Research On Self-adaptive Recommendation Algorithm Based On Empirical Learning

Traditional recommendation algorithms generally use users'browsing history, friends recommend or some other ways to recommend. Its main purpose is to solve the information overload problem in the field of electronic commerce. However, in practical application, it is generally presented as auxiliary function, and the main search algorithm is probably the web pages content-based sorting index algorithm. While in the content-based search, there are few personalized recommendation schemes, and the search quality has many problems. How to combine the above two points is the key to design a new high quality search algorithm.Considering the fact that the content-based search ranking can represent the correlation between the query and documents in a certain extent, and in recommendation theory, users' historical behaviors, such as user clicks, plays the guide recommended role on later users. This paper well combined the above two advantages, and proposes a recommendation algorithm named IMUSE. The main contributions of this paper are as follows. Firstly, It proposes a method to handle the historical user click data, which gives the processing means of the user clicks on the timeliness and documentation as well as heat. Secondly, It presents an experience based search recommendation algorithm, which combines the advantages of content-based retrieval recommendation algorithm and experience based recommendation algorithm. Especially, it gives a very good solution to the balance of query inputs and some other issues. Lastly, It puts forward a self-adaptive ACO based on PSO (Particle swarm Optimization), termed SAACO, which makes the parameters of ant colony algorithm in PSO dynamic, at the same time, applies the algorithm to the parameter estimation in search recommendation algorithm, in order to make the proposed algorithm adaptive parameter setting.This paper applies SAACO estimate the parameter of IMUSE, and compares them with the experimental results, the results show that those two values are basically similar. In addition, this paper selects accuracy and nDCG as two authoritative evaluation basis to evaluate the performance of IMUSE, and verifies the algorithm in the superiority of the performance through experiments.