Research On Online Recommendation From Microblogs Flow Using Improved Influence Of Bloggers

Information recommendation is one of the most important problems in Information Retrieve(IR).With the development of online social network services,research on social recommendation technology become the main direction.On one hand,different from traditional media,microblogs like tweet shows great timeliness.How to find recommendable information in an effective and timely way from massive online blogs flow is a problem worthy of studying.On the other hand,influence of bloggers plays a critical role when recommending in social media services.Therefore,design an effective and precision influence computing model is important for a better recommendation performance.This paper mainly studies technologies of online recommendation of microblog using influence of bloggers as a key feature.We developed an algorithm that measures influence of users taking both pairwise topical similarity and timeliness feature of information into account.First,this paper introduces research background and related research methods of microblogs recommendation.After that,we propose an algorithm computing semantic relatedness based on Word2 Vec model and the classical Vector Space Model in IR.The experimental results shows that our algorithm improved the classification performance of microblog and user's interesting topic,since the Word2 Vec model trained on Wikipedia corpus using deep learning method solved the common polysemy problem in short text.Secondly,this paper developed an EIP algorithm which measures influence of bloggers taking both topical similarity of users and timeliness of information into account.EIP is proposed to identify influencers in social network based on a concept related with user's forwarding activity,we call it passivity.Then it measures influence of users in network based on the relation between influence and passivity,also it considered topical character and timeliness feature of blog.An evaluation performed with dataset crawled from Sina micro-blog shows that EIP outperforms than other algorithms,including the original IP and TwitterRank.Finally,the semantic relatedness between blog and retrieval topic,influence of bloggers and other related features are used in online microblog recommendation based on learning to rank model.We then design and analysis each module of the recommendation system,which contains microblog classification,importance evaluation and pushing.