Datasets for Aspect-Based Sentiment Analysis in French

Datasets for Aspect-Based Sentiment Analysis in French Marianna Apidianaki1, Xavier Tannier2, C´ecile Richart1 1. LIMSI, CNRS, Universit´e Paris-Saclay, Rue John von Neumann, 91403 Orsay, France 2. LIMSI, CNRS, Univ. Paris-Sud, Universit´e Paris-Saclay, Rue John von Neumann, 91403 Orsay, France firstname.surname@limsi.fr Abstract Aspect Based Sentiment Analysis (ABSA) is the task of mining and summarizing opinions from text about specific entities and their aspects. This article describes two datasets for the development and testing of ABSA systems for French which comprise user reviews annotated with relevant entities, aspects and polarity values. The first dataset contains 457 restaurant reviews (2365 sentences) for training and testing ABSA systems, while the second contains 162 museum reviews (655 sentences) dedicated to out-of-domain evaluation. Both datasets were built as part of SemEval-2016 Task 5 “Aspect-Based Sentiment Analysis” where seven different languages were represented, and are publicly available for research purposes. This article provides examples and statistics by annotation type, summarizes the annotation guidelines and discusses their cross-lingual applicability. It also explains how the data was used for evaluation in the SemEval ABSA task and briefly presents the results obtained for French. Keywords: French Aspect-Based Sentiment Analysis, opinion mining, user reviews. 1. Introduction Aspect Based Sentiment Analysis (ABSA) is the task of mining and summarizing opinions from text about specific entities and their aspects (Hu and Liu, 2004; Ganu et al., 2009). We have created and present two new datasets for the development and testing of ABSA systems for French which have been released in the framework of the SemEval- 2016 Task 5 “Aspect-Based Sentiment Analysis”. The first dataset comprises French restaurant user reviews annotated with relevant entities, aspects and polarity values. For this domain, both training and test data was provided. The second dataset is a smaller set of museum reviews for which no training data is available, dedicated to out-of-domain ABSA evaluation. The ABSA task was first introduced for English in the SemEval 2014 evaluation campaign (Pontiki et al., 2014), where restaurant and laptop reviews annotated with aspect terms, categories and their polarity were provided for training and testing ABSA systems. The task was repeated in SemEval 2015 with a different, more unified, framework where aspect categories were defined as combinations of an entity type, an attribute type and a polarity value (Pontiki et al., 2015): (1) The fajitas were delicious, but expensive. {FOOD#QUALITY, TARGET: fajitas}→POSITIVE {FOOD#PRICES, TARGET: fajitas}→NEGATIVE (2) Great for a romantic evening. {AMBIENCE#GENERAL, TARGET: NULL}→ POSITIVE An out-of-domain subtask was also proposed where annotated hotel reviews were provided for testing but no training data was released. In 2016, the SemEval ABSA task became multilingual (Pontiki et al., 2016). New datasets were released for English allowing systems to be tested on the same domains as in previous years (laptops, restaurants and hotels), but datasets were also developed in new languages using a common set of annotation guidelines. More precisely, the multilingual dataset created for this year’s task comprises Restaurant reviews in six languages (English, French, Dutch, Russian, Spanish and Turkish), Hotel reviews in Arabic, …

Content truncated to 500 words