Exploring the Use of Lexical and Psycho-Linguistic Resources for Sentiment Analysis

摘要

Twitter data have been used for monitoring customer opinions in different domains. In this paper, we investigated the role of a wide set of lexical and psycho-linguistic resources for discovering a set of features that can be potentially useful for this task. Different experiments were carried out, the obtained results allow us to demonstrate that by using a small set of features it is possible to obtain competitive results against more complex approaches. The experimental evaluation was carried out on a collection of tweets corresponding to the airline domain. The obtained results allow us to observe a significant reduction in the vector representation dimension, maintaining satisfactory results using lexical resources for capturing different kinds of information to perform sentiment analysis in tweets on the aforementioned domain.