Research Of Collaborative Filtering Algorithms Based On Context

With the application and popularization of network technology, the rapiddevelopment of e-commerce, more and more information on the Web is flooded. Peopleby the lack of information into the information age of fast-growing, the era of informationoverload. How to find information suited to their needs in a number of resources tobecome one of the core problems of many scholars which is experts and Internet usersconcerned. Recommended system came into being in such a context.Collaborative filtering technology is the recommended system(RecommenderSystem), one of the most core technology, is currently the most widely used andsuccessful technique, theory and practice in red have made rapid development, it is basedon user selection information and historical similarities relations, collection and evaluationof user interests the same information for other users to generate recommendations.However, the traditional collaborative filtering algorithms faced sparse data, usersimilarity is difficult to measure, in real time and can be adjusted to expand and poor areas,affecting the quality of the recommendation system.We use collaborative filtering algorithm recommendation system for the study of thegoal, to address these issues, collaborative filtering algorithms as well as a comprehensiveintroduction to the corresponding improvements. The main work is as follows：describesthe current mainstream traditional algorithms, mainly based on matrix decompositionalgorithm and algorithm based on probabilistic models or field. We focus on thecollaborative filtering algorithm based on user-based collaborative filtering algorithmbased on the article, and the corresponding improved algorithms for which they are givento improve the quality of recommendation. Recommendation system for cold-startproblem, according to the algorithm the user registration information to solve the coldstart problem. Because content-based filtering algorithm can not usually present a flexiblecombination of various useful information (such as user interest, etc.), and collaborativefiltering algorithm depends on the display or implicit rating data. Recommendationalgorithms often are faced with a cold start (how to new users recommendations and howto carry out new projects to recommend to the user), data sparseness, the algorithm scalability problems. Proposed use UPFM matrix decomposition method recommendedcontextual advertising, not only in the use of algorithms to predict this new methodcompared to other matrix decomposition algorithm has improved the accuracy, but in theface of sparse data, the amount of data has better performance when these issues.