Listen to me, my neighbors or my friend? Role of complementary modalities for predicting business popularity in location based social networks

Businesses are one of the major stakeholders of the location-based social network (LBSN) platform. Predicting popularity of businesses in LBSN can help in future business planning and design of effective marketing strategies for business owners. In this paper, we introduce four simple modalities that serve as prime indicators of business popularity, namely, (a) social influence, (b) geographical proximity, (c) customer preference and (d) textual content of tips and reviews posted for a business and investigate their role on business popularity. Characterizing business popularity is ambiguous as there exist different viewpoints to measure popularity. We propose a principled methodology to properly label the popular and unpopular businesses by systematically defining popularity metrics specific to business categories. Social influence essentially represents the peer pressure for visiting certain businesses and eventually contributing to its popularity. Geographical proximity reveals that there are two types of customers visiting a business, namely, local & foreign customers and subsequently, we define the local dominated as well as foreign dominated businesses. On the other hand, customer preference reflects the inclination of a customer towards certain types (say ‘Shopping’, ‘Hotels’, ‘Religious Organizations’, ‘Mass Media’ etc.) of businesses. Further, textual content reveals information about customers’ experiences in visiting corresponding businesses. Interestingly, our investigation reveals that customers visit popular businesses amidst social pressure, however they visit unpopular businesses driven by their own personal preference. Moreover, we observe that local (foreign) customers play a major role in shaping the popularity of businesses in local (foreign, respectively) dominated cities. In addition, we observe that textual content of tips and reviews play a significant role in attracting customers to a business adding to its popularity. In a nutshell, this paper puts forward an important message that social influence, geographical proximity, customer preference and textual content exhibit strong complementary signals to predict business popularity. We evaluate the predictive power of the proposed modalities in estimating business popularity through the Naive Model using state-of-the-art machine learning algorithms that shows reasonable improvement over baseline algorithms. We also propose DeepPop, a novel deep learning model to predict business popularity by leveraging on proposed modalities. Our evaluation shows that DeepPop outperforms several state-of-the-art baseline algorithms achieving significant improvement in performance for predicting business popularity.