Research On Movie Recommendation Algorithm Based On Ontology

Considering the spiritual needs of inhabitants and the rapid development of the film, personalized recommendation technology is applied to the film fields in order to change the user habits that obtain the resources by searching to searching unnecessary. However, there are issues like cold starting, scarcity of the user-score matrix, ignoring the semantic association between projects and the change of the user preference following time in the traditional personalized recommendation algorithm. For these questions and the purpose of obtaining lower MAE and improving the accuracy of the recommendation algorithm, this thesis introduces the ontology technology during the research of the personalized recommendation algorithm. The main research contents and innovations of the thesis are as follows:Firstly, film ontology and user model based on the film ontology were built in this thesis. On the basis of the traditional ontology construction method and the iterative development model of software life cycle, this thesis proposed ontology construction method based on the iterative development model. And by using the existing ontology construction tools, the film ontology that applied to the personalized recommendation algorithm was built. Then, the user model based on film ontology was obtained, which added user information to the part nodes of film ontology.Secondly, in order to solve the cold starting of new user, the users were clustered by BK-Means algorithm based on the user background information. This method not only reduced the scope of the searching of user’s nearest neighbor but also effectively solved the cold starting of new user.Thirdly, the traditional Ebbinghaus forgetting function was improved by curve fitting according to user forgetting law in film field and using it to modify the user-score matrix so that the recent ratings by user were more prominent during recommendation.Fourthly, two film recommendation algorithms based on ontology were proposed. In order to solve the scarcity of the user-score matrix, RAFO1 filled the user-score matrix based on the semantic similarity between projects. Then, we find the nearest neighbor set on the modified matrix by traditional user similarity calculation formula and obtained the film recommendation. In RAFO2, the users’ semantic similarity was introduced to improve user similarity calculation formula, in order to find the nearest neighbor sets and obtain the film recommendation based on user preferences nearest neighbor. We can see the RAFO solving the cold starting of the new users by the experiment on Movielens dataset. RAFO1 is helpful for the scarcity of the user-score matrix, but ineffective in improving precision of algorithm. RAF02 can significantly improve the quality of recommendation.