Adapting conversational strategies in information-giving human-agent interaction

Area of Science:

• Human-Computer Interaction
• Artificial Intelligence
• Social Robotics

Background:

• Optimizing information delivery in human-agent interactions is crucial for effective communication.
• Socially interactive agents require adaptive strategies to maintain user engagement.
• User engagement is influenced by factors like personality, interest, and attitude.

Purpose of the Study:

• To develop a dialog manager that adapts conversational strategies to user preferences.
• To maximize user engagement during human-agent interactions.
• To investigate the impact of adaptive strategies on user perception.

Main Methods:

• Utilized reinforcement learning to train an agent for adaptive dialog management.
• Measured user engagement through non-verbal behaviors and turn-taking status.
• Incorporated engagement metrics into the reward function to balance task and social goals.
• Conducted a subjective study with 120 participants to assess observer perception of adaptation.

Main Results:

• The adaptive dialog model demonstrated an influence on user engagement.
• Personalized conversational strategies were perceived by third-party observers.
• The reward function effectively balanced information delivery with maintaining user engagement.

Conclusions:

• Adaptive conversational strategies enhance human-agent interactions.
• Reinforcement learning is a viable approach for developing socially interactive agents.
• User perception is a key factor in evaluating the success of adaptive dialog systems.