Researches For Recommendation Methods Based On Topic-oriented Users' Interests

The fast development of information technology and deeper Internet popularization have caused a global social network environment,the huge amount of information which made by the user bring about a lot of worry when the user wants to make more like-minded friend.Thus it can be seen that the research subject is very necessary.To solve the microblog platform encountered above problems,this thesis is described mainly in two points.On the one hand,the user's characteristic of interest is reflected by the attribute information and the labels of interest that are extracted when they registered for an account,but these information can not be used to reflect the user's interest at present.However,users' microblog content written recently can reflect their current interest.On the other hand,the recommendation algorithm of FOAF uses ‘common friends' as the indicators,it still ignores the difference among candidates who have the same ‘common friends' with target user.In order to considerate both precision and acceptance.The model in this thesis study the friends recommendation problem integrating the topic interests with social interests.In this thesis,the main contents includes as following:1.In order to reduce the limitation of the recommendation results which based on social relationships and to solve the defect of low accuracy of the result which based on contents,an algorithm called topic-oriented recommendation research based on user's interest is proposed2.In the process of analyzing people's social relationships,the prior probabilities are used to compute probabilities recommending candidates with the different number of common neighbors to target users.Words co-occurrence technology and k-core analysis technology are introduced to the recommendation algorithm based on text content in order to remedy the deficiencies of ignoring word semantics in traditional word vector space model.Then users' interests are measured by the feature of keyword distribution.3.An algorithm called improved topic-oriented recommendation research based on user's interest is proposed to solve the disadvantage of ITOR First,to reduce the dimension of keywords matrix,the MTC algorithms is used;then the semantic environment is considered based on words co-occurrence technology to calculate the topic similarity of user's;the degree of interaction between users is merged into the algorithm based on common neighbor friend recommendation to compute the social similarity of user's.;At last the linear weighting method calculated with interest and social similarity is used as a judgement for recommendation.Finally,we use microblogging data to verify the proposed algorithms in this thesis and experiments proved that the performance have improved in such precision,recall and F-measure respect compared with other typical methods.