The Study Of Collective User Behaviors On Social Media Under Emergencies

Benefiting from the development of Internet technology and the popularity of so-cial media,the study of the collective behavior of large-scale users has become possible.In recent years,social media has begun to show its prominence in response to disaster events due to its real-time nature.Compared with traditional communication methods,taking advantage of the collective wisdom generated on social media can have a positive impact on disaster relief behavior;on the other hand,the irrationality,infectivity,and conformity features of online users can easily lead to the occurrence of Internet pub-lic opinion.Therefore,the collective user online behaviors under emergencies should receive special attention.The dissemination of information in social media and the interaction between in-dividuals depend on the social network behind it.The behavior of collective users in social media has a natural connection with the large-scale network researches.The concepts and methods in the network science were utilized in the dissertation to study various collective behavior patterns of social media users under emergencies,moreover,the dissertation explores the effectiveness of online improvised logistical system under emergencies.The research results are helpful to guide users in regulating their own behaviors in the event of an emergency,helping users to improve the efficiency of in-formation acquisition,and assisting relevant departments in improving their regulatory capabilities.The results could provide new insights for studying large-scale collective behaviors of online users in emergencies.The detailed contents and main results of the dissertation are summarized as follow:1、The study of the dynamics of the public’s attention under emergencies.By employing hashtags as a proxy for users’ attended topics,the dissertation proposed an attention graph model to study the collective attention shift process under multi-ple exogenous events.The results reveal five salient patterns of collective attention under emergencies:expansion,topic agglomeration,attention concentration,mutable-ness,and re-mixing.Meanwhile,the collective attention shift patterns are related to the nature of the event itself:collective attention with asymmetric activity patterns before and after the peak is more likely to be associated with unexpected events;the gradual build-up of attention before reaching a peak usually can be seen in scheduled social events.2、The user sampling problem of efficient public opinion monitoring system.Due to the massive users of the social media,an effective user sampling method can improve the efficiency of public opinion monitoring.The dissertation proposed a sampling cri-teria that simultaneously measures the activeness,connectedness,and adaptiveness of a user.By transferring the user sampling problem into a new graph sampling problem,a random walk based graph sampling algorithm was proposed.The results show that by balancing between user diversity and user similarity,a small subset of users can effectively represent the attention dynamics of the overall user set.3.The study of topic prediction in emergencies.By using hashtags as represen-tations of user’s attended topics,the dissertation used the heterogeneous information network that simultaneously describes the user-hashtag bipartite relation and the co-occurrence/shifting relation between hashtags.Based on the heterogeneous graph,a bi-directional resource diffusion algorithm was proposed to compute the similarity be-tween users.Experimental results show that taking advantage of the user influence diffusion process in the hashtag network can greatly improve the prediction accuracy;using the hashtag co-occurrence/shifting relationship can effectively avoid the aggrega-tion of user nodes in the embedding space when using the graph embedding algorithm.Compared with the graph embedding method,the algorithms based on user similarity is more suitable for the scene where the user’s attention is concentrated.4.The study of the effectiveness of the improvised mutual assistance system in emergencies.In recent years,there has been a new trend in the use of social media under disaster events:When emergencies occur,users spontaneously collaborate with each other to carry out targeted offline actions.The dissertation proposed a content clas-sification scheme for such a system and studied the dynamic trend of each information category.The ability of automatic classification algorithms to detect the key informa-tion in the system was also explored.The results show that the online self-organizing system is filled with a large amount of noisy data.Meanwhile,significant information overload makes it difficult for resource seekers to discover related resource information.Compared to the noisy information.the users have the intention of broadcasting the key information through retweeting behavior,but the user’s attention distribution is highly skewed.The experimental results of automatic classification algorithms show that the use of word2vector model and contextual features can effectively predict the categories of texts.