Design And Implementation Of A Recommendation System For Mobile Internet Application Market

The nature of the recommender system is to establish the relationship between users and items, and recommend a specific user the items which he may be interested in. The recommender system tries to predict users’future behavior by excavating the history of users. As an important mean to solve the information overload, the recommendation algorithm works the following three steps:1. collect the users’ behavior history;2. establish an impeccable recommendation model;3. generate the final recommendation result. In the process of establishing the recommended model, the traditional methods often ignore the impact of time effect and the changes in users’ interest. All of those lead to recommend unreasonable and inaccurate results and other issues.Firstly we summary a fixed paradigm model for the most of the improved recommendation algorithm. This paradigm can guide our improvement ideas and goals.By meticulous observation in mobile applications market’s real data, we found the presence of a large amount of batch download. The behaviors have a strong correlation with each other in the same batch download. We quantify the time interval between batches, by the method of hierarchical clustering.We propose an improved collaborative filtering recommendation algorithm based on time effect. The improvements:1. improve the traditional collaborative filtering algorithm in computing performance by using positive and negative index table;2. improve the accuracy of the traditional collaborative filtering algorithm by introducing time hot decay function. Experiment in actual mobile application market data between the traditional collaborative filtering algorithm and collaborative filtering algorithm based on improved time effect. After adding time effect is verified by the experiments, the recommendation accuracy has improved. We propose an improved random walker recommendation algorithm based on time hot decay effect. The improvements:1. improve the traditional random walker algorithm in computing performance by using matrix calculation optimization method;2. improve the accuracy of the traditional random walker algorithm by introducing time hot decay function reflecting the users’changes in interest. Comparing experimental results show that the algorithm has greatly improvement on the recommendation accuracy and computational efficiency.