accuracy and efficiency of the clustering user scenario prediction algorithm and the clustering algorithm in the GSRM model are verified through experiments.

2 Sequential analysis of behavior patterns of social network users

In order to provide users with more personalized services in the context of social media networks, this article analyzes the user’s context in multimedia social networks. Carl K. Chang’s Situ theory is oriented towards software engineering and cannot be fully applied to multimedia social The emerging application scenarios under the network, so the social situation SocialSitu architecture is established on the basis of the Situ theory. Through the improved GSP (Generalized Sequential Pattern) algorithm, a user behavior pattern discovery based on context analysis in a multimedia environment is designed. Method: The SocialSitu (t) of the user in the multimedia social network is analyzed to obtain the user’s intention sequence.

2.1 SocialSitu framework

In a multimedia social network environment, a large number of users may be in different groups, and the users in the corresponding groups have different roles. The roles of users in the groups may cause users to have different desires. This article expands and enriches the Situ framework. The specific definitions are as follows:

Definition 1: Situation (t) = (d, A, E), indicates that the situation at time t is composed of a triplet, where d represents the predicted user’s intention, and A represents the action taken by the user to achieve the goal corresponding to d, E represents the scenario value of a subset of the scenario variables at time t.

Definition 2: SocialSitu (t): indicates the situation at time t. SocialSitu (t) is a quadruple SocialSitu (t) = {ID, d, A, E}, where d represents the user’s intention at time t (desire, A represents the user’s operation corresponding to d at this time (That is, behavior), E represents environmental information, including terminal information used by the user, and ID represents user’s identity information.

Definition 3: Desire (D) refers to what the user wants to do when using social media services, that is, the user’s purpose. It is composed of a series of atomic desire (d), which is {dl, d2 ... dn}, d_i (1 ≤ i ≤ n) indicates the user’s intention at time i.

Definition 4: Intention (I) represents the SocialSitu sequence within the period from the start point to the goal realization, that is, I={SocialSitu(1), SocialSitu (2) SocialSitu (n)}, SocialSitu (1) is the start point, and SocialSitu(n) is the end point of goal completion.

The SocialSitu sequences here are directly related to achieving the purpose. Users achieve their goals through the Intention sequence. As shown in Fig. 1. Each node in the graph represents SocialSitu(t) at a certain time, startj (1 ≤ j ≤ n, j ∈ N) represents the starting point of Intention(i), these starting points can be the same or different, End represents the ending point of Intention(i), each SocialSitu(t) the sequence represents a sequence composed of different SocialSitu (t) that the user passes from the start point to the end point. Except for the end point, each sequence may have the same node. In the multimedia social network, at least one sequence corresponding to the user’s Intention is i ∈ N, i ≥ 1.

Fig. 1

Intention sequence diagram.

2.2 Social network user behavior pattern discovery algorithm

All frequent SocialSitu sequences related to the achievement of a goal in the user’s historical visit record and group them into an Intention sequence. The user has at least one goal in the multimedia social network, that is, corresponds to an Intention sequence. as shown in Fig. 2.

Fig. 2

User intention prediction.

The user has at least one goal in the multimedia social network. The identified specific behavior pattern sequence of the current user is stored in the database, and the current sequence of the user is compared with the behavior pattern sequence of the user in the database to predict the current user behavior Intent in order to respond promptly and quickly to user requests and provide personalized services. The serialization algorithm is shown in Table 1. In the algorithm, the function Support (SocialSitu (t)= $g_{j}^{'}$ ) indicates the calculation of SocialSitu (t) as the antecedent of the association rule, and $g_{j}^{'}$ indicates the support degree of the association rule result. $P (SocialSitu (t) \cup g_{j}^{'}$ means calculating the probability that SocialSitu (t) and $g_{j}^{'}$ occur simultaneously.

Table 1

Intent serialization algorithm based on context awareness

Input: Transaction data set DS, minimum support threshold Min_sup, user target G
Output: SocialSitu(t) sequence
SituBehaviorAnalytics(DS, Min Support, G)
Begin
for j ← 1 to n //n indicates the number of user’s goal
:for t ← 0 to T
$Support (SocialSitu (t) \Rightarrow g_{j}^{'}) = P (SocialSitu (t) \cup g_{j}^{'})$ Endfor
Endfor
$(Support (SocialSitu (t) \Rightarrow g_{j}^{'}) ≻ {Min}_{S} upport$
if
L₁ =∪ SocialSitu (t) ;:
endif
for k → 2 to m and L_k-1 ≠ Null
Generate candidate sets Ck:
$Support (C_{k} \Rightarrow g_{j}^{'}) = P (C_{k} \cup g_{j}^{'})$ ;
If $Support (C_{k} \Rightarrow g_{j}^{'}) ≻ {Min}_{S} upport$
L_k = C_k
endif
endfor
$Intention (t) = L_{k} \cup g_{j}^{'}$
endfor
End

3 Social network user scenario prediction based on clustering algorithm

3.1 SocialSitu framework

On the basis of constructing an association topology model of social network user profile distribution, the feature extraction of the user profile is performed, and the segmented feature extraction method is used to extract the associated feature of the social network user profile. Any user profile transmission tree The non-root node’s scenario mode social network user’s scenario step prediction value is: ${CW}_{min}^{l_{i + 1}} = {\begin{matrix} {CW}_{min}^{0} \times D_{l_{0}} χ, i = 0 \\ {CW}_{min}^{l_{i}} \times D_{l_{i}} χ, 0 ≺ i ≺ M \end{matrix}$ (1)

The QoS control weighted adaptive learning coefficient for social network user scenario prediction is expressed as: $χ = {log}_{(1 + D)} A / {CW}_{min}^{0}$ (2)

Among them, ${CW}_{min}^{0}$ is the predicted initialization value of the lowest root node in the social network user’s contextual topology transmission tree; and ${CW}_{min}^{l_{i}}$ is the social network user’s contextual topology control value in step L. During the process of scenario mode prediction, between each step The weighted control composition is the same, assuming that the information code on the cluster head node of the social network user scenario pattern in the hybrid cloud environment is expressed as: $u_{w} = - Ksat (S (t) / u) / (λ g + g_{D})$ (3)

Let U be a quantitative universe represented by precise values, C is a qualitative concept on U, and use an undirected graph model structure to represent the original perception of social network user scenarios. Adaptive scheduling of social network user scenarios is integrated. The model consists of a multiple hypothesis testing problem with unknown parameters. Parallel set and variable axis sorting methods are used to partition the social network user scenario storage structure. The scheduling model is: ${\begin{matrix} max U = u_{1} + u_{2} + . . . u_{n} \\ u_{i} = p_{i} \\ \sum_{i = 1}^{n} p_{i} = 1, 0 ≺ p_{i} \end{matrix}$ (4)

Among them, K represents a weighted adjustment coefficient. Based on the average number of child nodes in the social network and the number of child nodes, feature extraction and distributed network design are performed. At this time, a new source of social network user scenarios is formed mapping. $x_{n} = [x (0), x (1) . . . x (N - 1)]^{T}$ (5)

According to the characteristics of the information flow of the partition information of the social network user’s scenario mode, the feature decomposition is performed, and the limited condition association rule mining volume that defines the feature value is: $cum (λ_{1} x_{1}, λ_{2} x_{2} . . . λ_{k} x_{k}) = \prod_{i = 1}^{k} λ_{i} cum (x_{1}, x_{2} . . . x_{k})$ (6)

Among them, the k-th order cumulant c_k (τ₁, τ₂ . . . τ_k-1) of the process extracts the feature scale of the task information flow. Under the limited conditions of the adaptive scheduling of social network user profile partitions, the feature quantity of association rules is extracted to provide user profile tracking for social networks Data entry basis.

3.2 Social network user scenario prediction algorithm based on clustering

The clustering user scenario prediction algorithm is a prediction algorithm based on the clustering results. Based on the cluster trajectory obtained by the clustering, the system can obtain the “hot spots” recognized by the user. Based on this, the clustering user scenario prediction algorithm predicts the Next position.

As shown in Fig. 3, all users are first divided into two groups. Old users are users who already have a trajectory pattern in this system. At this time, we learn the trajectories of old users to establish the attenuation function and prefix tree of the old users. The other is that new users experience the system’s location prediction service for the first time. We will perform a “cold start” of the prediction service for new users, learn the trajectory patterns of all old users in the area of the new user, and establish an attenuation function. The probability distribution of the user from one point to another can be obtained, and then the first three points with the highest probability are found to predict the position of the target user. After receiving the prediction information, the target user sends feedback on the quality of the prediction information. After the feedback subsystem obtains the feedback information of the target user, it calculates the optimal prediction parameters of the current target user. The entire clustering user scenario prediction algorithm framework includes the following:

Fig. 3

Algorithm framework for clustering user scenario prediction.

Prediction algorithm: Obtain the current location information of the target user, model the target user, and establish the attenuation function and prefix tree structure respectively. If the target user is a new user, then only the corresponding attenuation function is established;

Scoring subsystem: According to the target user model established in (1), the scoring subsystem calculates the score of the next possible position of the target user, and sends the first three positions with the highest score to the target user.

Feedback subsystem: According to the predicted position in (2), the target user sends feedback information on the quality of the prediction. The clustered user scenario prediction algorithm uses the feedback subsystem to iteratively modify the target user’s own prediction through the feedback function.

3.3 Prediction optimization algorithm

Based on the construction of the association topology model and feature extraction of the social network user profile distribution, an improved design of the user profile prediction algorithm is proposed. This paper proposes a social network user profile prediction method based on fuzzy partition clustering. Segment feature extraction method performs association data mining of social network user’s scenario patterns. According to the characteristics of the user’s scenario pattern, the ratio of small to large is used to describe the degree of similarity of the social network user’s scenario pattern. It can be expressed as: $P_{i} = f (d_{i}, d_{i}^{'})$ (7)

Feature extraction of optimized social network user scenario association rules is: $p_{i} = \frac{r_{i}}{m} = \frac{r_{i} w_{0}}{{mw}_{i}} = \frac{w_{i}}{w_{1} + . . . . w_{i} + . . . w_{n}}$ (8)

Get the information entropy of a single social network user node L: $H_{i} (x) = \sum_{k = 1}^{K} p_{k} ln \frac{1}{p_{k}} = - \sum_{k = 1}^{K} p_{k} ln p_{k}$ (9)

Using adaptive optimization to H_i (x) to obtain the maximum distribution information max(H_i (x)) of the social network user’s scenario mode, from which the fusion feature extraction optimization constraints of the social network user’s scenario mode are: $f_{1}, u_{1}) ≺ (f_{2}, u_{2}) = {\begin{matrix} f_{1} ≺ f_{2}, u_{1}, u_{2} \geq α \\ f_{1} ≺ f_{2}, u_{1} = u_{2} \\ u_{1} ≻ u_{2}, otherwise \end{matrix}$ (10)

Among them, f₁, u₁, u₂ represents the objective function value of the social network user’s situational pattern similarity feature, respectively. The parallel sets variable axis sorting method was used to sort the social network users’ scenario patterns, and the trajectory tracking cluster analysis was performed based on the data mining results.

Combined with the fuzzy partition clustering method to find the hidden patterns in the data set, the asynchronous progressive weighting coefficients for user scenario pattern mining are: ${CW}_{min}^{n} = {\begin{matrix} ((1 - B_{l (n)}) e^{1 - α} \times {CW}_{min}^{l (n)}, α ≻ 1 \\ {CW}_{min}^{l (n)}, 0 \leq α \leq 1 \end{matrix}$ (11)

To sum up, the fuzzy partition clustering method is used to find the hidden patterns in the user scenario data set. Based on the data fuzzy partition clustering and mining results, the social network user scenario patterns are adaptively predicted.

4 Experimental verification

This article conducts experiments on a multimedia social network platform. The platform is a social media website that integrates multimedia content management, copyright protection, and social trinity. Experiments are performed on the platform based on real user data. There are a total of 7,439 playback record data, of which 6136 pieces of data were used as the training set and 1303 pieces of data were used as the test set. The accuracy, recall and comprehensive evaluation index F-measure of the sequence algorithm and user-based collaborative filtering recommendation algorithm and popular recommendation algorithm were tested respectively. The comparison data are as follows: Tables 2 –4 show. The graphs corresponding to the three tables are shown in Figs. 4 –6, respectively.

Table 2
Comparison of the accuracy of the three recommended algorithms

Recommended list length Content recommendation algorithm based on SocialSitu User-based collaborative filtering Popular recommendation algorithms

6 0.320 0.25 0

11 0.550 0.25 0.06

16 0.550 0.145 0.07

21 0.490 0.165 0.12

Recommended list length	Content recommendation algorithm based on SocialSitu	User-based collaborative filtering	Popular recommendation algorithms
6	0.320	0.25	0
11	0.550	0.25	0.06
16	0.550	0.145	0.07
21	0.490	0.165	0.12

Table 3

Comparison of the recall rates of the three recommended algorithms

Recommended list length	Content recommendation algorithm based on SocialSitu	User-based collaborative filtering recommendation algorithm	Popular recommendation algorithms
6	0.048	0.028	0
11	0.139	0.056	0.02
16	0.232	0.057	0.045
21	0.259	0.097	0.07

Table 4

Comparison of comprehensive indicators F-measure of three recommended algorithms

Recommended list length	Content recommendation algorithm based on SocialSitu	User-based collaborative filtering recommendation algorithm	Popular recommendation algorithms
6	0.078	0.058	0
11	0.215	0.086	0.035
16	0.298	0.077	0.055
21	0.322	0.122	0.085

Fig. 4

Comparison of accuracy.

Fig. 5

Comparison of recall rate.

Fig. 6

F comprehensive index comparison chart.

The serial number of the next location in our data structure is defined as (location point ID, frequent number). In the same experimental environment, we use benchmark methods to compare the accuracy of location prediction to verify the effectiveness of the Mining MP algorithm and the user scenario prediction algorithm. Therefore, based on this data set, they proposed a unified framework on the impact of time. The predictive factors of the benchmark method are affected by the multiple effects of space-time and check-in data. Since our system does not use the check-in data to predict the user’s next location, we only compare the influence of time and space on the prediction with them. It can be seen from Figure 7 that there are two curves, which are the spatiotemporal effects of the comparison of our user scenario prediction algorithm model and the benchmark model. We call their method the benchmark model. It can be seen from the figure that the average accuracy rate of the user scenario prediction algorithm model is 0.479, which is better than the baseline model of 0.393. The reason is that the user scenario prediction algorithm model uses the user’s historical trajectory information to mine the user’s frequent patterns, so the predicted position points have a high frequency, which means that it is likely that the user’s next trajectory will still follow the previous trajectory was performed.

Fig. 7

Comparison of the spatiotemporal impact of the user scenario prediction algorithm model and the benchmark model.

Based on clustering, this paper proposes a user scenario prediction algorithm. Through the attenuation function and prefix tree structure model, the system can recommend services to users without historical trajectories. In the scoring subsystem and feedback subsystem, user feedback the information will be updated in a timely manner, and feedback is sent on the quality of the prediction information. After the feedback subsystem obtains the feedback information of the target user, the optimal prediction parameters of the current target user are calculated. Utilize the optimal prediction parameters to make predictions suitable for the user.

5 Conclusion

This paper automatically predicts the user’s scenario mode, combines big data analysis methods, analyzes the user’s behavior characteristics and user relationship analysis, and improves the service function of the social network. Prediction method, constructing the association topology model of social network user profile distribution, using segmented feature extraction method to extract the associated feature of social network user profile, and using Parallel Sets variable axis ranking method to partition the social network user profile storage structure Scheduling, combined with fuzzy partition clustering method to find hidden patterns in the data set, fuzzy C-means clustering method is used for data mining, and social network user scenario pattern prediction is realized. Research shows that the accuracy of social network user scenario prediction using this method is higher, and the data mining accuracy is higher, which has good practical value in social network user scenario pattern mining and behavior analysis.

References

Wang

, Liu

, Cong

, et al., Application of genetic algorithm to land use optimization for non-point source pollution control based on CLUE-S and SWAT[J], 560 (2018), 86–96.

Phimpraphai

Waraphon

, Tangkawattana

Sirikachorn

and Kasemsuwan

Suwicha

, Social Influence in Liver Fluke Transmission: Application of Social Network Analysis of Food Sharing in Thai Isaan Culture[J], Advances in Parasitology 101 (2018), 97–124.

Laleh

Naeimeh

, Carminati

Barbara

and Ferrari

Elena

, Risk Assessment in Social Networks Based on User Anomalous Behaviors[J], IEEE Transactions on Dependable & Secure Computing 15(2) (2018), 295–308.

Matthew

M.G.

, On the deep structure of social affect: Attitudes, emotions, sentiments, and the case of “contempt”[J], Behavioral & Brain Sciences 40 (2017), 1–18.

Mohamed

, The Relation of Artificial Intelligence with Internet Of Things: A survey, Journal of Cybersecurity and Information Management 1(1) (2020), 30–24.

Mohamed

, A comparative study on Internet of Things (IoT): Frameworks, Tools, Applications and Future directions, Journal of Intelligent Systems and Internet of Things 1(1) (2020), 13–39.

Sheeba

and Raj

T.A.B.

, Optimal Weighted Average Prediction and Correction in Big Sensor Data Using Fruit Fly Algorithm and Support Vector Machine[J], 14(7) (2017), 3615–3623.

Krishnaraj

P.M.

, Mohan

and Srinivasa

K.G.

, Performance of procedures for identifying influentials in a social network: prediction of time and memory usage as a function of network properties[J], 7(1) (2017), 34–41.

Xia

, Liang

and Xu

, The Operational Cost Minimization in Distributed Clouds via Community-Aware User Data Placements of Social Networks[J], 112 (2017), 263–278.

10.

Haibo

and Jonathan

J.H.Z.

, Social networks, mass media and public opinions[J], Journal of Economic Interaction & Coordination 12(2) (2017), 1–19.

11.

Iván

, Ruth

and César

A.C.

, An Approach Based on Social Network Analysis Applied to a Collaborative Learning Experience[J], IEEE Transactions on Learning Technologies 9(2) (2017), 190–195.

12.

Rocca

P.D.

, Senatore

and Loia

, A semantic-grained perspective of latent knowledge modeling[J], 36 (2017), 52–67.

13.

de Araujo Koury

Luisa Corrêa

, Ganser

Arnold

and Berliner

Nancy

, Treating acute promyelocytic leukaemia in Latin America: Lessons from the International Consortium on Acute Leukaemia experience[J], British Journal of Haematology 177(6) (2017), 134–145.

14.

Alvarez-Zuzek

L.G.

, La Rocca

C.E.

, Iglesias

J.R.

, et al., Epidemic spreading in multiplex networks influenced by opinion exchanges on vaccination[J], PloS one 12(11) (2017), 186492–186499.

15.

Gao

Lin

, Tang

Ming

and Pang

Haitian

, Multi-User Cooperative Mobile Video Streaming: Performance Analysis and Online Mechanism Design[J], IEEE Transactions on Mobile Computing 18(2) (2019), 376–389.

16.

Ventrella

Agnese V.

, Piro

Giuseppe

and Grieco

Luigi Alfredo

, On Modeling Shortest Path Length Distribution in Scale-Free Network Topologies[J], IEEE Systems Journal 9(8) (2018), 1–4.

17.

Rebhi

Wala

, Ben Yahia

Nesrine

and Ben Saoud

Narjès Bellamine

, Hybrid modeling approach for contextualized community detection in multilayer social network: emergency management case study[J], Procedia Computer Science 112 (2017), 673–682.

18.

Appice

, Towards mining the organizational structure of a dynamic event scenario[J], 50(6) (2017), 1–29.

19.

Mullai

, Sangeetha

, Surya

, Madhan kumar

, Jeyabalan

and Broumi

, A Single Valued Neutrosophic Inventory Model with Neutrosophic Random Variable, International Journal of Neutrosophic Science 1(2) (2020), 52–63.

20.

, Wang

, Yan

, et al., MapReduce-based adaptive random forest algorithm for multi-label classification, Neural Comput & Applic 31 (2019), 8239–8252.

21.

Torrecilha

R.B.P.

, Utsunomiya

Y.T.

and Batista

L.F.D.S.

, Prediction of lymph node parasite load from clinical data in dogs with leishmaniasis An application of radial basis artificial neural networks.[J], 234 (2017), 13–21.

22.

Kim

Young-Kyu

, Lee

Dongwon

and Lee

Janghyuk

, Influential Users in Social Network Services: The Contingent Value of Connecting User Status and Brokerage[J], Acm Sigmis Database 49(1) (2018), 13–32.

23.

Jung

Se-Hoon

and Kim

Jong-Chan

, C k LR Algorithm for Improvement of Data Prediction and Accuracy Based on Clustering Data[J], International Journal of Software Engineering and Knowledge Engineering 29(5) (2019), 631–652.

24.

Chen

Zhihao

and Zhang

Zhao

, What Network Topology Can Tell in Election Prediction[J], Discrete Mathematics Algorithms & Applications 10(2) (2018), 1–9.

25.

Franco

Mário

and Pais

Leonor

, Exploring the advantages of using social network sites (SNSs) in dental medicine organizations[J], Int J Health Care Qual Assur 30(4) (2017), 385–396.

26.

Dong

Suyalatu

, Deng

Yan-Bin

and Huang

Yong-Chang

, SEIR Model of Rumor Spreading in Online Social Network with Varying Total Population Size[J], Communications in Theoretical Physics 68(4) (2017), 545–556.

27.

Hassnain

, Yu

, Shahzad

M.A.

, et al., Available recovery time prediction in case of an accident scenario for NPP component[J], 97 (2017), 115–122.

28.

Chen

and Wang

, A congestion Detection and Control Algorithm for Clustering WSNs Based on GM(1,1)[J], Chinese Journal of Sensors & Actuators 31(2) (2018), 276–282.

29.

Baofang

, Wang

Hong

and Yu

Xiaomei

, Sparse network embedding for community detection and sign prediction in signed social networks[J], Journal of Ambient Intelligence & Humanized Computing 2017(1) (2017), 1–12.

30.

Chen

, Uliji

H.W.

, et al., Evolutionary heterogeneous clustering for rating prediction based on user collaborative filtering[J], (2017), 394–399.

31.

X.-X.

, Chen

L.-T.

and Zhang

, A Method to Construct Weibo User Behavior Relationship Network Using Dynamic Cognitioin[J], Journal of University of Electronic Science & Technology of China 47(2) (2018), 262–266.

Application of clustering algorithm in social network user scenario prediction

Abstract

Keywords

1 Introduction

2 Sequential analysis of behavior patterns of social network users

2.1 SocialSitu framework

3.1 SocialSitu framework

Table 2 Comparison of the accuracy of the three recommended algorithms Recommended list length Content recommendation algorithm based on SocialSitu User-based collaborative filtering Popular recommendation algorithms 6 0.320 0.25 0 11 0.550 0.25 0.06 16 0.550 0.145 0.07 21 0.490 0.165 0.12

References

Table 2
Comparison of the accuracy of the three recommended algorithms

Recommended list length Content recommendation algorithm based on SocialSitu User-based collaborative filtering Popular recommendation algorithms

6 0.320 0.25 0

11 0.550 0.25 0.06

16 0.550 0.145 0.07

21 0.490 0.165 0.12