Similarity Of Collaborative Filtering Algorithm Based On Weighted Information Entropy

With the rapid development of information technology and the World Wide Web, the information and resources on the network grow exponentially. People from the era of lack information gradually step into the era of information overload. In this day and age, regardless of the information producers or consumers of information encountered a huge challenge:For the producers of information, how to make their own stand out in the mass data and be concerned to the majority of consumers is very difficult; for information consumers, to find the information they are interested in the mass data is also difficult.Recommended system is an intelligent personalized information service system. With the help of user modeling techniques, and then through some recommended strategies to achieve targeted personalized recommendation, recommended system is an effective means to solve the information overload. Recommended system has been widely used in areas such as business, news reader, video, film and television, digital libraries and so on. In recommendation systems technology, collaborative filtering recommendation algorithm is easy to understand, easy modeling and well recommended effect, is the most widely used and the most popular technology.In the current background of most collaborative filtering recommendation system user rating data sparse. In this paper, to further improve the accuracy of collaborative filtering algorithm recommended as the goal, the main research contents are as follows:1) Recalled the development history and current status of the recommendation system, focuses on collaborative filtering algorithm recommended steps, detailing the collaborative filtering two categories:memory-based collaborative filtering and model-based collaborative filtering, summed up the collaborative filtering algorithm exists issues and Challenges.2) Similarity calculation as the core of collaborative filtering step, this paper focuses for the similarity calculation link. Found that in the case of data generally sparse, traditional similarity calculation method is easily exaggerated or narrow similarity, cause inaccurate recommended results.3) After analysis the causes of the problem of the traditional similarity calculation method, a new similarity calculation method:the similarity calculation method based on information entropy has been proposed. information entropy to calculate the score difference to calculate the similarity, by calculating the information entropy of score difference to calculate the similarity.4) Taking into account the differences in user ratings and the degree of user evaluation overlap, weighted the information entropy formula with rating differences and evaluation overlap degree, a similarity calculation method based on the weighted information entropy is formed.5) Using MovieLens dataset, we do several experiments to verify the multiple evaluation indexes. Experimental results show that the similarity calculation method based on the weighted information entropy alleviate the inaccurate problem the traditional similarity calculation method, and improve the accuracy and quality of recommendations.