Please wait a minute...
Advanced Search
数据分析与知识发现  0, Vol. Issue (): 1-     https://doi.org/10.11925/infotech.2096-3467.2018.2020.0206
  本期目录 | 过刊浏览 | 高级检索 |
基于双向长效注意力特征表达的少样本文本分类模型研究
徐彤彤,孙华志,马春梅,姜丽芬,刘逸琛 徐彤彤,孙华志,马春梅,姜丽芬,刘逸琛
(天津师范大学计算机与信息工程学院 天津  300387)
Few-shot text classification based on Bi-directional Long-term Attention feature expression Few-shot text classification based on Bi-directional Long-term Attention feature expression
Xu Tongtong, Sun Huazhi, Ma Chunmei, Jiang Lifen, Liu Yichen
(College of Computer and Information Engineering, Tianjin Normal University, Tianjin 300387, China)
全文: PDF (815 KB)  
输出: BibTeX | EndNote (RIS)      
摘要 

[目的] 针对当前文本分类任务中存在的训练数据匮乏以及模型泛化性能低等问题,在少样本环境下研究文本分类问题,提出一种少样本文本分类模型。[方法] 基于元学习中的分段训练机制将文本分类任务划分为多个子任务;为了捕捉每个子任务中文本的长效上下文信息,提出了双向时间卷积网络(Bi-directional Temporal Convolutional Network,Bi-TCN);为了捕获辨别力更强的特征,联合Bi-TCN和注意力机制提出双向长效注意力网络(Bi-directional Long-term Attention Network,BLAN);利用一种新的神经网络模型度量每个子任务中查询样本与支持集的相关性,从而实现少样本文本分类。[结果] 在ARSC数据集上进行了实验,实验结果表明,在少样本环境下,该模型的分类准确率高达86.80%,比现有先进的少样本文本分类模型ROBUSTTC-FSL和Induction-Network-Routing的精度分别提高了3.68%和1.17%。[局限] 仅针对短文本分类问题,对于篇幅较长的文本,其分类能力有限。[结论]双向长效注意力网络克服了训练数据匮乏问题且充分捕获文本的语义信息,有效提高了少样本文本分类性能。

服务
把本文推荐给朋友
加入引用管理器
E-mail Alert
RSS
作者相关文章
关键词 少样本文本分类注意力机制少样本学习双向时间卷积网络     
Abstract

[Objective] This paper studies text classification task in few-shot learning setting and proposes a few-shot text classification model to address the issues of data scarcity and low generalization performance. [Methods] We divide the text classification task into multiple subtasks based on episode training mechanism in meta-learning. We propose a Bi-directional Temporal Convolutional Network(Bi-TCN) to capture the long-term contextual information of the text in each subtask. We propose Bi-directional Long-term Attention Network(BLAN) to capture more discriminative features based on Bi-TCN and multi-head attention mechanism. Neural Tensor Network is used to measure the correlation between query samples and support set of each subtask to achieve few-shot text classification. [Results] Experiments are performed on the ARSC dataset, the experimental results show that the classification accuracy of BLAN model reaches 86.80% in few-shot learning setting, which is improved by 3.68% and 1.17% respectively compared with the existing advanced few-shot text classification models ROBUSTTC-FSL and Induction-Network-Routing. [Limitations] The performance of BLAN on long text is not satisfactory. [Conclusions] BLAN overcomes the issue of data scarcity and captures comprehensive feature information of text, which effectively improves the performance of few-shot text classification.

Key words Few-shot text classification    Attention mechanism    Few-shot learning    Bi-TCN
     出版日期: 2020-07-17
ZTFLH:  TP393  
引用本文:   
徐彤彤, 孙华志, 马春梅, 姜丽芬, 刘逸琛. 基于双向长效注意力特征表达的少样本文本分类模型研究 [J]. 数据分析与知识发现, 0, (): 1-.
Xu Tongtong, Sun Huazhi, Ma Chunmei, Jiang Lifen, Liu Yichen. Few-shot text classification based on Bi-directional Long-term Attention feature expression Few-shot text classification based on Bi-directional Long-term Attention feature expression . Data Analysis and Knowledge Discovery, 0, (): 1-.
链接本文:  
http://manu44.magtech.com.cn/Jwk_infotech_wk3/CN/10.11925/infotech.2096-3467.2018.2020.0206      或      http://manu44.magtech.com.cn/Jwk_infotech_wk3/CN/Y0/V/I/1
[1] 黄露,周恩国,李岱峰. 融合特定任务信息注意力机制的文本表示学习模型*[J]. 数据分析与知识发现, 2020, 4(9): 111-122.
[2] 尹浩然,曹金璇,曹鲁喆,王国栋. 扩充语义维度的BiGRU-AM突发事件要素识别研究*[J]. 数据分析与知识发现, 2020, 4(9): 91-99.
[3] 石磊,王毅,成颖,魏瑞斌. 自然语言处理中的注意力机制研究综述*[J]. 数据分析与知识发现, 2020, 4(5): 1-14.
[4] 薛福亮,刘丽芳. 一种基于CRF与ATAE-LSTM的细粒度情感分析方法*[J]. 数据分析与知识发现, 2020, 4(2/3): 207-213.
[5] 徐彤彤,孙华志,马春梅,姜丽芬,刘逸琛. 基于双向长效注意力特征表达的少样本文本分类模型研究*[J]. 数据分析与知识发现, 2020, 4(10): 113-123.
[6] 吴粤敏,丁港归,胡滨. 基于注意力机制的农业金融文本关系抽取研究*[J]. 数据分析与知识发现, 2019, 3(5): 86-92.
[7] 陶志勇,李小兵,刘影,刘晓芳. 基于双向长短时记忆网络的改进注意力短文本分类方法 *[J]. 数据分析与知识发现, 2019, 3(12): 21-29.
[8] 李钰曼,陈志泊,许福. 基于KACC模型的文本分类研究 *[J]. 数据分析与知识发现, 2019, 3(10): 89-97.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
版权所有 © 2015 《数据分析与知识发现》编辑部
地址:北京市海淀区中关村北四环西路33号 邮编:100190
电话/传真:(010)82626611-6626,82624938
E-mail:jishu@mail.las.ac.cn