O2O Recommender System Based On Online Shopping Behavior

In recent years, traditional retail market is facing the grate impact from e-commerce provider, and is in the transition to intelligent shopping area. In order to improve the user shopping experience in the area, and help user to quickly find points of interest in the area, the recommender system can provide effective help. While existing recommender systems in e-commerce has already developed very mature and has been widely used, there is no matured application of recommender system in offline retail industry yet. Existing offline recommender system mainly consists of sorting by popularity and location-based services. The former can only be applied to restaurant industry, because it has on ability to support personalized recommendations. Recently, the latter is hot in academia, but since it is heavily dependent on location data quality, therefore it is difficult to promote in the real scene.Given the currently strong demand for personalized recommendation system in the retail industry, and existing shortage of offline recommendations, in this paper we changed the recommendation ideas and then proposed recommender system combining O2 O mode. O2 O recommendation system analyze the user's consumer preferences according to the user online shopping behavior, and based on the preference, provide offline recommendations to users. The greatest advantage of O2 O recommender system is that, in comparison to offline location data, online data has great advantage in the quantity and quality of all aspects of the data, therefore O2 O recommender system can analyze user preferences more accurately. At the same time, O2 O recommendation system will face a series of challenges, because the recommended process has gone through the process from online to offline.In this paper, we first described the status of the retail industry, illustrates the research background. Then technical roadmap of recommender system was reviewed, and the research work of current location-based recommender system was introduced. Based on existing research, research ideas of this paper were proposed. Subsequently, this paper put forward the concept of O2 O recommender system, analyzed the advantages of O2 O recommender system and also listed the challenges the online shopping behavior based O2 O recommender system faces. In order to overcome these challenges, this paper presents a series of models and designs O2 O recommender system used, including time-recession behavior scoring model, apparent semantic user preference model based on interest terms, user- label matrix filling technique, offline shop recommendation model based on similarity, the recommendation result diverse processing technologies as well as principle of online and offline calculation separation. Then, this paper described in detail the process of converting these design to software implementation, proposed modularized O2 O recommender system framework, the principle of the separation of different real-time computing levels and the internal data management system that meet the demand of persistent and real-time in the same time. Finally, a series of experiments validated the correctness of model we proposed and the the real-time performance of O2 O recommender system.