Hybrid Recommendation System Research Based On Cauchy Quantum-behaved Particle Swarm Optimization

The rapid developments of various Internet services have made the web structure more and more complicate. Although the huge amount of information provides convenience and entirety to the user, it also lowers the utilization of the information, a common situation is that users often lose the point in the enormous messages which causes a typical problem of information overwhelming and overloading. Recommendation system helps to resolve this phenomenon effectively. However major problems such as data sparsity, cold start and change of users' interests limit the system to further development. The aim of this article is to solve these problems.A hybrid recommendation algorithm based on Cauchy Quantum-behaved Particle Swarm Optimization is proposed. The algorithm firstly constructs the hybrid recommendation model based on time factor, and then, on the basis of the model, the Cauchy Quantum-behaved Particle Swarm Optimization algorithm is applied for searching the optimal parameters of the model. Firstly, the hybrid recommender model is built by adding the features of the users and items to the traditional collaborative filtering algorithm, and the two main contributions are as follows: (1) when comparing the similarity of the items or users,if the user rating matrix is too sparse to calculate, we can use the content information, thus it can effectively overcome the defect; (2) when adding a new user or item, we can calculate the similarity with content information of them and perform recommendation, which can decrease the effect of cold start. In addition, the other improvements in building the hybrid recommendation model are as follows: one is to improve the way of calculating the similarity put forward; the other is to introduce a time factor represented the change of user's interest.Furthermore, in previous models, user profiles usually contain gender, age, profession, living place and many other attributes without considering whether the information has actual use to the specific recommendation or not. In fact, taking all attributes into consideration may not necessarily obtain the most accurate result, and on the contrary, it costs more redundant efforts. Thus, this paper optimizes the hybrid recommendation model based on time factor ,and searches the best user's attributes by using improved discrete Particle Swarm Optimization. Finally, the optimized model involves in 5 parameters ,such as the weight of user's ratings and contents, the weight of item's ratings and contents, the weight of user-based and item-based recommendation,the threshold of user's nearest neighbor and that of item's nearest neighbor. It is a huge job to do it by hand. Therefore Cauchy Quantum-behaved Particle Swarm Optimization algorithm is adopted to search the optimal parameters so that it can improve recommendation effects.The algorithm proposed in this paper is compared with the conventional collaborative filtering algorithm and the algorithm using Particle Swarm Optimization to search parameters and that using Artificial Bee Colony to do it. Experimental results show that in increasing recommendation accuracy, alleviating the data sparsity and cold start effect aspects, the proposed algorithm is superior to other algorithms.