Film Score Data Analysis And User Behavior Preference Modeling

With the rapid development of Internet technology, the phenomenon of "information overload" has become the biggest obstacle for people to get information. It has been attracted much interest in data analysis and knowledge discovery that how to construct User Preference Model through multidimensional information dataset, and then we can describe and make reasoning based on this model to help people get personalized information service.Bayesian network (BN) is one of the most effective theoretical models for representing and inferring uncertain knowledge. Latent variables (LVs) in BN is such variables which are objectively existent but can never be observed in related datasets, BN with LVs is generally called latent variable model (LVM) which can simplify the complexity of model structure, can improve reasoning efficiency and make the relation among these observed variables more clear. Lately there are many approaches that use LVM to model user’s behavior preference, due to LVM’s high efficient and concise representation and reasoning ability of uncertain knowledge. But mostly they used given structure to model user’s preference, so that their ability of representing and inferring uncertain knowledge has been limited.According to this problem, in this thesis we bring up this User’s Preference Model (UPM) based on a LVM with grouping information of these attribute nodes to model user’s preference from Movie-Lens dataset, and then we determine the cardinality of the LV through LCM-learning based on a simplified structure of the UPM. Next we propose a heuristic structure learning algorithm based on domain knowledge instruction called KBL. Finally, we give a rating prediction method by using marginal utility theory for some movies belong to multiple genres.The main work and contributions of this thesis can be summarized as follows:1) We built a LVM with grouping information from Movie-Lens dataset to describe the attributes of user and movie in a certain domain, called User Preference Model, or UPM. We propose a heuristic structure learning algorithm based on domain knowledge instruction called KBL to construct the UPM;2) We proposed an algorithm based on the Clique Tree Propagation for UPM’s inference to get user’s rating about a movie.3) Based on the analysis of the attributes of user, movie and rating data from Movie-Lens dataset, we built the UPM using KBL algorithm, the result of experiment indicates the correctness and efficiency of our method.