Sentimental Analysis for AIML-Based E-Health Conversational Agents

Conversational agents or chat-bots are emerging in various applications including finance, education and e-health. Recent research has highlighted the importance of the consistency between the response of the chat-bot and the sentiment of the input utterance. This is quite challenging as detecting the sentiment of an utterance often depends on the context and timing of the conversation. Moreover, whereas humans have complex repair strategies, encoding these for human-computer interaction is problematic. This paper presents five sentiment prediction models for conversational agents that are trained on a large corpus of smartphone application reviews and their sentiment ranks obtained from the Google playstore. These models are tested on collected, real-life conversations between a human and a machine. It is found that positive utterances are classified with a high accuracy but classifying negative utterances is still challenging.