Towards Building Voice-based Conversational Recommender Systems: Datasets, Potential Solutions, and Prospects

Voice-based systems can improve accessibility: The paper identifies that voice-based conversational recommender systems (VCRSs) could benefit users with visual impairments or limited reading and writing abilities, addressing a significant gap in current text-based recommendation systems.

New datasets are essential for development: The researchers created the first benchmark datasets for voice-based conversational recommenders in e-commerce and movie domains, highlighting that such datasets were previously unavailable and are necessary to advance this emerging technology.

Synthetic data generation requires careful quality control: The authors used ChatGPT-driven prompts and text-to-speech models to create voice conversations, but deliberately implemented "strategies to ensure the naturalness and high quality of voice conversations," suggesting that synthetic audio data requires thoughtful design to sound authentic and usable.

Natural conversation is a key advantage: The paper emphasizes that voice-based systems enable "natural, intuitive, convenient" interactions compared to text-based alternatives, potentially making recommendations more accessible and user-friendly for broader populations.

End-to-end voice systems remain an open challenge: While the paper explores potential solutions for building complete voice-based systems that "seamlessly extract and integrate voice-based inputs," the abstract suggests this remains an area requiring further research and development.