From opinion classification to recommendations: How texts from a social network can help

This paper is about opinion classification of posts from a social networks by supervised machine learning, in order to use them in a recommender system. We compare different pre-processings, representations and machine learning tools on real data about movies having specificities (very short texts in English, containing a lot of sms-like codes, abbreviations, misspelling...). We study in detail the results of different classifiers and the contribution of the pre-processings on this kind of data. Finally, we evaluate the best classifier with a recommender system based on collaborative filtering. MOTS-CLÉS : classification d’opinions, apprentissage supervisé, textes communautaires, cyberlangue, recommandation automatique, filtrage collaboratif, cold-start.