Using eye-tracking data to create a weighted dictionary for sentiment analysis: the eye dictionary

Zammarchi, Gianpaolo

Methodology

;Antoch, Jaromir

2021-01-01

---

Abstract

Extracting information from written texts is of paramount importance to many entities (e.g. businesses, public organizations, individuals), but the exponential growth of available data has made this task beyond any single human being or business. Sentiment analysis is a tool to automatically transform the information extracted into knowledge. One of the main challenges is to assess if a text is positive or negative, which can be tackled using a dictionary where each word has a positive or negative associated value and then combining single-words values to express an overall text sentiment. In order to use such lexicon-based approach, we need an existing dictionary or to build a new one. In this work we present a new dictionary for sentiment analysis developed using eye-tracking data to determine the relevance of words and we assess its performances against other existing dictionaries.