This paper presents an approach for the Opinion Finding task at TREC 2008 Blog Track. For the Ad-hoc Retrieval subtask, we adopt language model to retrieve relevant documents. For the Opinion Retrieval subtask, we propose a hybrid model of lexicon-based approach and machine learning approach for estimating and ranking the opinionated documents. For the Polarized Opinion Retrieval subtask, we employ machine learning for predicting the polarity and linear combination technique for ranking polar documents. The hybrid model which utilize both lexicon-based approach and machine learning approach to predict and rank opinionated documents are the focuses of our participation this year. Regarding the hybrid method for opinion retrieval subtask, our submitted runs yield 15% improvement over baseline.
|Title of host publication||NIST Special Publication|
|Publication status||Published - 2008|
|Event||17th Text REtrieval Conference, TREC 2008 - Gaithersburg, MD, United States|
Duration: 2008 Nov 18 → 2008 Nov 21
|Other||17th Text REtrieval Conference, TREC 2008|
|Period||08/11/18 → 08/11/21|
ASJC Scopus subject areas