The Research Of Personalized Information Recommendation For Microblog Based On Event Network

The development of online social network platforms has changed the way ofinformation dissemination on internet, but it also brings the problem of “informationexplosion” and “information fragmentation”. It has become more and more difficultfor people to obtain useful information from vast amounts of data, meanwhilediscrete and fragmented information that user receives are more like disassembledpuzzles, which can not present a panoramic view of topics and events. The purposeof information recommendation is to help user solve the problem on usefulinformation access. Traditional information recommendation methods on thenetwork are mostly based on subscription-channels or content classification. Userssubscribe to channels or categories of interest, and then receive content under thesubscription. There are some problems in this way:1. The contents of a channel or acategory that user selected is still broad, users are not interested in all of them.2.Information recommendation can’t be dynamically updated for changes in userinterests.3.It can’t recommend content that users have potential interest. Topic-basedrecommendation can better meet the information needs of individual users. By usinguser’s behavior data from online social networks (including posts, comments andre-posts, etc.) to dig out interest topics of user, and find topic related content forrecommendation, the user can effectively get content that they are interested in fromthe vast amounts of information, and learn the progress of topics from therecommendation. Analysis shows that, news topic is consist of a series of events. Byusing event as basic unit and relationship between events as a description of thedevelopment and changes of events in topic, news text can be represented by anevent network structure. Therefore our study is around semantic text processingbased on event network model: news text are represented by semantic event networkmodel, text content is mapped to events including various elements and relationshipsbetween events. Then we use hierarchical community discovering algorithm on event network for clustering events, the sets of similar events can be used to describetopics. Use those topics to establish the user interest model, then we can realizepersonalized information recommendation. The main contribution of this paper are:(1) Through semantic extension of content on online social network, werealized semantic and proactive information recommendation, provides a new wayfor information recommendation under the era of web2.0;(2) We proposed an event network construction method based on statementanalysis. After text segmentation and POS tagging on text, we find out eventdenoters and event elements based on statement analysis. Then tagging out eventdenoters, event elements and event relations, and finally build up event network.(3) We proposed a hierarchical community discovering algorithm based onevent network for detecting news topics contained in the text. Event network is asemantic text representation model. We use event network model to represent newsarticle, based on the idea of clustering in complex network, we use a hierarchicalcommunity discovering method on event network to realize hierarchical clustering ofevents. Communities discovered in each layer of this algorithm, which are sets ofsimilar events, can be regarded as a news topic. The hierarchical communitystructure created by the algorithm also corresponds with the hierarchical structure oftopics, the final result of this community discovering algorithm can be seen as a setof topics in minimum size.(4) We proposed a personalized interest modeling method based on topics. Aftersemantic text analysis and processing, user interest including topics and events hasbeen discovered. Each topic contains many events, including event elements andevent relations, these rich semantic information are used for modeling user interest.Compared with keyword-based user interest modeling, topic-based modeling havefiner granularity. The semantic relationships between events rise the match of textand user interest model onto semantic level, instead of simple keyword matching.Therefore the quality of information recommendation is greatly improved.