Research And Implementation Of Recommendation Algorithm Based On The PLSA Model

With the continuous development of Internet, access to information has gradually changed, but in the face of the vast sea of the Internet resources, people often feel helpless, and even panic, we want the computer to be more intelligent and proactive as we filter out valid resources, and make us easy access to useful information in their own needs. In this paper, personal recommendation technology unprecedented attention, through information and user information and other content of the site, using filter algorithms, data mining and machine learning, allow the system to active users "think".Collaborative filtering is the most widely used and most successful recommendation algorithm, which thinks the user’s behaviour with general contact, user interaction and can be used to predict user preferences and interest of the recommended results. But it’s not easy to dig users’interested in character flaw thus could allow the limitations on the accuracy., And collaborative filtering there are still two more difficult to solve problems,cold start issues, and scalability of the project, based on these three issues, the following aspects of work we do. We will model for latent semantic analysis using collaborative filtering algorithms and statistics knowledge through the use of user score vector space a potential vector space, that is, characteristics of a user or project space, and do not need to use external knowledge, to discover the potential interest of the user collaborative recommendation. Through the introduction of probabilistic latent semantic analysis model, a good solution to the scalability issues of the recommended system is proposed. A large number of facts show that single response model does not accurately represent user behavior and make a recommendation. This paper first studies on collaborative filtering algorithms, on probabilistic latent semantic and introduce collaborative filtering recommendation algorithm, on this basis, with mixed recommend advantage, respectively PLSA-CF model and item-based collaborative filtering recommendation algorithm fusion algorithm based on content and together. The later is called return to recommend potential groups, like collaborative filtering system using all user ratings data, but also have the advantage of content-based recommendation system, such as recommended by the project provide the recommended interpretation which is well understood by users, and new projects are recommended.Experiment results show that PLSA-CF model outperform the traditional of collaborative filter, and the model mix of recommendation outperform a single collaborative filtering recommendation and has been given a considerable increase in the accuracy; returned potential groups recommended system not only recommend new projects and can provide very good recommend explain,improve recommended system of transparency,and are better of user experience.