Using Multiple Resources in Graph-Based Semi-supervised Sentiment Classification

Abstract

For sentiment classification, there exist a heterogeneous mass of resources such as semantic dictionaries, unlabeled corpora, and heuristic rules. In this paper, based on a graph-based semi-supervised algorithm, we focus on exploiting multiple resources to construct similarity matrices which are fused by simple but effective schemes. We reported encouraging results of the experiments in sentiment classification, which indicate that the adopted algorithm can utilize multiple resources to improve performance.