AlloSat: A New Call Center French Corpus for Satisfaction and Frustration Analysis

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 1590–1597 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 1590 AlloSat: A New Call Center French Corpus for Satisfaction and Frustration Analysis Manon Macary1,2, Marie Tahon1, Yannick Est`eve3, Anthony Rousseau2 1LIUM / Le Mans Universit´e, 2Allo-M´edia, 3LIA / Universit´e d’Avignon m.macary@allo-media.fr, marie.tahon@univ-lemans.fr, yannick.esteve@univ-avignon.fr, a.rousseau@allo-media.fr Abstract We present a new corpus, named AlloSat, composed of real-life call center conversations in French that is continuously annotated in frustration and satisfaction. This corpus has been set up to develop new systems able to model the continuous aspect of semantic and paralinguistic information at the conversation level. The present work focuses on the paralinguistic level, more precisely on the expression of emotions. In the call center industry, the conversation usually aims at solving the caller’s request. As far as we know, most emotional databases contain static annotations in discrete categories or in dimensions such as activation or valence. We hypothesize that these dimensions are not task-related enough. Moreover, static annotations do not enable to explore the temporal evolution of emotional states. To solve this issue, we propose a corpus with a rich annotation scheme enabling a real-time investigation of the axis frustration / satisfaction. AlloSat regroups 303 conversations with a total of approximately 37 hours of audio, all recorded in real-life environments collected by Allo-Media (an intelligent call tracking company). First regression experiments, with audio features, show that the evolution of frustration / satisfaction axis can be retrieved automatically at the conversation level. Keywords: Speech Corpus, Call center, Emotion Recognition, Frustration Satisfaction 1. Introduction Information extraction is an important topic for research and industry. For instance, since in call center human agents receive hundreds of calls per day, it is interesting to provide indicators to help them analyzing this large amount of data. So we planned to work on emotion analysis in order to guide people into the analysis of massive conversations. The present work is the first step of a bigger project lo- cated at the cross-domain between semantic and paralin- guistic information modeling in call center spontaneous speech directly from audio signal. Call center speech, well- known for its negative emotions, is very interesting as these conversations are good representations of real-life expres- sions of speaker’s states. It also has the advantage of pro- ducing massive amounts of data even if they have to be anonymized in order to protect the private information of the callers which costs time and money partially explaining why there are not a lot of available corpora in this field. Nowadays, existing systems retrieve emotional information from the textual transcription of the conversations, in the field of sentiment analysis, opinion mining, or from audio signal, in the field of Speech Emotion Recognition (SER). To better model emotions dimensions, we first studied the two main psychological models used in affect anal- ysis to define emotions. The first one is composed of discrete emotional categories such as the “Big Six” (Ek- …

Content truncated to 500 words