Measurement Theory-Based Trust Management Framework for Online Social Communities

Abstract

We propose a trust management framework based on measurement theory to infer indirect trust in online social communities using trust’s transitivity property. Inspired by the similarities between human trust and measurement, we propose a new trust metric, composed of impression and confidence, which captures both trust level and its certainty. Furthermore, based on error propagation theory, we propose a method to compute indirect confidence according to different trust transitivity and aggregation operators. We perform experiments on two real data sets, Epinions.com and Twitter, to validate our framework. Also, we show that inferring indirect trust can connect more pairs of users.

Using Twitter trust network for stock market analysis

“…Therefore, unlike many existing works which treat all the authors equally important or ignore authors' identities, in addition to analyzing tweets' sentiment, we also take into account tweets' authors. We adapted our trust management framework [28] and constructed a user-to-user trust network for Twitter users based on their tweeting behaviors. Then, users were differentiated by their reputation or power in the whole community, where reputation or power is determined by the user-to-user trust network.…”

“…It did not differentiate their authors. In the remainder of this work, given historical Twitter data, we adapted our trust management framework [28] to measure user-to-user trust relationships and then construct a trust network for the whole community. Based on the trust network within the community, we then derived users' reputation or power, which is used later as weights in the process of Twitter sentiment aggregation.…”

“…In this paper, we call this trust inference. In [28], we proposed a measurement theory based trust management framework which addresses both trust modeling and trust inference. We represent our trust management framework in Figure 1.…”

“…As in [52], we select SentiStrength [32] as our sentiment analysis tool. To test the accuracy of SentiStrength, we used a Yelp dataset which includes both text reviews and ratings in [28]. We used the scalar outputs (from −4 to +4)…”

“…propagative. There are two types of inference operations used to capture trust propagation: trust transitivity and trust aggregation [28].…”