@inproceedings{TD:100429,
	address={Beijing, China},
	att_abstract={There is enough evidence that social media contains timely information that businesses could use to their benefits.
In this paper we discuss automatic extraction of descriptions of problems from twitter data. More specifically we present a system that filters tweets related to an enterprise and extracts descriptions of problems with their product/service. First step of this extraction process is to identify the tweets containing such descriptions. We view this as text classification problem. We propose that sentences describing problems can be characterized by their lexical and syntactic structure. Our experiments show that use of such structural features in classification models, results in the F-measure of 0.742. It is a significant improvement over a baseline F-measure of 0.66, obtained by using only word ngram features. Since twitter data is dynamic, classification models have to be adopted to changing nature of problems and language distributions. We describe a simple adaptation scheme, and experimentally demonstrate its effectiveness. Finally we discuss our method to pinpoint the phrases describing the problems in the identified tweets. We show that by using simple syntactic pattern an extraction F-measure of 0.434 is achieved. Considering the noise in the tweeter data this level of performance is quite encouraging.},
	att_authors={ng2836},
	att_categories={C_IIS.3, C_IIS.11},
	att_copyright={},
	att_copyright_notice={},
	att_donotupload={},
	att_private={false},
	att_projects={OpinionMining},
	att_tags={Twitter, NLP, Data Minining, Text Classification},
	att_techdoc={true},
	att_techdoc_key={TD:100429},
	att_url={http://web1.research.att.com:81/techdocs_downloads/TD:100429_DS1_2011-06-13T16:12:16.105Z.pdf},
	author={Narendra Gupta},
	booktitle={Proceedings of the 3rd Workshop on Social Web Search and Mining  (SWSM2011) in conjunction with SIGIR 2011 },
	institution={{The 3rd Workshop on Social Web Search and Mining (SWSM2011) in conjunction with SIGIR 2011, July 24&}},
	month={July},
	title={{Extracting descriptions of problems with product and service from twitter data}},
	year=2011,
}