Research Of Collaborative Recommendation Based On Formal Concept Analysis

With the rapid spread of the Internet and the development of E-Commerce, E-Commerce systems have made more convenient for users. At the same time, its structure becomes more and more complex. It is difficult for users to find the products and services wanted. Recommendation system can directly interact with users; it can recommend for users to simulate salesman and help users get the goods wanted. In the increasingly fierce competition, recommendation system can effectively keep users and improve the sales. With a good prospect of development and utilization, recommendation system has become an important research field.Recently, recommendation system has been very successful in both theory and practice. However, with the increase of the system, it also faces challenges. Aimed at the main problems of recommendation systems, this paper applies Formal Concept Analysis to improve the efficiency of recommendation. However, extract formal concept from formal context and establish concept lattice are time-consuming. Therefore, stick to the point, this paper carries through a useful explore and research, the two main points of the research as follows:1). In order to better apply concept lattice to recommendation system, this paper proposes a lower time complexity and simple method for constructing formal concept lattice, which based on matrix operation. On the basis of matrix operation, a new algorithm BCLMO is proposed for constructing formal concept lattice. Comparing with some traditional method, BCLMO could distinctly improve the speed of extracting formal concept, and decrease the time and space complexity.2). Aiming at recommendation system, this paper proposes two new notions-core-concept and core-concept lattice, and applies core-concept lattice to recommendation system. Based on the core-concept, the collaborative recommendation algorithm not only avoids the complexity of constructing formal concept lattice, but also filters the users who can't be "the latest neighbors". It decreases the computational scale, reduces the time and space complexity, and gets rid of the noise to a certain degree. Afterward, experiments prove this method can improve the recommendation efficiency and quality.