Identifying high value customers in a social network: individual characteristics vs. social influence

Firms are interested in identifying customers who generate the highest revenues. Typically, customers are regarded as isolated individuals whose buying behavior depends solely on their own characteristics (e.g., previous purchase behavior, demographics etc.). In a social network setting, however, customer interactions can play an important role in purchase behavior.

This thesis develops a generalizable methodology to identify high-value customers in a network. Previous work on social networks has focused most attention on modeling the interaction between individuals and understanding the positions of individuals in a network (e.g., measuring the influence of an individual based on his/her degree of network centrality). Little is known about how network influence directly translates into the benefits to the firm. In this study, the importance of taking into account both an individual characteristics and network effects when measuring customer value is argued. Drawing upon the spatial statistics literature, a spatial autocorrelation model is constructed that explicitly shows how these effects interact in generating firm revenue.

This model is applied to a unique user-level dataset from a popular online gaming company in Korea. The data contain information about demographics of individual gamer, interaction between gamers, behavior within the game environment, and revenues generated by each individual. First, we propose a static model studying gamers' revenue in one period. We quantify the relative impact of an individual characteristics and network effects on revenue. The proposed static model shows better forecasts of an individual's value within a network for the firm than the benchmark models. The empirical analysis shows that individuals who are most influential in a network sense are not necessarily individuals who have the highest customer value.

Next, we incorporate the spatio-temporal aspects of social influence in a network into the static model. This model is extended to construct the spatial dynamic model to forecast revenue in a social network. Second, we account for the homophily effects by separating the contemporaneous network effects out into the contemporaneous, temporal, and spatio-temporal effects. The proposed spatial dynamic model allows us to quantify an individual value in a network in a long-term perspective. The dynamic model is shown to outperform the static, and the other benchmark models in quantifying an individual value in revenue generation to the firm. Lastly, a dynamic coevolution model to account for homophily is suggested and discussed for future research.