Skip to Main content Skip to Navigation

Modèles thématiques pour la découverte non supervisée de points de vue sur le Web

Abstract : The advent of online platforms such as weblogs and social networking sites provided Internet users with an unprecedented means to express their opinions on a wide range of topics, including policy and commercial products. This large volume of opinionated data can be explored and exploited through text mining techniques known as opinion mining or sentiment analysis. Contrarily to traditional opinion mining work which mostly focuses on positive and negative opinions (or an intermediate in-between), we study a more challenging type of opinions: viewpoints. Viewpoint mining reaches beyond polarity-based opinions (positive/negative) and enables the analysis of more subtle opinions such as political opinions. In this thesis, we proposed unsupervised approaches – i.e., approaches which do not require any labeled data – based on probabilistic topic models to jointly discover topics and viewpoints expressed in opinionated data. In our first contribution, we explored the idea of separating opinion words (specific to both viewpoints and topics) from topical, neutral words based on parts of speech, inspired by similar practices in the litterature of non viewpoint-related opinion mining. Our second contribution tackles viewpoints expressed by social network users. We aimed to study to what extent social interactions between users – in addition to text content – can be beneficial to identify users' viewpoints. Our different contributions were evaluated and benchmarked against state-of-the-art baselines on real-world datasets.
Complete list of metadatas

Cited literature [202 references]  Display  Hide  Download
Contributor : Thibaut Thonet <>
Submitted on : Monday, December 4, 2017 - 5:24:02 PM
Last modification on : Friday, January 10, 2020 - 9:09:24 PM


Files produced by the author(s)


  • HAL Id : tel-01655278, version 1


Thibaut Thonet. Modèles thématiques pour la découverte non supervisée de points de vue sur le Web. Informatique et langage [cs.CL]. Université Toulouse 3 – Paul Sabatier, 2017. Français. ⟨tel-01655278v1⟩



Record views


Files downloads