Method Of Constructing And Updating User Interest Model For Document Recommendation

With the diversification of user needs, personalized recommendation systems are applied not only for e-commerce, but also for recommended pages, movies, documents, and so on. In order to help people to obtain documents they need accurately and conveniently, research on document-based recommendation technology becomes necessary. Document recommendation system collects and analysis of the documents people already read. Then the system understands the users’ interest through those documents and builds user interest model base on those documents. The documents which are highly matching with the model will be recommended to the users.In document recommendation system, the module of user interest model is one of the key modules. In process of model constructing, the document for model constructing, which is represented directly by the vector from the document getting characteristic feature words, only takes the same point on the word form but ignores the same point on the meaning of the word. Then the system can only recommend the documents with the same word form, but can’t recommend the ones with the different word form with the same meaning. The recall would be low. The information user already read will increase with the time of using the system. When the user interest model updated, the computation of extracting the user interest model from the documents for model constructing will become a great amount. The system responses slowly, which is not conducive to the user experience.To solve the above problems, this paper has done the following tasks:1) Designed a document recommendation system, and described the overall structure and the various modules of the system. This paper participated in system frame,focused on user interest model module.2) Proposed a method to construct user interest model based on semantic relevance between words. Generally believed that every word has multiple meanings, depending on the word itself is unable to determine the correct meaning of word. But the meanings of the words in the same text are interrelated. It means that the words may have similar hypernym concepts(the more general concept). The main idea of the method is to determine the correct meaning of the word based on the semantic relevance between words of the document, and then select hypernym concepts of the correct meaning to construct user interest model. This paper proposed a method to recognize words’ correct meaning based on semantic relevance between words. Every meaning of a word has a set of hypernym concepts inherently, so the semantic relevance between words also reflects in the relations of their hypernym concepts. Then the method of recognizing words’ correct meaning utilized the relations of their hypernym concepts to determine the correct word meaning. Finally getting the right hypernym concepts of the correct word meaning through the recognizing words’ correct meaning method, and using the hypernym concepts represented the word of the document to construct the document vector, based on those work construct user interest model.3) This paper studied on rapid updating method of user interest model. Usually updating model add(or delete) information in the previous model. The incremental update was based on the previous intermediate results stored by system of generating the previous user interest model to avoid redundant counting, to achieve rapid updating purposes. The general idea of rapid updating method for the model was described. And three kinds of user interest model rapid updating method were realized. In the structure of document recommendation system, this paper realized a incremental updating user interest model based on semantic relevance between words.Finally, the experiment shows that the user interest model constructing approach based on semantic relevance between words proposed in this paper has a higher recall rate. The incremental updating method in this paper has low computational complexity and faster updating speed advantages over the original update method.