Recommend system is a powerful business tool to enhance e-commerce sales for it has the potentiality to turn browsers into buyers by helping them finding the items they may wish to purchase. Recommender systems also increase cross-sell and enhance customers' loyalty. Due to the enormous good of recommender systems, a lot of e-commerce businesses began to utilize this technology to assist their customers buying process. In our research, firstly, we analyze the influence factors in customer buying processing. We use collaborative filtering and agent technology to anticipate the social influence and psychology tendency. Particularly, multi-agent system has been used in analyzing personal factors hi customer buying.Also in the study of recommender system, we noticed mass marketing has been replaced by data-driven one to one marketing. One-to-one strategy values customer individual preferences and it works on the base of tailoring marketing to each customer. On the other hand, Pareto Principle indicates the major part of revenue for a company comes from a small fraction of customers. That is what so called 80/20 Rule. From the study of Reicheld and Sassur, we even know for some company 100% revenue comes from 5% important customers. It is not exactly new that customers varying in profitability. However, segmenting customers in terms of their profit contribution remains an underutilized approach in many recommender systems nowadays. In such systems, they are likely to believe every browser might be a buyer later, so they give the same care to all of the customers without calculating whether the expected profit from a customer is greater than the cost of marketing to her or not. Providing a high quality recommendation service to all customers is totally not economically logical. So in this thesis, our research is based on the understanding of customer profitability, Pareto's principle and the customer pyramid. A segmenting method based on customer value calculation has been proposed here.This paper has addressed two main issues based on a serial of experiments. First, segmenting customers according to their contributions. Second, proposing a hybrid recommendation strategy which used multi-agent system, collaborative filtering, and Top-N together to generate right recommendations for customers in different profitability tiers.In the first part, we have defined customer value from two categories: intrinsic value and network value. Based on customer's historical behavior, segment them with considering their recency, frequency, and monetary. Worth to mention, customer's network value has first been used to do segmentation in the research.In the second part, we have proposed a hybrid recommendation strategy. Compared with traditional ones, it has combined multi-agent system, collaborative filtering, and a simplified Top-N algorithm together. Personalized recommendation service has been generated to high profitability customers, while high-efficient recommendation algorithm takes care of the others. Experiments have been performed to investigate the performance of our system. Also, agent-based recommendation strategy selection architecture has been proposed and experimented in our research. |