Desigh And Implementation Of Recommendation System Based On User's History And Comments Of Products

Since the early 1990s, commercial enterprises entered the Internet, the huge demand for commercial applications continue to promote the rapid development of Internet. Internet has become a necessary tool for people to obtain information because of its interoperability, ease of operation and low cost and convenience with technological progress, and directly leads to the transformation of human society from industrial society to information society. But at the same time, the advent of the information explosion era makes people face a lot of information every day, and difficult to extract from their valuable part, which makes the effective utilization of information is extremely low. Thus, personalized recommendations for user characteristics and history are called today's important issues.Collaborative filtering recommendation technology is one of the earliest applications and the most successful technologies in existing recommendation systems. However, with the continuous updating of Internet technology and the need of the development of the times, it is necessary to continuously improve the collaborative filtering algorithm and discover the bottleneck in this process, That is, cold start problems and tend to recommend hot items (not recommended long tail items) problem.This article begin from the user browsing records and product reviews information,combine with the traditional collaborative filtering algorithm and the emotional analysis of the product reviews to the user commodity recommendation. Specifically completed the following work:First of all, carried out on a business platform for users to browse records and commodity review information crawl;Secondly, the traditional collaborative filtering method is used to calculate the propensity value of the product to the user.Thirdly, the emotional analysis of the product reviews, the characteristics of the goods and the favorable evaluation of the characteristics of the characteristics, combined with the history of comprehensive consideration, to calculate a user of the goods tendencies;Finally, the user will be twice the weight of the goods tend to value, linear regression is recommended to the highest accuracy of the final recommendation listThe final results show that the recommended accuracy and unpaid goods account for higher than the traditional collaborative filtering recommendation algorithm.