Product Recommendation Algorithm Refinement Based On Collaborative Measures Of Product Properties

E-commerce is one of the most important applications in information age. With dramatic increment of product amount, it becomes harder for a user to pick out what he/she want. Product recommendation algorithms draw more and more attention for generating list of products to recommend.In this research, product properties are studied and quantitatively described with statistical summaries of customer behavior patterns, or collaborative descriptions. The benefit is to achieve a formal and quantitative measure of product properties, without sacrificing Operability. A product recommendation algorithm is refined with these measures. Simulation with benchmarking dataset shows that the refined algorithm has better performance.The research includes the following works:1. Propose new research approach, where product properties are quantitatively measured by collaborative descriptions.Product property is critical information for improving product recommendation. But it is demanding to obtain knowledge of product properties directly. On the other hand, customer behaviors are relatively easy to be collected and summarized. Using collaborative information describing product properties provides a way to tackle the problem.2. Define three collaborative product property measurements.The three measurements reveal, respectively, product properties considering time, space and uncertainty. Mathematical definitions are given and semantics of the measurements are discussed.3. Refine a product recommendation algorithm with help of one of the three measurements. And performance improvement is proven by simulation with benchmarking dataset.4. Give a general work flow of applying the research approach to other fields.The work flow consists of three stages, data collection & storage, information analysis and evaluation & result release. Attention should be paid especially to abstraction procedure of customer collaborative behavior pattern from raw data, relations between collaborative behavior and product properties, and the release and sharing of the result.By adopting the collaborative measurements, a product recommendation algorithm is refined to obtain better recommendation result. This benefits all e-commerce participants. For customers, it would be easier to pick out the product he/she want. For on-line shops, it's a channel to better cross selling. For e-Commerce platforms, it leads to better customer service.