Inference Annotation of a Chinese Corpus for Opinion Mining

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4991–4999 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 4991 Inference Annotation of a Chinese Corpus for Opinion Mining Liyun Yan1, Danni E1, Mei Gan1, Cyril Grouin1,2, Mathieu Valette1 1INaLCO, ER-TIM, 2 rue de Lille, F-75007 Paris, France 2Universit´e Paris-Saclay, CNRS, LIMSI, Bˆat 507, Campus universitaire, F-91405 Orsay, France liyun.yan@inalco.fr, danni.e@paris-sorbonne.fr, gan.mei@outlook.com, cyril.grouin@limsi.fr, mvalette@inalco.fr Abstract Polarity classification (positive, negative or neutral opinion detection) is well developed in the field of opinion mining. However, existing tools, which perform with high accuracy on short sentences and explicit expressions, have limited success interpreting narrative phrases and inference contexts. In this article, we will discuss an important aspect of opinion mining: inference. We will give our definition of inference, classify different types, provide an annotation framework and analyze the annotation results. While inferences are often studied in the field of Natural-language understanding (NLU), we propose to examine inference as it relates to opinion mining. Firstly, based on linguistic analysis, we clarify what kind of sentence contains an inference. We define five types of inference: logical inference, pragmatic inference, lexical inference, enunciative inference and discursive inference. Second, we explain our annotation framework which includes both inference detection and opinion mining. In short, this manual annotation determines whether or not a target contains an inference. If so, we then define inference type, polarity and topic. Using the results of this annotation, we observed several correlation relations which will be used to determine distinctive features for automatic inference classification in further research. We also demonstrate the results of three preliminary classification experiments. Keywords: Inferences, Natural Language Processing, Opinion Mining 1. Introduction The Internet provides users with the possibility to share their opinions and sentiments on various aspects of daily life. This has become a challenge for companies (reputa- tion, customer satisfaction, etc.) and alters the way clients behave because they can freely express their negative com- ments whilst hiding behind their screens. These comments can be interpreted differently according to different cultural and social codes, whose diversity has a large influence on vocabulary usage (generic vs domain-specific), judgment criteria (room size, facilities and hotel services), as well as means of expression (especially elements considered nega- tive). There is no doubt that analysis of massive amounts of com- ment data requires automatic opinion mining tools to an- alyze the content and identify expressed trends. Further, these tools should be able to find the hidden meaning of inferences in specific contexts. Existing tools/models can accurately predict the writer’s at- titude in simple explicit sentences which contain opinion words, such as 酒店地理位置很好 (The hotel location is very good) or 房间空间小 (The room is small).1 However, polarity is not obvious when there is inference involved. The hotel is close to the Eiffel Tower also means The hotel location is very good. In other words, the lack of opinion words in sentences may increase the difficulty of detecting the attitude of the writer. Interpreting …

Content truncated to 500 words