The Research And Implementation Of Long Tail Recommendation Based On Electronic Commerce

With the development of internet technology, electronic commerce platforms have increasingly become the main source of network traffic. At the same time, online consumption and offline services have turned out to be the mainstream of public life. Among all the technology applied in this trend, the recommendation systems from the electronic commerce platforms are the main connection between customers and business merchants. This kind of technology was applied as a result of combining online data collection, offline data analysis,online data recommendation and feedback closed-loop conduction mode to guide users' behavior, dig their interests, and improve profits efficiently at last. The current recommendation systems have been widely used in the major electronic commerce platforms, within which the applied technology is becoming more and more mature. However, most recommendation systems are influenced by the long tail effect more or less, which is because they are all based on recommendations oriented by collaboratively filtering users and content.This dissertation discussed the existing problems of recommending long tail product from common electronic commerce platforms through specific experiments. By given corresponding theme models of utilizing long tail data, the performance evaluation criteria of long tail recommendation was offered.After analyzing the characteristics of the data sets, the first step was to create a cascade model to recommend products according to the commonly applied classification algorithms. The results of the experiments showed that on the condition of insuring the recommendation effect, F1 value was stable at 34% approximately. With the data collected, this dissertation pointed out the insufficient utilization of those algorithms in analyzing long tail product, and offered a new standard for evaluating long tail recommendation rate and stability, by which in turn to assess the utilization rate and comprehensive performance of existing algorithms. In the cascade model, the index values for the recommendation rate and comprehensive performance were 29.5% and 1.105 respectively. Based on the theory of using theme models to broaden the source of information channels, a new EN-LDA algorithm according to the users' algorithm was proposed. The aim of this kind of algorithm was to solve the common problems in recommending product.The final experiment result showed that the new theme model increased the long tail recommendation rate and stability rate to 47.6% and 1.452 respectively, and long tail-rank was improved to a higher level, while ensuring traditional recommendation performance. As could be seen from the entire experiment, this kind of new algorithm would be able to solve the utilization problems in commonly used recommendation systems.