Research On Electronic Commerce Recommendation Algorithm Based On User Behavior

With the rise of the Internet industry, man’s basic necessities of life can settlement by the mobile phone or computer. More and more transactions happen from offline to online, Internet plus has penetrated into all aspects of people’s lives, people’s quality of life has improved. At the same time, every minute of every day, the Internet also produce huge amounts of data, in the history of mankind never received such a large amount of data in a short term, these data show the hint of people’s willingness to buy, the direction of public opinion and some valuable information, if use these data correctly, we can produce more high-quality services.How to dig out valuable information from these data? It is clear that only by manpower is unable to complete, which requires some recommendation algorithm to support. In this paper, we first introduce some existing recommendation algorithm, then we analysis the user purchase behavior. Common recommendation algorithms are based on content-based recommendation algorithm and collaborative filtering recommendation algorithm. Recommendation algorithm based on content generate a user preference document by the historical data, then compare the user preference document with recommendation projects, select the most similar project and recommend to the user; collaborative filtering recommendation algorithm’s basic idea is to find the same kind of users, and the users’ existing features to recommend. Collaborative filtering algorithm is proposed based on the following hypothesis: if the users get a similar score of a part of the projects, then they are also roughly the same score to other projects. These two common recommendation algorithms have their own advantages, however, due to the huge amount of data in the field of electronic commerce, and different users may buy very different goods, so the score matrix is sparse, the original algorithm needs some improvements.The original collaborative filtering algorithm need each user to generate a score matrix for the projects, in other application area matrix score can be very intuitive. For example if we predict the commodity score, score evaluation matrix is the user evaluation of goods which is generally the numbers from 1 to 5. But in the electronic commerce application scenarios user score on the project is not so intuitive, user ’s score of goods mainly influence by user’s behavior, but only according to the behavior to estimate user score of goods is not enough, sometimes user behavior frequency can reflect the user’s preference for a product in a certain extent, and therefore should be given a higher score, so the frequency should be as a factor to consider. In this paper, we put forward an improved collaborative filtering algorithm based on the above two points.In order to solve the data sparseness of collaborative filtering algorithm, and improve the accuracy of the recommendation, we need to merge other recommendation techniques to form a recommendation algorithm that can provide a higher accuracy rate and recall rate. Some studies show that the hybrid recommendation strategy is better than the single recommendation strategy. This paper analyzes the feasibility of the hybrid recommendation algorithm, and gives a hybrid recommendation algorithm based on the behavior model of e-commerce users. The experimental results show that the proposed algorithm has good recall rate and accuracy rate.