An archive for my past research in conversational recommendation.
Conversational recommendation (2018-2020)
Users have a hard time “telling” recommendation systems what they like. Recent advances in dialogue systems unlock new possibilities for user interactions. My co-authored papers in 2020 studied this topic.
Even though I am not studying this topic currently, I wish following aspects👇 can be considered by our community in the future.
- Taxonomy creation: Existing feature taxonomies are handcrafted. Is there a smart/soft way to build a taxonomy for better system design and evaluation?
- Discourse structure: Existing works (ours included) consider multi-round dialogue in a superficial manner. What features/products in previous rounds contrast with those later? What information users heard in previous rounds caused their later decision?
- Real-world applications: Many works (ours included) are developed and evaluated in the sandbox – user simulation. How can we develop systems with (and for) real-world users (with diverse backgrounds, multiplex intent, and language variations)?
Recommendation: Tutorials by Wenqiang et al. A pretty cool integrated CRSLab environment by RUC. Talk to my group mate Victor Li who is currently studying this topic.