Research And Application Of Personalized Recommendation System Based On Mixed Mode

With the popularity of the Internet and the development of global information process, the network has become the main way of people getting information; at the same time, resources are becoming grown explosively. Then users often feel it is very difficult to find information which is really valuable for them, and some information which is rarely concerned about will be easily become isolated. The personalized recommendation system can solve these problems effectively, it analyzes the user's characteristics and interests in buying behavior, and then recommends them information that they are interested in and satisfied with in order to help them make the right decision. But the current e-commerce recommendation system is not mature with some general drawbacks exsisting in recommending efficiency, quality, automation and so on. Four aspects will be discussed and researched in the following thesis.This thesis studies the mainly personalized recommendation technology, and emphasis on content-based filtering and collaborative filtering technology to study and compare their strengths and weaknesses and to explore the combination of two recommended techniques for mixed-recommended ideas, on this basis put forward a combination recommendation engines under the framework model.The research of interesting models and the clustering of users has been the focus, and this thesis will put special attention to the interesting model. Survey shows that most users are willing to provide their name, gender, occupation and other less sensitive personal information to the website, and users with similar backgrounds always have the same interests. In addition to this, users' score and query keywords can effectively represent the their interest, therefore, we can combine the three together to establish the users'interest model, then calculate the similarity between them, and cluster users in the off-time, thus greatly shortening the online recommend time and improve the recommendation system in real time.Respond to the quality drawbacks of current personalized recommendation system, this thesis will use tags to mask the content properties and combine with the concept knowledge based on specific domain, to calculate the items' similarity fully utilize their properties, thus solve the problem of cold-start, then combined with content filtering technology to forecast the missing values in user-item matrix, so as to solve the data sparse problem and improve the quality of recommendation.Lastly, design and implement a personalized recommendation system based on mixed model. The prototype system which based on multi-model recommendation engine will provide personalized, top picks, new items, etc. recommendation, integrated the keyword searching and user ratings functions to help users get access to personalized information via different levels, in order to enhance the user experience.