A social search model and algorithms based on the location aspect model for mobile-commerce services

Abstract

The location-based social network services (LBSNSs) are getting more popular with a wide spread of smart devices, most of which have a GPS module enabling people to share their places of interests with their friends. In addition, the mobile Q&A services, such as Naver KIN mobile, have been growing as the number of smart device users increases. However, the current mobile Q&A services expose questions to all users regardless of the questions` content or users` interests. In other words, each user sees questions of various categories and has to select the questions to answer. For this reason, askers sometimes wait for a long time to get answers for the questions, and not all the answers have high quality. In this paper, to overcome these limitations, we propose to use social search that finds experts for a given question and pushes the question to them. Our main contribution is to develop a social search model, which we call the location aspect model, tailored to location-based questions in mobile-commerce (M-Commerce). The model calculates the expertise score of each user about a given question. To reflect the characteristics of those questions, we identify the M-Commerce questions` categories and components by analyzing real question sets archived in Jisiklog. Thus, our location aspect model is expected to achieve higher accuracy for M-Commerce questions than other general models. For the validation of our model, we collect the user and check-in information from Foursquare and conduct three types of evaluations. First, for the users judged as experts, we compare their profiles with the questions to confirm that they are eligible to be selected as experts. Second, we create ten synthetic users with particular visit behaviors and check if they are properly selected by our model. Third, by user studies, we conclude that the users who are judged as experts respond sooner with higher-quality answers than those who are not. Overall, we believe our model will be prac...