Minimally Supervised Hierarchical Domain Intent Learning for CRS

TL;DR

The paper tackles the escalating data and labeling burden of maintaining domain intent in dynamic conversational recommender systems by introducing a minimally supervised hierarchical clustering framework that blends neural attention with adaptive entropy merging, building on DEC and NAM. Using a 44,112-question business-domain corpus, the method demonstrates that a stable, multi-level intent hierarchy can be constructed with about $2.04 \times 10^4$ utterances, significantly reducing labeling and retraining needs. The approach comprises four phases—feature refinement, ANN-based clustering, entropy-guided merging with occasional contrastive refinement, and prototype selection—achieving scalable, self-learning domain knowledge without frequent full retraining. Empirical results show favorable stability and cluster quality metrics (e.g., NMI ≈ 0.80, ARI ≈ 0.65) and strong expert coherence, suggesting practical impact for cost-effective, equitable CRS deployment across evolving domains.

Abstract

Modeling domain intent within an evolving domain structure presents a significant challenge for domain-specific conversational recommendation systems (CRS). The conventional approach involves training an intent model using utterance-intent pairs. However, as new intents and patterns emerge, the model must be continuously updated while preserving existing relationships and maintaining efficient retrieval. This process leads to substantial growth in utterance-intent pairs, making manual labeling increasingly costly and impractical. In this paper, we propose an efficient solution for constructing a dynamic hierarchical structure that minimizes the number of user utterances required to achieve adequate domain knowledge coverage. To this end, we introduce a neural network-based attention-driven hierarchical clustering algorithm designed to optimize intent grouping using minimal data. The proposed method builds upon and integrates concepts from two existing flat clustering algorithms DEC and NAM, both of which utilize neural attention mechanisms. We apply our approach to a curated subset of 44,000 questions from the business food domain. Experimental results demonstrate that constructing the hierarchy using a stratified sampling strategy significantly reduces the number of questions needed to represent the evolving intent structure. Our findings indicate that this approach enables efficient coverage of dynamic domain knowledge without frequent retraining, thereby enhancing scalability and adaptability in domain-specific CSRs.

Figures (5)

• Figure 1: Minimal Data Requirement for Optimal Hierarchical Intent Clustering
• Figure 2: Dataflow of Hierarchical clustering algorithm
• Figure 3: Stability and Performance of the Hierarchical Clustering
• Figure 4: Clustering statistics
• Figure 5: Qulaity metrics