情感可解释性分析是近年来比较新颖的研究方向,其目标是在预测文本的情感极性的同时给出决定情感极性的证据片段。该文在仅有情感分类任务数据集的基础上,提出了基于擦除的情感可解释性片段抽取方法,通过被擦除单词对情感极性逻辑判断的波动影响来决定证据的抽取。随后,利用擦除的方法使用模型对公开情感分析数据集中的部分数据进行片段抽取并人工过滤得到有监督数据,再使用T5序列生成式模型进行有监督训练,从而进一步提升证据抽取的性能。最终在“百度2022语言与智能技术竞赛: 情感可解释评测”中获得第三名的成绩。
Abstract
Related studies pay more and more attentions to sentiment interpretability analysis, which aims to predict the text's emotional polarity and the evidence segments that determine the emotional polarity. Based on the dataset of sentiment classification tasks, this study proposes an erasure based method for extracting evidence segments, which analysizes the logits influence of masked words on the emotional polarity. Subsequently, the model is used to extract the evidence of partial data in the public sentiment analysis dataset and the supervised data is manually filtered form the auto-extracted dataset. To further improve the performance of evidence extraction, this study fine-tunes the T5 sequence generative model on supervised data. This study won the third place in the Baidu 2022 Language and Intelligent Technology Competition: Sentimental Interpretable Assessment.
关键词
情感可解释性 /
基于擦除 /
序列生成式模型
{{custom_keyword}} /
Key words
sentiment interpretability /
erasure-based /
sequence generative model
{{custom_keyword}} /
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 贾闻俊,张晖,杨春明,等.面向产品属性的用户情感 模型[J].计算机应用,2016,36(1):175-180.
[2] 刘新星,姬东鸿,任亚峰.基于神经网络模型的产品属性情感分析[J].计算机应用,2017,(06):221-226.
[3] PONTIKI M, GALANIS D, PAPAGEORGIOU H, et al. Semeval task 4: Aspect based sentiment analysis[C]//Proceedings of the 8th International Workshop on Semantic Evaluation, 2014: 27-35.
[4] PONTIKI M, GALANIS D, PAPAGEORGIOU H, et al. Semeval task 12: Aspect based sentiment analysis[C]//Proceedings of the 9th International Workshop on Semantic Evaluation, 2015: 486-495.
[5] PONTIKI M, GALANIS D, PAPAGEORGIOU H, et al. Semeval task 5: Aspect based sentiment analysis[C]//Proceedings of the International Workshop on Semantic Evaluation, 2016: 19-30.
[6] SAEIDI M, BOUCHARD G, LIAKATA M, et al. Sentihood: Targeted aspect based sentiment analysis dataset for urban neighbourhoods[J]. arXiv preprint arXiv:1610.03771, 2016.
[7] LIU P, YUAN W, FU J, et al. Pre-train, prompt, and predict: A systematic survey of prompting methods in natural language processing[EB/OL] https://arxiv.org/pdf/2107.13586[2021-7-28].
[8] RAFFEL C, SHAZEER N, ROBERTS A, et al. Exploring the limits of transfer learning with a unified text-to-text transformer[J]. Journal of Machine Learning Research,2020, 21(140): 1-67.
[9] STROBELT H, GEHRMANN S, PFISTER H, et al. Lstmvis: A tool for visual analysis of hidden state dynamics in recurrent neural networks[J]. IEEE Transactions on Visualization and Computer Graphics, 2017, 24(1): 667-676.
[10] STROBELT H, GEHRMANN S, BEHRISCH M, et al. Seq2seq-vis: A visual debugging tool for sequence-to-sequence models[J]. IEEE Transactions on Visualization and Computer Graphics, 2018, 25(1): 353-363.
[11] VIG J. BertViz: A tool for visualizing multihead self-attention in the BERT model[C]//Proceedings of the ICLR Workshop: Debugging Machine Learning Models, 2019.
[12] BASTINGS J, FILIPPOVA K. The elephant in the interpretability room: Why use attention as explanation when we have saliency methods?[EB/OL]https://arxiv.org/pdf/2010.05607[2023-3-9].
[13] LI J, CHEN X, HOVY E, et al. Visualizing and understanding neural models in nlp[EB/OL]https://arxiv.org/pdf/1506.01066[2016-1-8].
[14] ARRAS L, MONTAVON G, MüLLER K R, et al. Explaining recurrent neural network predictions in sentiment analysis[EB/OL]https://arxiv.org/pdf/1706.07206[2017-6-22].
[15] LI J, MONROE W, JURAFSKY D. Understanding neural networks through representation erasure[EB/OL]https://arxiv/org/pdf/1612.08220[2016-12-24].
[16] WANG L, LIU H, PENG S, et al. DuTrust: A sentiment analysis dataset for trustworthiness evaluation[EB/OL]https://arxiv/org/pdf/2108.13140[2021-8-30].
[17] RIBEIRO M T, SINGH S, GUESTRIN C. "Why should i trust you?" Explaining the predictions of any classifier[C]//Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016: 1135-1144.
[18] ZHANG L, WANG S, LIU B. Deep learning for sentiment analysis: A survey[J]. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, 2018, 8(4): e1253.
[19] YANG Z, YANG D, DYER C, et al. Hierarchical attention networks for document classification[C]//Proceedings of the Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,2016: 1480-1489.
[20] XU J, CHEN D, QIU X, et al. Cached long short-term memory neural networks for document-level sentiment classification[EB/OL]https://arxiv/org/pdf/1610.04989[2016-10-17].
[21] QIAN Q, HUANG M, LEI J, et al. Linguistically regularized lstms for sentiment classification[EB/OL]https://arxiv/org/pdf/1611.03949[2016-11-13].
[22] YU J, JIANG J. Learning sentence embeddings with auxiliary tasks for cross-domain sentiment classification[C]//Proceedings of the Conference on Empirical Methods in Natural Language Processing, 2016: 236-246.
[23] TAY Y, TUAN L A, HUI S C. Dyadic memory networks for aspect-based sentiment analysis[C]//Proceedings of the ACM on Conference on Information and Knowledge Management, 2017: 107-116.
[24] CHEN P, SUN Z, BING L, et al. Recurrent attention network on memory for aspect sentiment analysis[C]//Proceedings of the Conference on Empirical Methods in Natural Language Processing, 2017: 452-461.
[25] LIU Y, OTT M, GOYAL N, et al. Roberta: A robustly optimized bert pretraining approach[EB/OL]https://arxiv/org/pdf/1907.11692[2019-7-26].
[26] WU L, LI J, WANG Y, et al. R-drop: Regularized dropout for neural networks[J]. Advances in Neural Information Processing Systems, 2021, 34: 10890-10905.
[27] MIYATO T, DAI A M, GOODFELLOW I. Adversarial training methods for semi-supervised text classification[EB/OL]https://arxiv/org/pdf/1605.07725[2016-5-25].
[28] WEI J, REN X, Li X, et al. Nezha: Neural contextualized representation for chinese language understanding[EB/OL]https://arxiv/org/pdf/1909.00204[2019-8-31].
[29] ZHANG Z, ZHANG H, CHEN K, et al. Mengzi: Towards lightweight yet ingenious pre-trained models for chinese[EB/OL]https://arxiv/org/pdf/2110.06696[2021-10-13].
[30] SU J. Wobert: Word-based Chinese bert model-zhuiyiai[R]. Technical Report, 2020.
{{custom_fnGroup.title_cn}}
脚注
{{custom_fn.content}}
基金
国家自然科学基金(61976148)
{{custom_fund}}
PDF(1218 KB)
