Participation de l’IRISA à DeFT 2018 : classification et annotation d’opinion dans des tweets (IRISA at DeFT 2018: classifying and tagging opinion in tweets )

IRISA at DeFT 2018: classifying and tagging opinion in tweets This paper describes the systems developed at IRISA by the LinkMedia team for the challenge DeFT 2018. The challenge focuses on opinion mining in French tweets about transports. The team has participated in 3 out of the 4 tasks: (i) classification of the tweets whether they are about transports or not, (ii) classification of the tweets according to their polarity and (iii) fine grained annotation of the sentiment expression and the object about which an opinion is given. For the tasks 1 and 2, we have used a boosting algorithm as well as recurrent neural networks (RNN). For the 3rd task, we have experimented the use of recurrent neural networks combined with some CRF. All the approaches give close results, with a slight advantage when using RNN, and yields among the best results for every tasks. MOTS-CLÉS : analyse d’opinion, boosting, arbres de décision, réseau de neurones récurrents, plongement de mots, CRF.