Title: Sentopics: An R package linking topic models and sentiment analysis
Authors: Olivier Delmarcelle - Ghent University (Belgium) [presenting]
Abstract: The R package sentopics provides a new approach to analyse textual sentiment. By integrating topic modelling in the framework, the package breaks down classical sentiment values into topical-sentiment components. The package implements the Joint Sentiment-Topic model, a two-layer topic model incorporating sentiment lexicons to determine topical-sentiment word clusters. Sentopics also enables to join the results of a simple topic model with sentiment computed from another tool. In addition, the package defines several functions designed to enhance interpretability by vizualizing topic models results or preparing topical-sentiment time series. Finally, a built-in parallel framework is included to speed up the estimation of multiple topic models at once. The user may then compare the different estimates using coherence metrics, a tool implemented by the package to assess the quality of topic models.