This article presents a text mining approach based on linguistic knowledge to automatically detect opinion targets in relation with topic elements, for competitive intelligence. The identification of opinions and sentiments expressed in texts is currently studied a lot, but few works are focused on the identification of opinions whose target is associated to a predefined topic. We present in a first time the information detection task with linguistic patterns, and a state of the art on opinion detection and opinion targets detection. Then we describe the French corpora: one contains transcription of telephone requests related to Energy, the other contains extracts of internet forum related to Video Games. Next we detail the linguistic knowledge used to annotate those texts. The knowledge is based on the identification of explicit relations between topic and opinion ("Transformers is great") and on the identification of implicit opinions ("they intervene quickly"). At last, an example of result is presented, as a first evaluation.
展开▼
