Improving Dialogue Processing with EASE-DRCBot: A Comprehensive Approach to Understanding User Utterances

EASE-DRCBot is a cutting-edge dialogue system developed for DRC2023, with the primary goal of facilitating natural and efficient conversations between humans and machines. The system utilizes three key techniques to understand user utterances: keyword extraction, example-based method, and sentiment analysis.
Keyword Extraction: EASE-DRCBot employs keyword extraction to identify specific words or phrases in a user’s utterance, enabling it to better comprehend the user’s intentions. This approach helps the system address errors in sentiment analysis by identifying relevant keywords.
Example-Based Method: When the defined keywords are not present in the user’s utterance, EASE-DRCBot uses an example-based method, which involves analyzing similar sentences or phrases to understand the user’s intentions. This approach leverages BERT models and cosine similarity to identify relevant examples.
Sentiment Analysis: To determine whether a user’s utterance is positive or negative, EASE-DRCBot utilizes sentiment analysis, which involves analyzing the tone and context of the user’s words using BERT models. This approach helps the system understand the user’s emotions and respond accordingly.
Phases of Dialogue Flow: EASE-DRCBot divides the dialogue flow into three distinct phases: recommending tourist routes, answering questions, and providing a response using GPT-3.5. These phases ensure a smooth conversation and minimize dialogue breakdowns.
Key Features: Some key features of EASE-DRCBot include inducing user utterances to encourage a smooth conversation, responding using GPT-3.5 when necessary, and utilizing OpenAI’s GPT-3.5 API for signaling to the user that a response is being prepared.
Evaluation: The system’s scores on informativeness, naturalness, appropriateness, likeability, satisfaction with dialogue, trustworthiness, usefulness, and correctness were relatively high, suggesting that EASE-DRCBot’s strategy of presenting two sight-seeing spots and explaining the reasons based on the user’s information is effective.
In conclusion, EASE-DRCBot represents a significant breakthrough in dialogue systems, offering a more natural and efficient conversational experience for users. By leveraging keyword extraction, example-based method, and sentiment analysis, the system can understand user utterances with remarkable accuracy, making it an excellent tool for various applications.