Design And Implementation Of Client-orieted Recommendation Engine For Baidu Groupon Module

This Project was participated by author during the internship in Baidu Inc. This Personalize Recommendation System is designed to enhance the user experience of users, reduce the cost of the user’s selecting, provide users resources with needing to search, and to increase the user engagement, improve the sales of Nuomi. This system includes the offline calculation module and online recommendation module. The author participated in the design and implementation work of the entire project.In the requirements analysis phase, the author communicated with the relevant personnel, to determine the overall requirements of the project. Then, determined the overall design of the project according to the demand analysis, and divided the function module. The whole system has been devided into two primary modules and six secondary modules. The two primary modules are offline calculation module and online recommendation module. Offline calculation module is responsible for the offline calculation part. It has been devided into three secondary modules, including topic model offline calculation module, collabrative filter offline calculation module and logit regression offline calculation module. These three secondary modules correspond to three algorithem that has been used in the system. Each of the modules is responsible for the the model traning of the corresponding algorithm model. Online recommendation module is responsible for the online recommendation part. It also has been divided into three secondary modules. The topic model online recommendation module and collaborative filtering online recommendation module is to generate recommendations for users using model calculated offline. Recommendation reprocessing module is responsible for a series of reprocessing work of the recommendations, including resort the recommendations using the logit regression model trained offline. After the completion of the development, the author has evaluated the effect of each algorithm in the system, and has tested the whole system online.The project in this paper has performed well in the test. Recommendations generated by the system have a higher click rate than the random content. This indicates that this personalized recommendation system can generate recommendations that users would like using history information of users. The system has met the expecatation.