Automatic construction of domain-specific sentiment lexicon based on constrained label propagation

Domain-specific sentiment lexicon has played an important role in most practical opinion mining systems. Due to the ubiquitous domain diversity and absence of domain-specific prior knowledge, automatic construction of domain-specific sentiment lexicon has become a challenging research topic in recent years. This paper proposes a novel automatic construction strategy of domain-specific sentiment lexicon based on constrained label propagation. The candidate sentiment terms are extracted by leveraging the chunk dependency information and prior generic lexicon. The pairwise contextual and morphological constraints are defined and extracted between sentiment terms from the domain corpus, and are exploited as prior knowledge to improve the sentiment lexicon construction. The constraint propagation is applied to spread the effect of local constraints throughout the entire collection of candidate sentiment terms. The final propagated constraints are incorporated into the label propagation for the domain-specific sentiment lexicon construction. Experimental results on real-life datasets demonstrate that our approach to constrained label propagation could dramatically improve the performance of automatic construction of domain-specific sentiment lexicon.