Multi-Granularity Open Intent Classification via Adaptive Granular-Ball Decision Boundary
Authors: Yanhua Li, Xiaocao Ouyang, Chaofan Pan, Jie Zhang, Sen Zhao, Shuyin Xia, Xin Yang, Guoyin Wang, Tianrui Li
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments conducted on three public datasets demonstrate the effectiveness of our proposed method. |
| Researcher Affiliation | Academia | 1School of Computing and Artificial Intelligence, Southwestern University of Finance and Economics, Chengdu 611130, China 2Engineering Research Center of Intelligent Finance, Ministry of Education, Chengdu 611130, China 3Key Laboratory of Big Data Intelligent Computing, Chongqing Key Laboratory of Computational Intelligence, Chongqing University of Posts and Telecommunications, Chongqing 400065, China 4National Center for Applied Mathematics in Chongqing, Chongqing Normal University, Chongqing 401331, China 5School of Computing and Artificial Intelligence, Southwest Jiaotong University, Chengdu 611756, China |
| Pseudocode | No | The paper describes methods using definitions, equations, and figures, but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks. |
| Open Source Code | Yes | Code https://github.com/Liyanhuaa/MOGB |
| Open Datasets | Yes | In accordance with prior studies (Zhang, Xu, and Lin 2021; Zhang et al. 2023a), we conduct experiments on three commonly used datasets: Stack Overflow (Xu et al. 2015) is a dataset containing 3,370,528 programming question titles across 20 categories. SNIPS (Coucke et al. 2018) is composed of 14,484 spoken utterances with 7 distinct categories of intent classes. BANKING (Casanueva et al. 2020) is an online banking inquiry with 13,080 instances and 77 intents classes. |
| Dataset Splits | Yes | Dataset #Class #Train/Valid/Test Length Stack Overflow 20 12,000/2,000/6,000 9.18 SNIPS 7 13,084/700/700 9.05 BANKING 77 9,003/1,000/3,080 11.91 ... The experimental results are presented based on three datasets, with known class ratios set at 25%, 50%, and 75%, respectively. |
| Hardware Specification | No | The paper mentions using a pre-trained BERT model but does not specify any hardware details (e.g., GPU, CPU models, or cloud computing specifications) used for training or inference. |
| Software Dependencies | No | The paper mentions using the "pre-trained BERT model" but does not specify any software libraries or frameworks with their version numbers that are critical for reproducing the experiments. |
| Experiment Setup | Yes | The training batch size is set to 128, with a learning rate of 2e-5 to finetune the final layer. We control the adaptive granular-ball clustering by two attributes of the granular-ball: purity limit pl and sample count limit nl. We set pl = 0.9 for all datasets and varying nl for different datasets depending on the number of samples within each class. In addition, granular-balls with high quality are selected to represent the distribution by new purity limit pt and sample count limit nt. We set pt = 1 for all datasets and set varying nl based on the dataset. |