Extracting Social Dimensions Using Fiedler Embedding

Abstract

In this paper, we present and evaluate the use of a Fiedler embedding representation for multi-label classification of social media. Networked data, such as data from social media, contains instances of multiple types that are related through different types of links. The network structure causes these data instances to no longer remain independently identically distributed (i.i.d.). Relational learning succeeds in improving the classification performance by leveraging the correlation of the labels between linked instances. However, instances in a network can be linked for different causal reasons, hence treating all links in a homogeneous way limits the performance of relational classifiers on such datasets. Social-dimension based approaches address this problem by extracting a feature space which captures the pattern of prominent interactions in the network. In this paper, we propose an alternate low-dimensional social feature representation that can be extracted from edge-based social dimensions using Fiedler embedding. This embedded feature space encodes the relations between people and their connections (nodes and links). Experiments on two real-world social media datasets demonstrate that our proposed framework offers a better feature representation for multi-label classification problems on social media.