Development and Evaluation of an Automated e‑Counselling System for Emotion and Sentiment Analysis

Abstract

Given the challenges associated with the analysis of emotions in text by counsellors, we present an intelligent e‑counselling system for automatic detection of emotions and sentiments in text. The system‑ EmoTect‑ was developed using a supervised support vector machine learning classifier. Therefore, students’ life stories were collected and developed into a corpus for training and evaluating of the classifier. EmoTect allows users to label instances of the training data based on their own perception of emotions, and then gradually learns to classify emotions according to the user’s perceptions. The EmoTect interface provides a visualization of the emotional changes from automatically analysed students’ submissions over a selectable period. In this paper, the EmoTect classifier is evaluated with a gold standard corpus obtained from students but annotated by counsellors. In addition to the classifier evaluation, the EmoTect prototype was evaluated with counsellors in their settings. From the experimental results, the EmoTect classifier for the sentiment classification achieved comparable accuracy to that achieved with a gold standard when presented with unknown data. The contextual evaluation of the system indicates counsellors' satisfaction and sense of enthusiasm for using EmoTect for counseling delivery.