Functional-Oriented Relationship Strength Estimation: From Online Events to Offline Interactions

Abstract

Link mining/analysis over network has received widespread attention from researchers. Recently, there has been growing interest in measuring relationship strength between entities based on attribute similarity. However, limited work has assessed the competitive advantage of functional elements in relationship strength quantification. The functional elements embody the growth/development nature of the relationship. Motivated by the availability of large volumes of online event records that can potentially reveal underlying functional socio-economic characteristics, we study the problem of offline relationship strength estimation with functional elements awareness from online events. Two major challenges are identified as follows: (1) informal information, online events are of high dimensions, and not all the learnt functions of online events are predictive to offline interactions; (2) heterogeneous dependency, it’s hard to measure the relationship strength by modeling functional elements with network effects jointly. To handle these challenges, we propose generalized relationship strength estimation model (gStrength), a novel approach for relationship strength estimation. First, we define the combination of latent roles and observed groups as generalized roles, and present generalized role constrained latent topic model to make the extracted latent functions compatible with offline interactions. Second, we model the functional elements and further extend them to structural dependency settings to quantify relationship strength. We apply this approach to the political and economic application scenario of measuring international investment relations. The experimental results demonstrate the effectiveness of the proposed method.