The business model of the smart trading area is turning the potential demand of online consumer to offline goods. And this eventually contributes to real consumption. Recommender system can recognize user’s requirements and give recommendations. Result of the recommended sequence affects the user experience. In smart trading area, such a complex environment, we need to provide different scenarios with the corresponding sorted recommended results.In order to solve the above problems, we need to design and implement an adaptive framework for smart trading area. This framework can be trained for different scenarios. After training, it is suitable for different needs, and can provide a corresponding sorted recommended list of results.In this paper, the main work is following.Firstly, we use learning to rank technology into recommender system. We propose a boosting merging algorithm. This algorithm uses the result of each recommendation algorithm as a weak ranker. Then merge them by optimizing the NDCG evaluation index with boosting method.Secondly, we propose an updating algorithm bases on user feedback information. This algorithm uses the output of boosting merging algorithm as a base model. And its core is LambdaMART algorithm. We divide this updating algorithm into three parts: mass updating, small batch updating and real time updating. By doing this, the algorithm meets the real-time requirements.Thirdly, we design and implement an adaptive framework. After training with data from different scenarios, this framework is suitable for different needs. We divide this framework into two parts: online and offline, to meet the design requirements of real-time. And this framework can collect and store the user feedback information for the updating algorithm.Fourthly, we experiment our algorithm. We implement the adaptive framework using our algorithm. And we test and analysis our algorithm. |