基于Bi-LSTM和TFIDF的工单事件提取

Event Extraction of Power Customer Service Order Based on BiLSTM-CRF and TFIDF

电网工单数据是电网运行情况以及客户满意程度的主要信息来源,近年来,有学者将深度学习的方法应用于工单数据的关键信息提取,但是提取出的关键词、词还不足以完整描述工单反映的具体情况。本文提出了一种事件抽取模型,先通过一定的方式进行文本预处理,确定标签体系和特征模板,再用Bi-LSTM和CRF相结合的模型进行实体识别和标注,最后通过TFIDF模型提取出事件表达,将该模型用于电网工单数据的事件抽取,用准确率、召回率和F1得分作为模型的评价标准,证明了该模型在工单数据分析中的可用性。

Power Customer Service Order data is the main source of information on grid operation and customer satisfaction.In recent years,some scholars have applied the method of deep learning to the key information extraction of work order data.However,the extracted keywords and words are not enough to fully describe the specific situation reflected by the work order.This paper proposes an event extraction model,which firstly performs text preprocessing in a certain way,determines the label system and feature template,and then uses Bi-LSTM and CRF model for entity recognition.Finally,the event expression is extracted by TFIDF model.This paper uses the model for the event extraction of the power customer service order.This paper chooses the accuracy,recall rate and F1 score as the evaluation criteria of the model,and prove the availability of the model in the analysis of work order data.