Exploiting Domain Knowledge in Aspect Extraction

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1655–1667, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Exploiting Domain Knowledge in Aspect Extraction Zhiyuan Chen, Arjun Mukherjee, Bing Liu Meichun Hsu, Malu Castellanos, Riddhiman Ghosh University of Illinois at Chicago HP Labs Chicago, IL 60607, USA Palo Alto, CA 94304, USA {czyuanacm,arjun4787}@gmail.com, liub@cs.uic.edu {meichun.hsu, malu.castellanos, riddhiman.ghosh}@hp.com Abstract Aspect extraction is one of the key tasks in sentiment analysis. In recent years, statistical models have been used for the task. However, such models without any domain knowledge often produce aspects that are not interpreta- ble in applications. To tackle the issue, some knowledge-based topic models have been proposed, which allow the user to input some prior domain knowledge to generate coherent aspects. However, existing knowledge-based topic models have several major shortcom- ings, e.g., little work has been done to incor- porate the cannot-link type of knowledge or to automatically adjust the number of topics based on domain knowledge. This paper pro- poses a more advanced topic model, called MC-LDA (LDA with m-set and c-set), to ad- dress these problems, which is based on an Extended generalized Pólya urn (E-GPU) model (which is also proposed in this paper). Experiments on real-life product reviews from a variety of domains show that MC- LDA outperforms the existing state-of-the-art models markedly. 1 Introduction In sentiment analysis and opinion mining, aspect extraction aims to extract entity aspects or features on which opinions have been expressed (Hu and Liu, 2004; Liu, 2012). For example, in a sentence “The picture looks great,” the aspect is “picture.” Aspect extraction consists of two sub-tasks: (1) extracting all aspect terms (e.g., “picture”) from the corpus, and (2) clustering aspect terms with similar meanings (e.g., cluster “picture” and “pho- to” into one aspect category as they mean the same in the domain “Camera”). In this work, we adopt the topic modeling approach as it can per- form both sub-tasks simultaneously (see § 2). Topic models, such as LDA (Blei et al., 2003), provide an unsupervised framework for extracting latent topics in text documents. Topics are aspect categories (or simply aspects) in our context. However, in recent years, researchers have found that fully unsupervised topic models may not pro- duce topics that are very coherent for a particular application. This is because the objective functions of topic models do not always correlate well with human judgments and needs (Chang et al., 2009). To address the issue, several knowledge-based topic models have been proposed. The DF-LDA model (Andrzejewski et al., 2009) incorporates two forms of prior knowledge, also called two types of constraints: must-links and cannot-links. A must-link states that two words (or terms) should belong to the same topic whereas a cannot- link indicates that two words should not be in the same topic. In (Andrzejewski et al., 2011), more general knowledge can be specified using first- order logic. In (Burns et al., 2012; Jagarlamudi et al., 2012; Lu et al., 2011; Mukherjee and Liu, …

Content truncated to 500 words