Opinion Summarization using Entity Features and Probabilistic Sentence Coherence Optimization: UIUC at TAC 2008 Opinion Summarization Pilot

摘要

This paper talks about participation of the University of Illinois at Urbana-Champaign (UIUC) in TAC 2008 Opinion Summarization pilot. We mainly explored two ideas: (1) use of entity recognition and parsing to enhance a standard retrieval method for opinion retrieval, and (2) use of a coherence language model to optimize the ordering of sentences in a summary. Our result showed that use of entity recognition during retrieval led to mixed results and re-ordering with coherence language model was not as good as heuristic polarity-based ordering using guiding phrases. Our additional experiments showed that the performance of coherence language model can be different depending on probability function and word selection.