Identifying Sources of Opinions with Conditional Random Fields and Extraction Patterns

By Yejin Choi, Claire Cardie, Ellen Riloff, Siddharth Patwardhan. Human Language Technology - The Baltic Perspectiv, 2005

416 citations recorded by the source corpus.

Recent systems have been developed for sentiment classification, opinion recognition, and opinion analysis (e.g., detecting polarity and strength). We pursue another aspect of opinion analysis: identifying the sources of opinions, emotions, and sentiments. We view this problem as an information extraction task and adopt a hybrid approach that combines Conditional Random Fields (Lafferty et al., 2001) and a variation of AutoSlog (Riloff, 1996a). While CRFs model source identification as a sequence tagging task, AutoSlog learns extraction patterns. Our results show that the combination of these two methods performs better than either one alone. The resulting system identifies opinion sources with 79.3% precision and 59.5% recall using a head noun matching measure, and 81.2% precision and 60.6% recall using an overlap measure.

Subfields: Artificial Intelligence

View Paper: https://www.semanticscholar.org/paper/4bb8b75308b5e9021c02285584e1851479089d91

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. B. Liu
2. Bing Liu
3. Yejin Choi
4. Claire Cardie
5. Noah A. Smith
6. Erik Cambria
7. Mohit Bansal
8. Jianfeng Gao

Citations in the corpus

• OpinionFinder: A System for Subjectivity Analysis (2005, background): “Please see (Wiebe and Riloff, 2005; Choi et al., 2005; Wilson et al., 2005) for discussions of related work in automatic opinion and sentiment analysis.”
• Minimum-Risk Training of Approximate CRF-Based NLP Systems (2012, background): “CRF-based models have improved the state of the art in a number of natural language processing (NLP) tasks ranging from part-of-speech tagging to information extraction and sentiment analysis (Lafferty et al., 2001; Peng and Mc-Callum, 2006; Choi et al., 2005).”
• Structure-Aware Review Mining and Summarization (2010, methodology): “For example, with CRFs, Zhao et al (2008) and McDonald et al. (2007) performed sentiment classification in sentence and document level; Breck et al (2007) identified opinion expressions from newswire documents; Choi et al. (2005) determined opi-654 nion holders to opinions also from newswire data.”
• An Introduction to Conditional Random Fields for Relational Learning (2007, background): “…2005], segmenting addresses in Web pages [Culotta et al., 2004], finding semantic roles in text [Roth and Yih, 2005], identifying the sources of opinions [Choi et al., 2005], Chinese word segmentation [Peng et al., 2004], Japanese morphological analysis [Kudo et al., 2004], and many others.”
• Text Mining in Big Data Analytics (2020, methodology): “A source identifier combining a conditional random field sequence tagging model [60] and extraction pattern learning [61] to determine the sources of the speech events and subjective expressions [62];”
• Using Sequence Kernels to identify Opinion Entities in Urdu (2011, methodology): “Related Work Choi et al., (2005) consider opinion entity identification as an information extraction task and the opinion holders are identified using a conditional random field (Lafferty et al., 2001) based se-quence-labeling approach.”
• Structured Models for Fine-to-Coarse Sentiment Analysis (2007, methodology): “Choi et al. (2005, 2006) use CRFs to learn a global sequence model to classify and assign sources to opinions.”
• Phrase Dependency Parsing for Opinion Mining (2009, background): “, 2002; Dave et al., 2003; Kim and Hovy, 2004; Takamura et al., 2005). While we focus on extracting product features, opinion expressions and mining relations in this paper. Kobayashi et al. (2007) presented their work on extracting opinion units including: opinion holder, subject, aspect and evalua”
• Opinion Mining with Deep Recurrent Neural Networks (2014, methodology): “Variants of conditional random field (CRF) approaches have been successfully applied to opinion expression extraction using this token-based view (Choi et al., 2005; Breck et al., 2007): the state-of-the-art approach is the semiCRF, which relaxes the Markovian assumption inherent to CRFs and…”
• Building a Sentiment Summarizer for Local Service Reviews (2008, methodology): “The model we employ for sentiment classification is a hybrid that uses both lexicon-based and machine learning algorithms.”
• Topic sentiment mixture: modeling facets and opinions in weblogs (2007, neutral): “Natural Language Processing (see e.g., [26, 2 ]).”
• FactBank: a corpus annotated with event factuality (2009, methodology): “Hence, a corpus annotated with this kind of data is of great value for informing tools devoted to identifying opinion sources, in line with other work in the field, e.g., Bethard et al. (2004) and Choi et al. (2005) .”
• Temporal Patterns of Happiness and Information in a Global Social Network: Hedonometrics and Twitter (2011, background): “, OpinionFinder [62,63]) but suffer from a degree of inscrutability.”
• Opinion Mining and Sentiment Analysis (2008, methodology): “They share some sub-tasks in common, such as entity recognition; for example, as mentioned above, determination of opinion holders is an active line of research [37, 42, 61, 158].”
• A Survey of Opinion Mining and Sentiment Analysis (2012, background): “Opinion holders are more important in news articles as they often explicitly state the person or organization that holds an opinion [5, 13, 49].”
• Joint sentiment/topic model for sentiment analysis (2009, methodology): “…mixture of topics in the text, which is less informative to users and may limit the usefulness of the results; (2) most of the approaches [16, 15, 4, 3, 13, 25] are favored in supervised learning, which require a labeled corpus for training and potentially restrain their applicability to other…”