ReaderBench – Sentiment Analysis on Texts

The component analyses a text and is focused on the identification of six major sentiment valences (excited, sad, scared, angry, tender and happy). Our model integrates Cohesion Network Analysis from the ReaderBench framework with a wide variety of lexical resources (e.g., General Inquirer, Laswell dictionary, SenticNet, EmoLex, SenticNet, GALC, ANEW, LIWC, and Hu-Liu polarity lists for English language; Affective norms for French words and emoLex for French language), as well as the sentiment analysis model from Stanford Core NLP, based on recursive deep networks. Given an input text, this component returns a normalized score expressing the strength of each sentiment dimension. The component is available in the following languages: English and French. Dutch and Romanian languages will be available soon, too.


Current status

  • Working on providing support for additional languages besides English and French
  • Fine-tuning word vectors for particular positive/negative corpora