Using Predicate-Argument Structures for Context-Dependent Opinion Retrieval

摘要: Current opinion retrieval techniques do not provide context-dependent relevant results. They use frequency of words in documents or at proximity to query words, such that opinionated containing the are retrieved regardless their contextual semantic relevance topic. Thus, for qualitative analysis products, performance measurement companies, and public reactions political decisions can be largely biased. We propose a sentence-level linear model is based on subjective similarities between predicate-argument structures. This ensures only but semantically The performs combination popular model, our proposed transformed terms similarity subjectivity mechanism. Evaluation experimental results show structures improves task by more than 15% over TREC baselines.