A Survey of Personality, Persona, and Profile in Conversational Agents and Chatbots

Advances and Challenges in Developing Conversational Agents with Personality

Introduction to Conversational Agents with Personality

The evolution of artificial intelligence has led to the sophisticated development of neural conversational agents (CAs), notably chatbots, capable of simulating human-like conversations. The integration of personality into these agents marks a significant stride towards creating more relatable and engaging interactions. This paper categorically surveys recent advancements in infusing personality into conversational agents, delineating personality schemes, datasets, and methods employed in the embodiment of personality. It also discusses the implications of these developments and posits future research directions in the domain.

Personality Schemes in Conversational Agents

The conceptualization of personality within conversational agents borrows extensively from psychological models, with schemes ranging from the widely acknowledged Big-5 (OCEAN) model to chatbot-specific schemes like Descriptive Sentences and Attribute-Value Pairs. The paper discusses various implementations of these schemes, revealing a trend towards the utility of Descriptive Sentences in defining chatbot personalities. This scheme's flexibility and expressiveness allow for the creation of rich, character-like personas for chatbots, amplifying user engagement.

Datasets for Personality in Chatbots

A plethora of datasets have been developed to train chatbots with personality traits, with the Persona-Chat and Image-Chat datasets exemplifying the Descriptive Sentences approach. These datasets consist of dialogues or textual descriptions designed to imbue chatbots with identifiable personality traits, catering to the nuanced demands of personalized interaction. The paper highlights the ingenuity behind these dataset designs and their crucial role in advancing chatbot personality research.

Embodiment of Personality in Models

The surveyed paper discusses a variety of methods for integrating personality into conversational agents, including pre-existing NN architectures like Seq-to-Seq and Transformer models, as well as memory networks and novel approaches such as few-shot learning. It emphasises the diversity in methodological approaches to embodying personality, from encoding personality traits directly into model architectures to leveraging previous dialogue histories for implicit personality inference.

Implications and Future Directions

The integration of personality into conversational agents has practical implications for enhancing user experience across various applications, from social companionship to customer service. Theoretically, it presents an intriguing intersection between computer science and psychology, offering insights into simulating human-like attributes in AI agents. The paper speculates on future developments which could focus on improving the fidelity of personality traits expressed by chatbots, optimizing interaction dynamics, and advancing the evaluation metrics for assessing personality consistency and user engagement.

Conclusions

This systematic survey illustrates the substantial progress made in the development of personality-infused conversational agents. Despite notable advancements, the challenge of accurately evaluating and ensuring the consistency of chatbot personalities remains. Nonetheless, the ongoing research in creating nuanced, personality-driven chatbots promises to significantly enhance human-machine interaction.

The paper serves as a comprehensive overview for researchers engaged in the ongoing quest to develop more human-like conversational agents. It not only captures the current state of the art but also delineates the trajectory for future explorations in integrating personality into conversational AI.