| The Internet has brought great convenience to people’s daily life, more and more consumers tend to purchase merchandise online. There are a series of advantages in shopping online. At first, consumers have more widely choices. Secondly, shopping online is a simple process, which could be easily operated. The last and the most important, the time cost in shopping online is lower than shopping than in physical store. It has saved the time that consumers cost on the way. However, there exist some defects on traditional e-commerce platform. As so many goods are exhibited, it is not easy for consumers to find what they want and what they need, this is the problem of information overloading.For the purpose of recommending products to customers precisely and improving the sales of products, the recommendation system appeared. However, the historical data of user behavior, such as the time that user stay at the product pages, browse times, number of adding to shopping cart,and so on.These all belong to the implicit feedback. Designing effective personalized recommendation algorithms according to implicit feedback of users, is the main content of this paper.In this paper, some classical algorithms of personalized recommendation system are introduced, Such as content-based recommendation, collaborative filtering recommendation and hybrid recommendation, etc..Then in the field of electronic commerce, We take user behavior data that including time information in consideration, mainly "click, collection, add to cart, buy" four kinds of behaviors. A series of features are extracted from user behavior sequence, which are the input parameters of logistic regression model. Then we obtain a probability matrix. We view the obtained probability matrix as the scores, and use collaborative filtering recommendation strategy to recommend products to customers. The traditional collaborative filtering methods tend to ignore the impact of consumption time. Comparatively, this paper pays attention to the temporal behavior, which makes the personalized recommendation more reasonable.The experiment make use of the data that consumers left when they shopped online in some year. Our experimental results show that behavior sequence combined with collaborative filtering recommendation strategy has the ideal effect in recommendation. Besides, it has solved the problem that the strategy of collaborative filtering couldn’t take advantage of implicit feedback directly. What’s more, our algorithm perform well with sparse data. At last, beginning from the busness features and the angel of statistics, this paper take some measures to adjust algorithm. Therefore, the result of the recommendation are optimized and the accuracy of the algorithm is improved. |
