Research On Tag-based Personalized Recommendation Algorithm

With the rapid development of Internet, it produces a large amount of information. How to find valuable information from these vast amounts of information in order to meet users'needs has been a hot issue which is concerned by researchers. The traditional information search technology can not well meet the user needs in this area, therefore, personalized recommendation system gradually rise. Collaborative filtering recommendation is one of the most widely used recommendation technology, which has achieved a rapid development both in theory and practical application. But the traditional CF also faces many problems, such as data sparsity, cold start, single-interest model, difficulty to measure similarity and scalability problems, which can affect the quality of the recommendation system. With the emergence and popularity of Web2.0, Social Tag arose at this time. Social Tag is different from the general key words, it can accurately analyze the characteristics of information resources and the preferences of users, it can constitute a network of relationships with users and resources.To implement the personalized information recommendation service, it provides a new solution, with better research and application values.After in-depth research and analysis, the traditional collaborative filtering algorithm,with the introduction of Social Tag semantic similarity, is improved in this paper. Firstly,considering the semantic information of tags, this paper analyzes the characteristics of social tag, and finds the problems of ambiguity tags and low-matched synonymy tags. Than a method is proposed, which measures the semantic similarity of tags by combining WordNet with co-occurrence distribution. It represents the correlation of tag both from the perspective of semantic and statistics, and expands the relationship between tags, items and users, thereby improving the coverage of information. Secondly, based on the relationship between tags and items, a collaborative filtering based on semantic similarity between tags and items is proposed. We can take use of the tag semantic similarity and establish the links between tags and items according to the tag set associated with the item. Then we can get the associations between items. The score matrix is filled by these associations to alleviate the problem of sparse data and improve the predictive accuracy of algorithm. Thirdly, according to the relationship between tags and users, a collaborative filtering based on semantic similarity between tags and users is proposed. The links between tags and users are established the links between tags and items according to the tag set associated with the user. We rebuild the user interest model by using tags and ratings and measure the user similarity based on the tag similarity. Then user-nearest neighbor set is generated and recommendation is given. Finally,we use MovieLens data sets to validate the improved collaborative filtering algorithm, and compared with the traditional ones. The experiment result is evaluated both in the accuracy of predictive rating and item recommended ,and it shows that the collaborative filtering algorithm based on tags semantic similarity has better quality than the traditional ones.