为解决匮乏资源下多意图识别语义、语境信息易受到不相关意图信息干扰的问题,该文提出一种基于原型网络在语义上嵌入意图信息的多意图识别方法。首先设计意图融合特征提取机制,通过结合话语和意图信息构建具有区分度的支持集、查询句和意图集表征,缓解短话语往往遭遇意图相关信息的语义混淆的问题;其次设计原型意图分离机制,计算所属意图话语对该意图原型的权重信息,联合意图权重得到分离式意图原型表征,降低支持集和查询句中不相关意图带来的噪声。实现了在低资源多意图场景下捕获高质量的原型表征。实验结果表明,该方法可有效提高小样本多意图识别的效果。
Abstract
To address the few-shot multi-intent recognition, the issue of irrelevant intention interference in the semantic and contextual information, a prototype network based multi-intention recognition method is proposed with intention embedding in the input. Firstly, the intention feature extraction mechanism is designed to construct representative support set, query sentence and intention set by combining discourse and intention information. Secondly, the intent prototype detection mechanism is designed to calculate the weight of the intent discourse belonging to the intent prototype, and the final intent prototype representation to associated with such weight and the likelihood of the intent. High-quality prototype characterization is thus achieved for capturing low-resource multi-intent scenarios. The experimental results show that the method can effectively improve the performance of few-shot multi-intent recognition.
关键词
多意图识别 /
小样本学习 /
语义混淆 /
低资源 /
原型网络
{{custom_keyword}} /
Key words
multi-intent recognition /
few-shot learning /
semantic obfuscation /
low-resource /
prototype networks
{{custom_keyword}} /
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 熊伟,宫禹.基于元学习的不平衡少样本情况下的文本分类研究[J].中文信息学报,2022,36(01): 104-116.
[2] WU M. Few-shot text classification with distributional signatures[D]. B.S. Thesis, Boston: Massachusetts Institute of Technology, 2020.
[3] OHASHI S,TAKAYAMA J, KAJIWARA T, et al. Distinct label representations for few-shot text classification[C]//Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, 2021: 831-836.
[4] LUO Q, LIU L, LIN Y, et al. Don’t miss the labels: Label-semantic augmented meta-learner for few-shot text classification[C]//Proceedings of the Association for Computational Linguistics, 2021: 2773-2782.
[5] HAN C, FAN Z, ZHANG D, et al. Meta-learning adversarial domain adaptation network for few-shot text classification[C]//Proceedings of the Association for Computational Linguistics, 2021: 1664-1673.
[6] 陆恒杨,范晨悠,吴小俊.面向网络社交媒体的少样本新冠谣言检测[J].中文信息学报,2022, 36(01): 135-144.
[7] SIMON C,KONIUSZ P, HARANDI M. Meta-learning for multi-label few-shot classification[C]//Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2022: 3951-3960.
[8] HU M, ZHAO S, GUO H, et al. Multi-label few-shot learning for aspect category detection[C]//Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, 2021: 6330-6340.
[9] HOU Y, LAI Y, WU Y, et al. Few-shot learning for multi-label intent detection[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2021, 35(14): 13036-13044.
[10] LAI V, DERNONCOURT F, NGUYEN T H. Learning prototype representations across few-shot
tasks for event detection[C]//Proceedings of the Conference on Empirical Methods in Natural Language Processing, 2021: 5270-5277.
[11] FINN C B. Learning to learn with gradients[D]. PHD Thesis, Berkeley: University of California, 2018.
[12] QIN L, XU X, CHE W, et al. TD-GIN: Token-level dynamic graph-interactive network for joint multiple intent detection and slot filling[J] .arXiv preprint arXiv: abs/2004.10087.2020.
[13] CHOI H, KIM J, JOE S, et al. Evaluation of bert and albert sentence embedding performance on downstream nlp tasks[C]//Proceedings of the 25th International Conference on Pattern Recognition. IEEE, 2021: 5482-5487.
[14] HOU Y, LIU Y, CHE W, et al. Sequence-to-sequence data augmentation for dialogue language understanding[C]//Proceedings of the 27th International Conference on Computational Linguistics, 2018: 1234-1245.
[15] HOU Y, CHE W, LAI Y, et al. Few-shot slot tagging with collapsed dependency transfer and label-enhanced task-adaptive projection network[C]//Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 2020: 1381-1393.
[16] KENTON J D M W C, TOUTANOVA L K. BERT: Pre-training of deep bidirectional transformers for language understanding[C]//Proceedings of NAACL-HLT, 2019: 4171-4186.
[17] REIMERS N, GUREVYCH I. Reporting score distributions makes a difference: Performance study of lstm-networks for sequence tagging[C]//Proceedings of the Conference on Empirical Methods in Natural Language Processing, 2017: 338-348.
[18] SNELL J, SWERSKY K, ZEMEL R. Prototypical networks for few-shot learning[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems, 2017: 4080-4090.
[19] VINYALS O, BLUNDELL C, LILLICRAP T, et al. Matching networks for one shot learning[C]//Proceedings of the 30th International Conference on Neural Information Processing Systems, 2016: 3637-3645.
{{custom_fnGroup.title_cn}}
脚注
{{custom_fn.content}}
基金
国家自然科学基金(U1836103);四川省科技重点研发项目(18ZDYF2039)
{{custom_fund}}
PDF(1777 KB)
