Personalised Conversational Agents: A Systematic Review

Abstract

The article discusses state-of-the-art developments in personalized conversational agents and identifies future research opportunities. Despite being in its infancy, this field aims to enhance conversational systems by incorporating auxiliary information, such as persona data. The study employed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework to analyze 257 papers from databases like Scopus, Web of Science, and Cornell University database library from 2000 to 2022. The selected articles focused on computer science and information technology, emphasizing full papers and conference contributions published in English. Notably, the study examined personal interlocutors with both implicit and explicit information, considering global representation. Among the popular corpora for training and evaluation were Persona-Chat, Personal-Dialog, and ConvAI2. Response generation techniques included autoregressive models, Generative Adversarial Networks (GANs), and Transformer-based approaches. Metrics for persona-based conversational agents included perplexity, N-gram overlap, distinct n-grams, model-based metrics, and subjective evaluation. China led in publications during 2021. In the future, persona-based conversational agents will aim for more personalized interactions by integrating multimodal capabilities, emotional intelligence, and voice assistant integration.