Unsupervised Negation Focus Identification with Word-Topic Graph Model

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 1632–1636, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. Unsupervised Negation Focus Identification with Word-Topic Graph Model Bowei Zou Qiaoming Zhu Guodong Zhou* Natural Language Processing Lab, School of Computer Science and Technology Soochow University, Suzhou, 215006, China zoubowei@gmail.com, {qmzhu, gdzhou}@suda.edu.cn Abstract Due to the commonality in natural language, negation focus plays a critical role in deep understanding of context. However, existing studies for negation focus identification ma- jor on supervised learning which is time- consuming and expensive due to manual preparation of annotated corpus. To address this problem, we propose an unsupervised word-topic graph model to represent and measure the focus candidates from both lexi- cal and topic perspectives. Moreover, we propose a document-sensitive biased Pag- eRank algorithm to optimize the ranking scores of focus candidates. Evaluation on the *SEM 2012 shared task corpus shows that our proposed method outperforms the state of the art on negation focus identification. * 1 Introduction Negation is used to reverse the polarity of part of statements that are otherwise affirmative by de- fault (Blanco and Moldovan, 2011), which is common in natural language. Negation focus is defined as the special part in sentence, which is most prominently or explicitly negated by a neg- ative expression. For example, sentence (1) could be interpreted as He stopped, but not until he got to Jackson Hole with a positive part he stopped and a negative part until he got to Jackson Hole. (1) He didn't stop until he got to Jackson Hole. Our previous work (Zou et al., 2014) showed that contextual information plays a critical role on negation focus identification. For better illus- tration of this conclusion, they manually analyze the evidences for 100 negation focuses. It is sur- * Corresponding author prising that 77 focuses can be identified with help of contextual information. This indicates that negation focus is often related with what authors repeatedly states in context. In this paper, we thus focus on graph-based ranking methods (Mihalcea and Tarau, 2004) which first build a word graph according to word co-occurrences within document, and then use random walk al- gorithms (e.g., PageRank) to measure word im- portance. However, for negation focus identification, the graph-based methods may suffer from the fol- lowing two problems: (a) the words in graph- based methods are strongly connected by co- occurrence rather than semantic content, which do not necessarily guarantee that they are rele- vant to the negation focus in context; and (b) identifying a negation focus may be affected by not only the relatedness of surrounding words but also its importance in current document which is not considered in standard random walk algorithms. To address the above problems, we propose a word-topic graph model by adding a topical layer on the original word layer to capture the seman- tic clues from both lexical and topic perspectives. Besides, a document-sensitive PageRank algo- rithm is also proposed to optimize the graph model by …

Content truncated to 500 words