Research Of Recommendation System For The Internet Insurance Based On Hybrid Model

In recent years, the internet finance is booming and rapidly infiltrating into all kinds of traditional financial field. The internet insurance adapted to the trend of economic boom in the network age, and keenly captures the risk guarantee requirements of internet economy. It is springing up! There are many different kinds of the existing internet insurance products, however, the function of the existing internet insurance platform is simple. And ordinary users are relatively lack of knowledge about the insurance. It is difficult to find the best products they want.As one of the effective methods to solve the overloading problem of information, the recommendation system goes into our eyes. The recommendation system attracts the extensive attention for it can provide information on potential demands to users. In order to let each user access to specific recommendation lists, the personalized recommendation system arises at a historic moment.In this paper, the main research works is to proposed a hybrid recommendation model for the internet insurance. The subalgorithms which involved in the hybrid model were improved as follows.(1) In user-based collaborative filtering algorithm, in order to get the directed influence relationships between users, we built a consumer network based on time sequence. Then we can find the neighbor set more accurately. We call it TsUCF. It improves the accuracy of UCF.(2) In item-based collaborative filtering algorithm, considering the effects of product type on item similarity, we built the TyICF. It improves the accuracy of ICF.(3) We improved the network inference algorithm NBI, considering the effects of purchase number on resource allocation. The WNBI was built. It improves the accuracy of NBI. After the improvement, their advantages and disadvantages were analyzed by off-line experiment. In the end, we built the hybrid model by combining the TsUCF with the WNBI to improve the quality of the recommendation system. Then we proved that our hybrid model was better in accuracy, diversity and expansibility than traditional algorithms.