| Summary: | 碩士 === 國立成功大學 === 製造資訊與系統研究所 === 102 === Recommender System is implementing in so many e-commerce websites like Amazon and Taobao which are the most famous C2C websites all over the world, and the main algorithm they used is Collaborative Filtering (CF), which could realize the recommender system quickly since the parameters are few. However, the disadvantages are still there, like Scalability problem, Data Spartsity problem, and Cold-start problem, which are discussing in most thesis in the past.
To improve the Recommender efficiency, this research implement a C2C shopping website - ”IMI Lab Global” for customers to sell and also purchase products, and there are 3 solutions to solve the problems that CF algorithm have, 1. for scalability problem, this research not only use the RDB but Apache Hadoop environment to deal with NoSQL data to realize the processing speed, high throughput and low latency, 2. for data sparsity problem, this research gives a Rating time-filter solution to give ratings on the user-item matrix which we will use for CF memory-based model, 3. for Cold-start problem, this research uses Facebook access for customers to login into the shopping website ”IMI Lab Global”, and then we may take the user’s preference as our user’s similarity concept.
|
|---|
