-
摘要: 提出基于雙向長短期記憶網絡（bidirectional long short-term memory，BiLSTM）和前向神經網絡的融合模型完成公共安全事件的觸發詞識別任務.首先通過BiLSTM提取整段文本的高層語義特征，避免了以往機器學習方法需要人工提取特征的問題，其次采用特征拼接并在前向神經網絡中識別并分類事件觸發詞.實驗結果表明相較于基準模型，本文方法在中文突發事件語料庫（Chinese emergency corpus，CEC）上取得了更為突出的性能，Micro-F1值為78.47%.此外本文討論了不同拼接特征在觸發詞識別任務中的重要性，對文本分析中3類特征（詞性、句法、實體）的重要程度進行了比較和分析，得出句法特征對于事件觸發詞識別任務助益最大的結論.Abstract: As the internet coverage continues to expand, obtaining valuable information from a large amount of fragmented semi-structured text data has become a huge challenge considering the vast amount of social public information. Event trigger identification technology can effectively mine and refine text information so that the users can quickly and accurately get what they need; thus, it has gradually become an active research area in the field of natural language processing. An event trigger word is generally a word or phrase that marks the occurrence of the event, then trigger word identification has been applied to many aspects and plays an important role in the fields of knowledge base construction, intelligent search engine, automatic question answering robot, and automatic summarization. However, the text data are characterized by high dimensionality and ambiguity. The existing identification methods are mostly based on manual complex feature engineering or only consider the features in a certain text window. In this process, manual analysis and selection of a large number of features are required. Considerable reliance on natural language processing tools leads to the inability of applying the model on a large scale, and there are problems of erroneous cascade communication and complicated feature engineering. This paper proposed a fusion model based on the bidirectional long short-term memory (BiLSTM) and feed-forward neural networks to complete the trigger identification task for public security events. First, the high-level features of the entire text were extracted through BiLSTM to avoid manual feature extraction, which was associated with the existing machine learning methods. Then, contacted features were used to input feed-forward neural networks and identify event triggers. The experimental results show that the proposed method achieves good performance in the Chinese emergency corpus, CEC, and the Micro-F1 is 78.47%. In addition, the importance of different contacted features was also discussed in trigger word recognition tasks, and the importance of three types of features, namely part of speech, syntax, and entity, in text analysis was analyzed. It is concluded that syntactic features are most helpful to the task of event-trigger word recognition.
-
表 1 實驗環境所需軟件及下載地址
Table 1. Required software and download link
軟件 版本號 地址 TensorFlow 1.10 www.tensorflow.org Stanford CoreNLP 3.9.2 stanfordnlp.github.io/CoreNLP/ LTP 3.4.0 ltp.ai libSVM 3.23 www.csie.ntu.edu.tw/~cjlin/libsvm/ 
下載: 導出CSV
表 2 CEC事件語料庫及常見觸發詞
Table 2. Statistics of CEC and common trigger words
類型 篇數 句子 事件 常見觸發詞 地震 45 292 682 震級、震源、震感 交通事故 49 265 798 酒后駕駛、相撞、碰撞、追尾 恐怖襲擊 30 273 456 恐怖組織、自殺式襲擊、劫持 食物中毒 45 288 701 食物中毒、腐爛、嘔吐、腹瀉 火災 31 260 496 火災、燒毀、濃煙、燃燒
下載: 導出CSV
表 4 特征性能比較
Table 4. Comparison of features performance
序號 特征 Micro-F1/% 1 Ew，g 76.61 2 Ew+Eo，g(詞性標簽) 77.59 3 Ew+Eo，g(實體類型) 78.18 4 Ew+Eo，g(依存關系) 78.47
下載: 導出CSV
www.77susu.com<span id="fpn9h"><noframes id="fpn9h"><span id="fpn9h"></span> <span id="fpn9h"><noframes id="fpn9h"> <th id="fpn9h"></th> <strike id="fpn9h"><noframes id="fpn9h"><strike id="fpn9h"></strike> <th id="fpn9h"><noframes id="fpn9h"> <span id="fpn9h"><video id="fpn9h"></video></span> <ruby id="fpn9h"></ruby> <strike id="fpn9h"><noframes id="fpn9h"><span id="fpn9h"></span> -
參考文獻
[1] Bj?rne J. Biomedical Event Extraction with Machine Learning [Dissertation]. Turku: University of Turku, 2014 [2] Xuan X X, Liao T, Gao B B. Automatic extraction of Chinese event trigger word. Comput Digit Eng, 2015, 43(3): 457 doi: 10.3969/j.issn1672-9722.2015.03.026軒小星, 廖濤, 高貝貝. 中文事件觸發詞的自動抽取研究. 計算機與數字工程, 2015, 43(3): 457 doi: 10.3969/j.issn1672-9722.2015.03.026 [3] He X Y, Li L S. Trigger detection based on bidirectional LSTM and two-stage method. J Chin Inform Process, 2017, 31(6): 147 doi: 10.3969/j.issn.1003-0077.2017.06.020何馨宇, 李麗雙. 基于雙向LSTM和兩階段方法的觸發詞識別. 中文信息學報, 2017, 31(6): 147 doi: 10.3969/j.issn.1003-0077.2017.06.020 [4] Pyysalo S, Ohta T, Miwa M, et al. Event extraction across multiple levels of biological organization. Bioinformatics, 2012, 28(18): i575 doi: 10.1093/bioinformatics/bts407 [5] Zhou D Y, Zhong D Y, He Y L. Event trigger identification for biomedical events extraction using domain knowledge. Bioinformatics, 2014, 30(11): 1587 doi: 10.1093/bioinformatics/btu061 [6] Wei X M, Zhu Q, Lyu C, et al. A hybrid method to extract triggers in biomedical events. J Digit Inform Manage, 2015, 13(4): 298 http://smartsearch.nstl.gov.cn/paper_detail.html?id=1bf664c0d6c16db5dc030d6e4a7d251f [7] Lousteau-Cazalet C, Barakat A, Belaud J P, et al. A decision support system for eco-efficient biorefinery process comparison using a semantic approach. Comput Electron Agric, 2016, 127: 351 doi: 10.1016/j.compag.2016.06.020 [8] Chen Z Y, Huang Y, Wang Y, et al. Unsupervised method for event trigger identification and classification. Foreign Electron Meas Technol, 2016, 35(7): 91 doi: 10.3969/j.issn.1002-8978.2016.07.022陳自巖, 黃宇, 王洋, 等. 一種非監督的事件觸發詞檢測和分類方法. 國外電子測量技術, 2016, 35(7): 91 doi: 10.3969/j.issn.1002-8978.2016.07.022 [9] Wang Y, Wang J, Lin H F, et al. Bidirectional long short-term memory with CRF for detecting biomedical event trigger in FastText semantic space. BMC Bioinf, 2018, 19(Suppl 20): 507 http://www.ncbi.nlm.nih.gov/pubmed/30577839 [10] Amami M, Elkhlifi A, Faiz R. Biological event extraction using SVM and composite kernel function//The Extraction and Knowledge Management Conference. Bordeaux, 2012: 333(Amami M, Elkhlifi A, Faiz R. Biological event extraction using SVM and composite kernel function//Conférence Internationale Francophone sur l'Extraction et la Gestion des Connaissances. Bordeaux, 2012: 333 [11] Vanegas J A, Matos S, González F, et al. An overview of biomolecular event extraction from scientific documents. Comput Math Methods Med, 2015, 2015: 571381 http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4637451/ [12] Wang Y, Wang J, Lin H F, et al. Biomedical event trigger detection based on bidirectional LSTM and CRF//2017 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). Kansas City, 2017: 445 [13] Bengio Y, Ducharme R, Vincent P, et al. A neural probabilistic language model. J Mach Learn Res, 2003, 3: 1137 http://search.ebscohost.com/login.aspx?direct=true&db=aph&AN=11468885&site=ehost-live [14] Peters M E, Neumann M, Iyyer M, et al. Deep contextualized word representations//Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics. New Orleans, 2018: 2227 [15] Srivastava N, Hinton G, Krizhevsky A, et al. Dropout: a simple way to prevent neural networks from overfitting. J Mach Learn Res, 2014, 15(1): 1929 http://dl.acm.org/citation.cfm?id=2670313&preflayout=flat [16] Lai S W. Word and Document Embeddings Based on Neural Network Approaches [Dissertation]. Beijing: The University of Chinese Academy of Sciences, 2016來斯惟. 基于神經網絡的詞和文檔語義向量表示方法研究[學位論文]. 北京: 中國科學院大學, 2016 [17] Devlin J, Chang M W, Lee K, et al. Bert: Pre-training of deep bidirectional transformers for language understanding//Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics. Minneapolis, 2019 [18] Wang J, Li H L, An Y, et al. Biomedical event trigger detection based on convolutional neural network. Int J Data Min Bioinf, 2016, 15(3): 195 doi: 10.1504/IJDMB.2016.077067 -
點擊查看大圖
圖(3) / 表(4)
計量
- 文章訪問數: 1269
- HTML全文瀏覽量: 414
- PDF下載量: 49
- 被引次數: 0
