Research And Implementation Of Personalized Recommendation Algorithm Based On User Segmentation And Combination Similarity

With the continuous development of mobile Internet technology, people have entered the "information overload" era in advance. However, the real needs of users is buried in the mass of data, how to provide users with efficient and valuable information and services has become one of the academic direction for further research.For China Mobile Phone of certain province,whose management system recommended reading method has a problem, this paper presents a personalized recommendation based on user segmentation strategy recommended in combination with the recommended method for sparse and cold-start problems province of China Mobile to achieve a personalized mobile reading platform. The main work is as follows:1. presenting a new segment-based approach recommended by Item-Based user.New Item-Based user segments based on the recommended method, based on the user’s reading books, which put users into two categories- the depth reading users and non- depth reading users; at the same time, in order to improve user loyalty and reduce churn, the depth reading user has been classified into high value user groups, user groups and the low value of the value of user groups, we can apply different strategies recommended personalized book recommendations. Experimental results show that the method recommended quality has improved, and the user clicks on the purchase rate than in the past there was a marked improvement.2. presenting a new segment-based approach recommended SlopeOne users.The new difference of user segments recommended method between SlopeOne and Item-Based is that, for low-value customer, it is not necessary to segment, but directly use SlopeOne to recommend. On the recommendation performance, user clicks and purchase rate, which have a very good improvement and upgrading.3. presenting a new method based on association recommended a combination of similarity.With the implement of a combination of association rules recommended and collaborative filtering mode, this method can deal with sparse data and efficient cold-start problems, it has good practical value.4. The user is given a new method to enhance the activity.By selecting the target user, read Internet users read content preferences and reading method for calculating time mobile Internet users to build a user preference calculation for silencing activity of the user with the new user upgrade method. Under the same test data and test environments, silent user activity has been improved significantly.