Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Granularity Open Intent Classification via Adaptive Granular-Ball Decision Boundary

Yanhua Li, Xiaocao Ouyang, Chaofan Pan, Jie Zhang, Sen Zhao, Shuyin Xia, Xin Yang, Guoyin Wang, Tianrui Li

TL;DR

Open intent classification requires correctly labeling known intents while detecting unknown ones. The paper introduces MOGB, a two-module framework that combines hierarchical representation learning via adaptive granular-ball clustering with a nearest sub-centroid classifier, and multi-granularity decision boundaries to separate known from open intents. This approach addresses non-spherical distributions and intra-open intents, achieving strong results across StackOverflow, SNIPS, and BANKING, especially when unknowns predominate. It provides a scalable, end-to-end training paradigm that integrates representation learning and boundary acquisition, yielding reduced empirical and open-space risks and improved interpretability through multi-scale granular-balls. Overall, MOGB offers a robust solution for open-world intent classification with practical impact for dialogue systems.

Abstract

Open intent classification is critical for the development of dialogue systems, aiming to accurately classify known intents into their corresponding classes while identifying unknown intents. Prior boundary-based methods assumed known intents fit within compact spherical regions, focusing on coarse-grained representation and precise spherical decision boundaries. However, these assumptions are often violated in practical scenarios, making it difficult to distinguish known intent classes from unknowns using a single spherical boundary. To tackle these issues, we propose a Multi-granularity Open intent classification method via adaptive Granular-Ball decision boundary (MOGB). Our MOGB method consists of two modules: representation learning and decision boundary acquiring. To effectively represent the intent distribution, we design a hierarchical representation learning method. This involves iteratively alternating between adaptive granular-ball clustering and nearest sub-centroid classification to capture fine-grained semantic structures within known intent classes. Furthermore, multi-granularity decision boundaries are constructed for open intent classification by employing granular-balls with varying centroids and radii. Extensive experiments conducted on three public datasets demonstrate the effectiveness of our proposed method.

Multi-Granularity Open Intent Classification via Adaptive Granular-Ball Decision Boundary

TL;DR

Open intent classification requires correctly labeling known intents while detecting unknown ones. The paper introduces MOGB, a two-module framework that combines hierarchical representation learning via adaptive granular-ball clustering with a nearest sub-centroid classifier, and multi-granularity decision boundaries to separate known from open intents. This approach addresses non-spherical distributions and intra-open intents, achieving strong results across StackOverflow, SNIPS, and BANKING, especially when unknowns predominate. It provides a scalable, end-to-end training paradigm that integrates representation learning and boundary acquisition, yielding reduced empirical and open-space risks and improved interpretability through multi-scale granular-balls. Overall, MOGB offers a robust solution for open-world intent classification with practical impact for dialogue systems.

Abstract

Open intent classification is critical for the development of dialogue systems, aiming to accurately classify known intents into their corresponding classes while identifying unknown intents. Prior boundary-based methods assumed known intents fit within compact spherical regions, focusing on coarse-grained representation and precise spherical decision boundaries. However, these assumptions are often violated in practical scenarios, making it difficult to distinguish known intent classes from unknowns using a single spherical boundary. To tackle these issues, we propose a Multi-granularity Open intent classification method via adaptive Granular-Ball decision boundary (MOGB). Our MOGB method consists of two modules: representation learning and decision boundary acquiring. To effectively represent the intent distribution, we design a hierarchical representation learning method. This involves iteratively alternating between adaptive granular-ball clustering and nearest sub-centroid classification to capture fine-grained semantic structures within known intent classes. Furthermore, multi-granularity decision boundaries are constructed for open intent classification by employing granular-balls with varying centroids and radii. Extensive experiments conducted on three public datasets demonstrate the effectiveness of our proposed method.

Paper Structure

This paper contains 24 sections, 7 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Open Intent Classification
  4. Granular-ball Computing
  5. Our Methodology
  6. Problem Statement and Model Overview
  7. Problem Statement
  8. Model Overview
  9. Hierarchical Representation Learning
  10. Feature Extraction
  11. Adaptive Granular-ball Clustering
  12. Nearest Sub-Centroid Classifier
  13. Multi-Granularity Decision Boundary in Inference
  14. Experiments
  15. Datasets
  16. ...and 9 more sections

Figures (3)

  • Figure 1: (a) Previous boundary-based methods struggle to learn a boundary for open intent classification by balancing open space risk and empirical risk. (b) Our proposed Multi-granularity decision boundary can effectively eliminate both two risks.
  • Figure 2: The architecture of MOGB. In the hierarchical representation learning module, we generate granular-balls on all known intents via adaptive granular-ball clustering and then use the nearest sub-centroid classifier to learn representation. In the boundary acquiring module, multi-granularity decision boundaries are established.
  • Figure 3: Effect of $n_t$ on BANKING with 50% known class. X-axis represents the value of $n_t$, the left Y-axis denotes the values of four metrics, and the right Y-axis indicates the number of established decision boundaries for all known classes.