Meta-Intents: Interaction Preferences in Conversational Recommender Systems
Ma, Y. (2025). [PhD thesis].
Abstract
With advancements in natural language processing, dialogue systems have become increasingly proficient in comprehending users’ intricate utterances. This convergence, when integrated with the application contexts of recommendation systems, has given rise to a burgeoning research domain known as conversational recommender systems (CRS). CRS can obtain user preferences through text conversations with users and recommends appropriate products. Different from traditional RS, CRS has a higher degree of interaction freedom, which makes us expect that CRS can interact with customers like a real-life salesperson. For example, it can not only ask users for features requirements, but also remind users to compare the current product with other candidates. More personalized interactive functions can emerge in CRS scenarios and make CRS more efficient and user-friendly, at the same time, it also puts forward new requirements for the user model. Different users and contexts necessitate distinct interactive functionalities, highlighting the importance of understanding and describing users’ interactive preferences—a relatively underexplored area of research. In this work, we propose the concept of meta-intents (MI) which represent high-level user preferences related to the interaction styles and decision-making support in conversational recommender systems (CRS). We developed a stable instrument to measure MI via a two-stage factor analysis, revealing that MI can be linked to users’ general decision-making styles. This connection allows for the translation of broad psychological user characteristics into more concrete design guidance for CRS. After proposing MI, we conducted a series of user studies to analyze how they impact various users’ interaction behaviors in a real instance of CRS. We analyzed the observed phenomena and extracted some helpful suggestions for interaction design. These suggestions can help us understand the role of MI in actual scenarios and build a CRS interaction strategy that is more in line with user habits.