Research On Group Attendance Event Prediction Method In Event-based Social Networks

With the rapid development of social media and information technology,Event-Based Social Networks(EBSN)have become a hot research and attention area.In EBSN,the emergence of new events increases the uncertainty when users choose events of interest,posing challenges for event organizers.By accurately predicting events of interest to users,event attendance prediction is expected to improve user experience,optimize resource planning,increase event success rates,and enhance platform stickiness.On the other hand,groups play a crucial role in prediction systems.Group members usually understand and trust each other,and their preferences and interests are often similar.The prediction system can utilize the behavior and feedback of group members to provide users with more personalized and targeted predictions.Users usually trust the suggestions of friends or colleagues more,and utilizing group information helps establish social influence and trust,thereby improving the accuracy of prediction systems.This thesis mainly studies the method of predicting group attendance events,aiming to improve the accuracy of predicting group attendance events,enhance the event organizer’s ability to host events and enhance the experience of participating users.Group attendance event prediction mainly faces the following main issues:(1)EBSN does not have explicit user group information.In order to effectively define and discover closely related user groups in EBSN,it is necessary to consider the interaction between group members online and offline during the group discovery process.A comprehensive consideration of the close connection between online and offline helps to determine the degree of closeness of user groups accurately.When searching for user groups with close online and offline connections,the characteristics of their close connections will impact the group discovery algorithm,and the resulting network structure will also impact the algorithm.Therefore,the critical issue is how to effectively utilize these correlations to improve the accuracy of group discovery.(2)The data reflecting user’s explicit preferences in EBSN is very sparse.In EBSN,combining contextual information can more accurately depict user event preferences.Because users rarely give event ratings after participating in events,their explicit preferences are often very sparse,making it particularly important to obtain user preferences in multidimensional contextual environments.In this case,the critical issue is how to mine and integrate contextual information of different forms and categories from multiple information sources to gain a more comprehensive understanding of user preferences.Meanwhile,the dynamic interaction between users and group members also affects their event preferences.Users may be influenced by their friends during the event decision-making process,leading to preference changes.Therefore,the critical issue is to consider the dynamic influence among group members and adopt methods to ensure accurate acquisition of user preferences for events to achieve better group participation in event prediction.(3)The EBSN users have diverse preferences and potential conflicts,making integrating difficult.Group participation in event prediction involves organically combining information from multiple dimensions and data sources to more comprehensively and accurately depict the characteristics and behavior of the group.When conducting effective group preference aggregation,it is necessary to deeply understand the information dissemination mechanism within the group,the social influence among members,the context of events,and the intertwined influence of personal motivation to form group preferences.The user preferences within a group are highly diverse,and members may have different tastes and interests.At the same time,there may be conflicts or differences among group members,making the process of group preference aggregation very complex.Therefore,the critical issue is to adopt a reasonable method to integrate the preference information of different group members to predict group participation events better.To address these challenges,the thesis conducted in-depth research from three aspects:user group discovery,user preference acquisition,and group preference aggregation.The main work and contributions of this thesis are as follows.(1)A maximum k fully connected user group discovery strategy was proposed,and three group discovery algorithms were designed: Mining the Maximum frequent itemset Based Search(MMBS)algorithm,Two Vector Based Search(TVBS)algorithm,and Divide and Conquer Parallel Search(DCPS)algorithm.The MMBS algorithm obtains two user sets by mining the maximum frequent itemsets of user attendanceterest groups and user attendance events in the EBSN dataset.The elements in the obtained two user sets are intersected pairwise,and unsuitable items are filtered to obtain the maximum k fully connected user group.The TVBS algorithm constructs a set of users into an enumeration tree and then searches on the enumeration tree.This algorithm utilizes the user attendance event vector and user attendance interest group vector during the search process,which can quickly calculate the number of shared events and shared interest groups of users.Pruning is performed when the number of coparticipating events and co-participating interest groups of group users do not reach the support threshold,making the algorithm more efficient in obtaining the maximum number of fully connected user groups.The DCPS algorithm first obtains the interaction subgroups of users and then uses the TVBS algorithm to mine the subgroups in parallel.It improves the efficiency of obtaining the maximum number of fully connected user groups.The experimental results on the Meetup platform dataset show that all three algorithms can effectively discover the maximum k fully connected user groups.It was verified that the group attendance event prediction algorithm based on the maximum k fully connected user group performed better than the group attendance prediction algorithm based on interest group member clustering.(2)The thesis proposes two group attendance event prediction algorithms based on preference acquisition: User Preference based Group Attendance Prediction(UPGAP)and Interactive Perception based Group Attendance events Prediction(IPGAP).The two algorithms aim to improve the accuracy of predicting group attendance events by comprehensively considering multiple factors.The UPGAP algorithm considers the characteristics of user attendance in events,such as event content description,event hosting time,and hosting location,and provides a method to obtain user preferences.In the IPGAP algorithm,in order to capture the dynamic interaction process between group members more accurately,a user preference adjustment method based on interaction perception was designed.This method reflects that user preferences are influenced by themselves,their friends,and event organizers,and it can more accurately represent user preferences.Experimental results have shown that compared with other algorithms,UPGAP and IPGAP algorithms perform better in multiple prediction indicators,including average accuracy,average recall,and F1 value,and has higher performance in predicting group attendance events.(3)The thesis proposes a weighted aggregation strategy based on user ratings and a weighted aggregation strategy based on group ratings.Based on these two strategies,two group attendance event prediction algorithms were proposed: User Rating Group Attendance Prediction Algorithm(URGAP)and Group Rating Group Attendance Prediction Algorithm(GRGAP).The two algorithms aim to improve the prediction performance of group attendance events.The URGAP algorithm first focuses on user preferences in predicting group attendance events.Analyzing user behavior and historical data in social networks more accurately captures their personalized preferences.Secondly,the algorithm considers the user’s position in the group and improves the accuracy of group preference aggregation by considering the social structure within the group.Finally,considering the level of user activity better reflects the level of user attendance group events.By comprehensively calculating these features,the algorithm obtains the weights of users in the group,thereby achieving group preference aggregation and providing better group attendance event prediction results.The GRGAP algorithm comprehensively considers the close relationship between users and other group members,the similarity of preferences between users,and user preferences.By analyzing the interaction behavior in social networks and the connectivity within groups,the algorithm captures the strength of the relationship between users and group members.Based on these features,the algorithm calculated the average rating of other members of the group on the user,using the average rating as the user’s weight,achieving group preference aggregation and improving the effectiveness of group attendance event prediction.The experimental results show that the proposed URGAP and GRGAP algorithms have better prediction performance compared to prediction algorithms using other aggregation strategies.This thesis proposes a multi-group participation event prediction strategy based on the characteristics of EBSN.On the one hand,it can effectively improve the accurate prediction ability of event organizers towards participating users and enhance user satisfaction.On the other hand,it provides new ideas for EBSN recommendation and prediction.