A Multi-cluster Boundary Learning Method for Out-of-Scope Intent Detection via MiniLM Embedding

Jul 16, 2026

Upload by Riley

Intent detection bridges user intents and system actions, yet out-of-scope (OOS) intent rejection remains challenging in open intent settings. Traditional approaches treat OOS detection as multi-class classification, causing performance to drop as known intent classes grow, while LLM-embedding methods are costly and difficult to deploy. The proposed multi-cluster boundary learning method uses MiniLM (all-MiniLM-L6-v2) embeddings in a one-class workflow to learn cluster boundaries and reject out-of-domain utterances as OOS. Experiments on CLINC150, StackOverflow, and Banking77 show state-of-the-art results and ablation gains with MiniLM, with code provided in supplementary materials.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

Abstract

Introduction

Motivation and challenges

Related work

Proposed approach and contributions

Experiments

A Multi-cluster Boundary Learning Method for Out-of-Scope Intent Detection via MiniLM Embedding Yihong Xu1 , Mingyu Kang1 * , Linyuan Lü1 * 1University of Science and Technology of China, Hefei, China [xuyihong@mail.ustc.edu.cn](xuyihong@mail.ustc.edu.cn)[ ](xuyihong@mail.ustc.edu.cn){kangmingyu, [linyuan.lv}@ustc.edu.cn](linyuan.lv}@ustc.edu.cn) arXiv :2607 .07974v 1 [ cs .CL] 8 Jul 2026 Abstract Intent detection is a critical task that bridges human intents and system actions in humanmachine interaction systems. However, there still exist challenges for detecting out-of-scope (OOS) intents. (i) The traditional methods view the OOS intent detection as a multi-class classiﬁcation, then the detection accuracy decreases as the class number of the known intents increases; (ii) LLM-embedding methods require large parameters, that makes them difﬁcult to train and practically deploy. Thus, this work proposes a multi-cluster boundary learning method to detect OOS intents via MiniLM embedding (i.e., all-MiniLM-L6-v2) in an one-class classiﬁcation workﬂow. The method learns the boundaries of multi-cluster embeddings generated by MiniLM from the training utterances, and then rejects the out-of-domain utterances as OOS intents. Experiments are conducted on public CLINC150, StackOverﬂow and Banking77 datasets. The results show that the method achieves the state-of-the-art OOS intent detection performance compared the other baselines. Ablation studies are also conducted and the results show that the used MiniLM can better adapt to the workﬂow and utterance embedding requirements. The code is available at supplementary materials. 1 Introduction Intent detection is a critical yet challenging task for human-machine interaction systems. It maps user utterances to system actions, which bridges human intents and system behaviors (Tur et al., 2010 ; Louvan and Magnini, 2020 ; Wölﬂein et al., 2025) . However, human intents are complex and cannot be fully enumerated in a closed-set intent detection system. That means if the system receives an out-of-scope (OOS) intent utterance but fails to reject it, that will trigger incorrect actions * Corresponding authors. Figure 1: A diagram of OOS intent detection and induce harmful consequences, as shown in ﬁgure 1. Thus, a more important and difﬁcult task is to detect the OOS intents and reject them in a timely manner, which is actually an open intent detection task (Hoffman et al., 2024 ; Muzahid et al., 2024) . There are three types of intent detection methods, including statistical methods, deep-learning methods and LLM-embedding methods. Among them, the statistical methods formulate intent detection as a supervised text classiﬁcation problem. These methods combine manually designed lexical features with traditional statistical classiﬁers, such as support vector machine (Haffner et al., 2003) and naive Bayes classiﬁer (Schuurmans and Frasincar, 2020) . But the deep-learning methods use deep neural networks, e.g., convolutional neural network (CNN) (Kim, 2014), recurrent neural network (RNN) (Liu and Lane, 2016) and transformer network (Vaswani et al., 2017), to perform intent detection via semantic feature extraction. Early deep-learning classiﬁers are trained on a closed-set mode, and then identify unknown out-of-domain utterances as known intents (Hendrycks and Gimpel, 2017 ; Guo et al., 2017) . Thus, some types of open intent detection are further proposed to conduct safe rejection for the unknown utterances with OOS in- tents. They are class-wise open classiﬁcation method (Shu et al., 2017), distance-based scoring method (Lee et al., 2018), contrastive representation learning method (Zeng et al., 2021) and adaptive boundary learning method (Zhang et al., 2021) . After that, with the strong semantic representation ability of LLMs, the LLM-embedding methods are proposed to extract utterance features by semantic embedding, prompt-based reasoning and uncertainty-aware agent rou

FAQs

How does the proposed method use MiniLM for OOS detection?

The method embeds utterances with MiniLM (all-MiniLM-L6-v2), learns boundaries among multiple embedding clusters, and rejects utterances that fall outside learned boundaries as OOS intents in a one-class classification workflow.

Jul 16, 2026

Upload by Riley