Proximity-based opinion retrieval

摘要: Blog post opinion retrieval aims at finding blog posts that are relevant and opinionated about a user's query. In this paper we propose simple probabilistic model for assigning scores to documents. The key problem is how capture expressions in the document, related query topic. Current solutions enrich general lexicons by query-specific using pseudo-relevance feedback on external corpora or collection itself. use lexicon proximity information order term relatedness We proximity-based propagation method calculate density each point document. position of document can then be considered as probability position. effect different kernels capturing also discussed. Experimental results BLOG06 dataset show proposed provides significant improvement over standard TREC baselines achieves 2.5% increase MAP best performing run 2008 track.