Research On Semantic-Based User Modeling And Its Applications

With the popularity of Internet and the development of information technology, the quantity of information increases in a geometric growth and the disorder and decentralization of information is aggravated further, which makes it difficult for users to find information needed in the huge amount of data. With the gap between information requirement and information acquisition, it is urgent to provide a new service mode for users, which can organize and regulate information according to users'interests.Personalized information service (PIS) is a kind of services that can supply information to users according to their interests. The quality of PIS is dependent on not only the technology of retrieval and recommend, but also the technology of user modeling, especially the latter is more important. The technology of user modeling is the key of PIS.At present, user modeling has some shortcomings such as lack of semantic in user model representation, difficulty in user interest acquisition and user model evolution. Try to solve these problems, the dissertation attempts to use behavior psychology theory to analyze the characteristics of user interest, and applies the semantic technique in user modeling, does researches on user model representation, user interest acquisition, user model evolution and semantic matching etc. The main work of the dissertation is as follows:1. An ontology-based representation method of user model is proposed.Utilizing the ontology's advantages on semantic analysis and representation, the dissertation proposes an ontology-based representation method of user model to solve the problems existed in traditional user model representations such as lack of semantic, no standard etc. The dissertation has designed a user model, domain ontology and document ontology in a computer science paper retrieval system (CSPRS). With the support of domain knowledge and representation standardization, the precision and compatibility of user model are improved obviously.2. The methods for acquiring user interest and quantifying degree of interest (DI) are proposed.Traditional explicit and implicit quantification methods for quantifying DI are easy to disturb the behavior of users, and rarely consider the psychological characteristics of user interest. According to the behavior psychology theory, the dissertation analyzes the inner correlations between interactive behavior and user interest, and proposes an acquisition method for acquiring user interest based on interactive behavior. According to the regularity of user interest, the dissertation proposes a quantification method based on logistic model for quantifying DI. In this method, time spent per page (TPP) is taken as measurement. The experiment result shows that this method has a good quantification effect.3. An evolution method of user model based on unidirectional spreading activation model is proposed.In traditional evolution methods, the adjustment often happens in the single changed node, these methods rarely consider the semantic relation between nodes. According to the characteristics of user model, the dissertation proposes an evolution method of user model based on unidirectional spreading activation model. This method updates the correlate nodes by restricting the direction of spreading and controlling the attenuation of strength, and achieves the evolution of user model, which lets user model reflect the transformation of user interest immediately.4. A computation method for measuring semantic similarity between user interest and document is proposed to rearrange the retrieved papers, and a computation method for measuring semantic similarity between paper and paper is also proposed to aggregate the reduplicate papers.The traditional methods measure semantic similarity by computing semantic distance between nodes, which are difficult to ensure the accuracy of computation. The dissertation integrates the DI of node with distance and level, proposes a computation method for measuring semantic similarity between user interest and document. The method is used to rearrange the retrieved papers. Moreover, the dissertation proposes a computation method based on paper properties for measuring semantic similarity between paper and paper. The method is also used to aggregate the reduplicate papers. The experiment result shows that these methods can improve the quality of PIS effectively.