To tackle issues including short length, strong technical specificity and challenges in intelligent reuse of signal equipment fault text data, a signal equipment fault text clustering method based on improved Biterm Topic Model and Word Vector Fusion (IBTM-TMW) is proposed. Firstly, to reduce noise of the data and improve data quality, a customized dictionary and gerund processing are introduced in the process of data preprocessing. Secondly, during the Gibbs sampling modeling process of word pairs, the differential importance of words is introduced as a weighting factor, and the Improved Biterm Topic Model (IBTM) is used to improve the learning capability of text topic features. The weight of Term Frequency-Modified Inverse Document Frequency (TF-MIDF) is embedded into the generation process of Word2vec word vectors. The text importance of words is integrated into the Word2vec word vector to refine the feature vector representation of text words. Finally, the text topic feature vector and the word feature vector are integrated to enhance the text feature representation capability. On this basis, the K-means ++ algorithm is used for fault cluster analysis. The results show that within the same data set, the quality of the text feature vector generated by IBTM-TMW model is significantly higher than those of LDA and Label-LDA models, and its diagnostic accuracy of Correct Classification Rate (CCR) reaches 89.9% (surpassing the 78.3%, 68.1%, 87.9% and 81.7% accuracies of K-means, GMM, AGNES and BIRCH, respectively). The proposed method improves the capability of analyzing the correlation between fault text features and their categories, thereby offering a valuable reference for text-data-driven fault diagnosis.
词向量是1种文本特征表示方法,用于将语料库中的词转换为向量形式。选择Word2vec模型来学习分布式词向量,该模型在生成词向量方面具有高效性。Word2vec 模型包含2种神经网络结构:连续词袋(Continuous Bag of Words,CBOW)模型和 跳字模型(Skip-Gram)[16],由于CBOW相较于Skip-Gram具有更快的训练速度[17],因此,选用CBOW模型进行词向量的训练,其模型结构如图3所示。
MAJianjun, LIPing, MAXiaoning, et al. Research on the Overall Framework and Key Technologies of Railway Integrated Information Platform [J]. China Railway Science, 2020, 41 (5): 153-161. in Chinese
YANGLianbao, LIPing, XUERui, et al. Intelligent Classification of Faults of Railway Signal Equipment Based on Imbalanced Text Data Mining [J]. Journal of the China Railway Society, 2018, 40 (2): 59-66. in Chinese
[5]
WANGF, XUT H, TANGT, et al. Bilevel Feature Extraction-Based Text Mining for Fault Diagnosis of Railway Systems [J]. IEEE Transactions on Intelligent Transportation Systems, 2017, 18 (1): 49-58.
LINHaixiang, BAIWansheng, LURenjie, et al. Named Entity Recognition of Fault Information of High-Speed Railway Turnout from BMBC Model [J]. Journal of Railway Science and Engineering, 2023, 20 (4): 1149-1159. in Chinese
ZHOULujie, DANGJianwu, WANGYuxin, et al. Fault Diagnosis for On-Board Equipment of Train Control System Based on CNN-CSRF Hybrid Model [J]. Journal of the China Railway Society, 2020, 42 (11): 94-101. in Chinese
LURenjie, LINHaixiang, XULi, et al. Fault Diagnosis Based On-Board Equipment of Train Control System CNN and PSO-SVM Hybrid Model [J]. Journal of Measurement Science and Instrumentation, 2022, 13 (4): 430-438. in Chinese
[12]
YANX H, GUOJ F, LANY Y, et al. A Biterm Topic Model for Short Texts [C]// Proceedings of the 22nd International Conference on World Wide Web. Rio de Janeiro, Brazil. New York: ACM, 2013: 1445-1456.
[13]
LIW J, FENGY M, LID J, et al. Micro-Blog Topic Detection Method Based on BTM Topic Model and K-Means Clustering Algorithm [J]. Automatic Control and Computer Sciences, 2016, 50 (4): 271-277.
[14]
LIX M, ZHANGA, LIC C, et al. Relational Biterm Topic Model: Short-Text Topic Modeling Using Word Embeddings [J]. The Computer Journal, 2019, 62 (3): 359-372.
[15]
ZHANGK, ZHOUY, CHENZ, et al. Incorporating Biterm Correlation Knowledge into Topic Modeling for Short Texts [J]. The Computer Journal, 2022, 65 (3): 537-553.
[16]
BLEID M, NGA Y, JORDANM I. Latent Dirichlet Allocation [J]. Journal of Machine Learning Research, 2003, 3: 993-1022.
[17]
CHENGX Q, YANX H, LANY Y, et al. BTM: Topic Modeling over Short Texts [J]. IEEE Transactions on Knowledge & Data Engineering, 2014, 26 (12): 2928-2941.
[18]
YINJ H, WANGJ Y. A Dirichlet Multinomial Mixture Model-Based Approach for Short Text Clustering [C]// Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York, USA. New York: ACM, 2014: 233-242.
[19]
ZHANGY, TSAIF S. Chinese Novelty Mining [C]// Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing. Singapore. Morristown: Association for Computational Linguistics, 2009: 1561-1570.
[20]
马治涛.文本分类停用词处理和特征选择技术研究[D].西安:西安电子科技大学,2014.
[21]
MAZhitao. Research on Stop Words and Feature Selection for Text Classification [D]. Xi'an: Xidian University, 2014. in Chinese
[22]
MIKOLOVT, SUTSKEVERI, CHENK, et al. Distributed Representations of Words and Phrases and Their Compositionality [EB/OL]. 2013: arXiv: 1310.4546.
[23]
MIKOLOVT, CHENK, CORRADOG, et al. Efficient Estimation of Word Representations in Vector Space ArXiv e-Prints, 2013: arXiv:
[24]
SABBAHT, SELAMATA, SELAMATM H, et al. Modified Frequency-Based Term Weighting Schemes for Text Classification [J]. Applied Soft Computing, 2017, 58: 193-206.
YANDuanwu, MEIXirui, YANGXiongfei, et al. Research on Microblog Text Topic Clustering Based on the Fusion of Topic Model and Word Embedding [J]. Journal of Modern Information, 2021, 41 (10): 67-74. in Chinese
[27]
YANGN, ZHANGY P. Railway Fault Text Clustering Method Using an Improved Dirichlet Multinomial Mixture Model [J]. Mathematical Problems in Engineering, 2022, 2022: 7882396.
[28]
FANY Q, LIUJ H, YEH, et al. TA-LSTM: a Time and Attribute Aware LSTM for Deep Flight Track Clustering [J]. IEEE Transactions on Aerospace and Electronic Systems, 2023, 59 (5): 7047-7060.