Research On Twitter Event Recommendation Method

In today's information age,social networks play a very important role in people's life.Not only do they enrich our daily life,but also provide a fast and powerful way of information dissemination.People can publish their own dynamics on social networks,and they can also learn about what is happening around them and around the world.However,while satisfying people's information needs,the massive information in social networks brings the problem of information overload.As a result,how to efficiently obtain information needed by users in massive information has become a popular research direction in the data mining field.The personalized recommendation system can effectively solve the above problems.Compared with traditional search engine,it can actively push information to users,and it doesn't have problems like poor search quality,too commercial search bids,and too many spam advertisements.Therefore,in order to help users to obtain useful information in social networks efficiently,this thesis selects Twitter,one of the most popular social networks,as the research object and studies the personalized Twitter event recommendation algorithm.The main contributions of this thesis include:(1)This thesis studies the recommendation algorithm of Twitter data in an innovative way.It no longer regard single tweets as recommend objects.Instead,it clusters the similar tweets in order to detect events which are happening in Twitter,and then regard events as the basic unit for recommendation research.The algorithm uses the idea of PCA algorithm to reduce the dimensions of tweets,analyzing the composition of tweets,removing redundant information and retaining the main useful information.This method can reduce the subsequent computational complexity and also improve similarity calculation accuracy to some extent.In addition,an improved cosine angle algorithm is used to calculate the similarity,taking the influence of vector dimension into account.This thesis analyzes the drawbacks of this fixed threshold method,and then proposes an adaptive threshold discovery method which can find the most appropriate threshold for every user model in order to increase recommendation accuracy and recall.Based on adaptive threshold and component filtering,a Twitter event recommendation algorithm is proposed.At last experimental verification is carried out on real Twitter dataset.The results show that compared with the traditional recommendation algorithm,the proposed algorithm in this thesis has greatly improved the recommendation performance.(2)For online recommendation,a Twitter event recommendation algorithm based on secondary clustering is proposed.The algorithm re-cluster the user model and events.And then by evaluating the clustering result we determine whether the event is of interest to user.If the user model and some of the tweets are clustered into sub-categories,the event is considered to be of interest to the user and should be recommended.The algorithm uses the K-Means++ algorithm for clustering analysis,which optimizes the initial clustering centers compared to K-Means algorithm,in order to get the global optimal solution.As experimental verification,it shows that the algorithm proposed in this thesis has better recommendation performance than traditional recommendation algorithm.(3)Based on the recommendation algorithm which based on secondary clustering,a Twitter event recommendation system is designed and implemented.First,the entire system architecture was designed.Then the implementation method of each function module was introduced in detail.Finally,the recommendation results were displayed in a visual way.