Study On Collaborative Filtering Technique And Application In Personalized Recommendation System For E-Commerce

Wide use of Internet and speedy development of E-commerce caused information overload, which made difficulties for consumers to find their needed products within a mass of product information, thus E-commerce recommendation systems emerge as the times require. But current E-commerce recommendation systems are considerably immature in practical use, and still have a lot of problems, like the quality of recommendation being seriously depressed by sparse consumers'ratings, bad system expansibility, being short of recommendation diversity to cover consumers'whole preference, etc. To solve these main problems of current recommendation systems, this dissertation valuably explores and researches the key techniques of recommendation strategies and algorithms in E-commerce personalized recommendation systems.In the aspect of recommendation strategies researching, the dissertation analyses different types of strategies of current E-commerce systems applying in different occasion and different situation. Aiming at the E-commerce system for frequently-buying products, the dissertation analyses its recommendation requirement, and brings forward the detailed recommendation strategies applying on this type of E-commerce systems, which is to generate a main top-N recommendation list to meet the demand of veracity and novelty in recommendation, to generate a secondary top-N recommendation list to meet the demand of integrality and diversity in recommendation, to generate new adding products list to draw more attention on new products, to generate current most popular products top-N list.In the aspect of recommendation algorithms designing, this dissertation researched current mainstream technique in personalized recommendation, collaborative filtering. The dissertation deeply analyzes its sparsity problem which depresses recommendation quality and recommending integrality problem which influences consumers'satisfaction, introduces"concept hierarchy"and"community filtering"techniques to improve collaborative filtering algorithm, and designes improved algorithms to realize the strategies proposed above. To solve the sparsity problem of collaborative filtering, by using"concept hierarchy"thought, new consumer profile is built, and the connotative relation between items is found. It alleviates the sparseness of consumers'rating data, helps users find right neighbors and get better recommendations. To solve the recommending integrality problem of collaborative filtering,"community filtering" technique is introduced, consumer's interest is divided into different"communities"according to items'categories, recommendations based on individual community is given to satisfy consumers'demand of understanding different type of products and embody consumers'whole interests sufficiently.We does imitation experiment on improved algorithms proposed by the dissertation, through experiment's validation, improved collaborative filtering algorithms based on concept hierarchy are better than traditional one in the aspect of veracity, integrality, diversity in recommendation, especially in sparse consumer rating datasets, the improved ones behave favorable recommendation performance.