Research And Application Of Overlapping Community Detection Based On User Interest

In the new era of the web when the number of netizens is increasing rapidly,people's communication and information transmission have overcome the limitations of time,space,and geographic location.This has led to a great extension of users' social networks and a full expansion of users' content information.The social network structure contains the relationship between users,and the content information contains the user's interest.This thesis focuses on topic modeling and overlapping community detection algorithms,taking Sina Weibo platform as the research object.In terms of the topic model,this thesis analyzes the problems of Weibo data,such as sparse information content and distracting factors such as emojis and spam.This makes the theme model perform poorly on the refuted data set,which is not conducive to the extraction of user interests by the theme model;in terms of overlapping community detection,due to the expansion of the number of users and the addition of information content,the original algorithm that simply discovers the community structure no longer meets the needs of the study.Therefore,this thesis proposes a method of overlapping community detection based on user interests.The main contents of this thesis are as follows:Firstly,this thesis proposes an interest extraction model based on the topic model to solve the problems of Weibo data set.Firstly,the model merges users' texts to reduce the impact caused by text sparsity;then preprocesses the data,removes interfering information and then divides the data into words,then uses keyword extraction techniques to construct custom stop words and removes stop words;finally,it uses the topic model to obtain the interest distribution of users.Experiments show that the model outperforms the traditional topic model.Secondly,this thesis proposes an overlapping community detection method based on users' interests to solve the problems of overlapping community detection algorithms.Firstly,it's based on a local extension algorithm in order to discover overlapping community structures;then,it defines relationship similarity,interest similarity,and node centrality to select seed nodes by combining the content information of nodes;then,it merges similar communities to reduce the amount of operations;finally,it divides isolated nodes and merges similar communities.The comparison experiments of this method on several datasets show that it has better performance than other algorithms.Thirdly,this thesis designs and develops a prototype system of overlapping community discovery method based on users' interests by analyzing and summarizing the above-mentioned research contents.The system includes the management function of data,user interest extraction function,and community detection result visualization function.