Design And Implementation Of Personalized Recommender System For Smart Trading Area

With the rapid development of mobile Internet and Internet of Thingstechnology, traditional trading areas are in the process of upgrading andtransformation to Smart Trading Areas. Based on the Online to Offlinebusiness model, Smart Trading Area makes an organic combination ofonline and offline services, and promotes prosperity by improving overalllevel of services. The key point of its business model is to transfer onlinepotential consumers to offline consumptions. Therefore, one of the mostimportant problems in building Smart Trading Area is that how to providerelevant information to consumers according to their interests.To address this problem, in this paper the design decisions of apersonalized recommender system is made by analyzing the Smart TradingArea scenario, and based on the design decisions, the system is furtherimplemented to provide recommendations. The main contributions of thispaper include:1. The recommendation environment of this system is studied anddetailed analysis is performed on external application, user and data in thescenario to recognize main constraints and make decisions on the choiceabout the recommender system architecture and the choice ofrecommendation algorithms.2. The principal and application of recommendation algorithms inSmart Trading Area scenario is analyzed. The importance of tag quality intag-based algorithm is presented and the tag quality is improved by cleaninglow quality tags and recommending high quality tags. In respect of the datasparsity problem in collaborative filtering algorithms, the concept of userfavorite item set and a set similarity algorithm is proposed. Based on that, auser favorite item set based collaborative filtering algorithm is proposed toaddress the data sparsity problem.3. In respect of the usage of context information in recommendationalgorithms, the context filtering approaches are introduced and the importance of location context and temporal context are discussed.Furthermore, a Geohash based solution is proposed to provide locationbased nearby recommendations.4. Based on the Smart Trading Area scenario, the architecture of thepersonalized recommender system is proposed which divides the process ofrecommendation computation into online parts and offline parts, thusreducing online computing time cost and ensuring real-timerecommendation to some extent. Finally, detailed design andimplementation of the recommender system is described, this systemsupports user data collection, tag recommendation, popular recommendation,nearby recommendation, personalized merchant and item recommendationand recommendation explanation. The system functionality is verifiedthrough application examples.The system implementation and verification shows that thepersonalized recommender system designed and implemented in this paperprovides a feasible and effective solution to support personalizedinformation services in Smart Trading Area scenario and hastheoretical and practical values.