Study On Document Recommendation Strategy Based On Query Network

With the increasing popularity of Internet technology, the web information explorsively grows. Because of a great wealth of online information resources, personalized information services have been drawing increasing attention. People’s purpose of accessing to information gradually shift from the recall to the precision, satisfying the specific needs of specific people has become the service target in new network environment, Under which the intelligent recommendation system appears and it can not only recommend documents and products but also guide customers to concern their interested information or document about some product information.The core of the recommendation system is the recommendation strategy, which is the focus of our study.Under the analysis of users’query behaviors and mining users’interest preferences, the query network graph is constructed. And a document recommendation model based on logs of users’queries, which combines the advantages of the traditional recommendation methods such as collaborative filtering-based and content-based recommendation methods while avoiding their shortcomings.This paper proposes a novel entity description model based on Click-Through. This model describes users and documents using keywords and index terms, which not only reduce the dimensions of vector space but also express users’interests and intensions, therefore it can be better apply to the entity description in recommendation systems. Due to the constantly changing of users’interests, in different time users’query interests may be different, thus this paper uses a behavior-interest vector to describe the interests of different queries.Based on the query network, this thesis proposes the recommendation algorithms based on the energy transfer. Every member in coordinated user group owns corresponding recommendation energy, who assigns the energy to documents along which the query networks, At last top-k documents with high energy are recommented to users.In the end, this thesis experimentally evaluated the query accuracy, coverage, popularity and diversity of the strategy. By which, the effectiveness is demonstrated...