Research On The Personalized Ranking Based On Similarities Between Users In Social Tagging System

With the development of the information technology and the advent of big data era,the resources of the web are exponentially increasing,and the search engine has become an important way for users to find resources from the network.In recent years,due to the rise of Micro Blog,We Chat and other We Media,the amount of information in the web has increased dramatically,which makes the traditional keyword based search results cannot satisfy the needs of information retrieval in the age of big data.Therefore,the personalized information retrieval technology is a hot point in the currently research.It can filter the original search results according to the user's interest preferences so that different users can get different search results.With the advent of the Web2.0 era,the social tagging system has sprung up,and many scholars have begun to explore the application of social annotation in personalized information retrieval.Many research findings revealed that social annotation can help improve the effectiveness of information retrieval,but there is always room for improvement.This dissertation mainly explores how to use the s ocial annotation to obtain the interest of the user accurately to further improve the ac curacy of personalized search.There are two main types of implementation strategy of personalized information retrieval: query expansion and result re-processing.In this paper,Social annotation information can effectively construct the user interest model and reorder the original search results according to the user's tastes and preferences.In the existing personalized ranking methods,the similarity between users is mostly calculated by using a common set of tags.However,in the real social tagging system,the user similarity calculation is not accurate becau se the number of tags and pages marked is limited.On the base of the existing research study,two contributions are made in this thesis:(1)A personalized ranking method utilizing tags and user similarity network is proposed.This method builds a tag sim ilarity network through the co-occurrence relation between tags,and uses the tag's similar network to find the tag synonyms and is used to extend the user's annotation information to the document.The user similarity is calculated by combining with the si milarity of users on the level of the tags and documents.The user similarity network is constructed based on the user similarity so that the scope of the user's interested documents in the query is extended according to the user's similar network,which c an improve the accuracy of personalized retrieval.The experimental results show that the method can improve the effect of personalized information retrieval to a certain extent in the case of reducing the time cost.(2)A kind of personalized ranking meth od utilizing similarities between topic domains is proposed.This method first separates the web pages and tags with different themes based on the division of the topic domains,and finds the tag synonyms by the constructed tag similarity network.And then it finds similar users according to user tags and theme preferences and expands the user's social annotation,which can better improve the effectiveness of information retrieval.The experimental results show that this method can effectively alleviate the problem of sparseness and tag synonyms,which can improve the user search experience.