In TREC 2007 Blog Track, we developed a three-step algorithm for the opinion retrieval task. An information retrieval step retrieves the query-relevant documents. A following opinion identification step identifies the opinionative texts in these documents. A ranking step identifies the query-related opinions in the documents and ranks them by calculating their opinion similarity scores. For the polarity task, our strategy is to find the positive and negative documents respectively, and then find the mixed opinionative documents in the intersection of the positive and negative document sets. We implemented our opinion retrieval algorithm in two special cases, one to retrieve the positive documents, and the other to retrieve the negative documents. A judging function labeled a subset of the documents, which were in the intersection of the positive and negative documents, as the mixed opinionative documents. We studied two parameters in our opinion retrieval algorithm, each of which had two values to compare. This resulted in four submitted opinion retrieval runs and their corresponding polarity runs.
[1]
Wei Zhang,et al.
UIC at TREC 2006 Blog Track
,
2006,
TREC.
[2]
Thorsten Joachims,et al.
Text Categorization with Support Vector Machines: Learning with Many Relevant Features
,
1998,
ECML.
[3]
Craig MacDonald,et al.
Overview of the TREC 2006 Blog Track
,
2006,
TREC.
[4]
Wei Zhang,et al.
Opinion retrieval from blogs
,
2007,
CIKM '07.
[6]
Wei Zhang,et al.
Recognition and classification of noun phrases in queries for effective retrieval
,
2007,
CIKM '07.
[7]
Clement T. Yu,et al.
An effective approach to document retrieval via utilizing WordNet and recognizing phrases
,
2004,
SIGIR '04.
[8]
H. Chernoff,et al.
The Use of Maximum Likelihood Estimates in {\chi^2} Tests for Goodness of Fit
,
1954
.