Research On Quality Evaluation And Revision For Emergency Plan Response Texts Based On Process Model Extraction

Emergency plans are used as effective instructions for emergency responses.Each topic of emergency plans deals with a specific type of emergencies.For a specific topic,emergency plans correspond to four administrative levels,i.e.national,provincial,municipal,and district levels.Expert review and revision for emergency plans is a time-consuming and labor-intensive task.Existing approaches have mostly established indicator systems through statistical analysis of historical data and produced subjective expert evaluation results.In addition,psychological factors,knowledge experience and decision-making level of review experts make it difficult to quantify indicators and perform evaluation operations.Until now,there is also a lack of theoretical evaluation about text quality of emergency plans before emergencies.The emergency plan response text describes how to performe an emergency response process,which involves response departments,task descriptions and task relation.Emergency response process model extraction automatically achieve process models from emergency plan response texts.The generated emergency response process model serves as a reference basis for quality evaluation and revision of the emergency plan response text.Therefore,it is of great significance to automatically evaluate and revise emergency plan response texts from a perspective of process description.The dissertation regards an emergency response process model as a bridge to address quality evaluation and revision for emergency plan response text by process model extraction and text similarity computing,which provides technical support for expert review and preparation.The main contributions of this dissertation are as follows:(1)A rule-based emergency response process model extraction approach is proposed to deal with the transformation from unstructured process description text to structured process models.First,an emergency plan response text is represented as a text tree according to layout markups and sentence-sequential relation.Second,according to syntax parsing and feature words,a series of rules are created to identify four-level response condition formulas,executors,task descriptions,and flow relation.Third,an emergency response process tree is generated from both the response text tree and extracted process elements,and then is transformed to an emergency response process BPMN model.Finally,a case study and extensive experiments on real-world data demonstrate that the extracted emergency response process models are highly consistent with both the process models extracted manually and the original response texts,and the proposed approach is capable of assisting domain experts in modeling emergency response processes.(2)A hybrid approach is proposed to automatically generate emergency response process task-view and organization-view models to meet different requirements.First,process elements regarded as continuous word sequences are identified by Bi-LSTM-CRF networks.Second,response tasks and task relation are generated based on feature words and position features,and then are used for generating emergency response process task-view models.Third,emergency response process organization-view models are generated from extracted response tasks and task relation.Finally,a case study and systematic experiments on real-world data illustrate that the proposed approach can be used to assist different participants in understanding emergency response processes from different perspectives and provides decision support for emergency responses.(3)A quality evaluation approach is proposed based on process model extraction,which can automatically analyze emergency plan response texts of different levels from a perspective of process description.First,quality elements for emergency plan response text are analyzed from a perspective of process description.Second,a framework for emergency plan response text quality evaluation is given based on extracted emergency response process multi-view models.Third,a series of quality evaluation rules are created and these rules focus on progressive and correlation relationship among four levels of emergency responses,completeness of response tasks,equilibrium of task allocation,rationality of department interaction,ambiguity and redundancy of response text,and cohesion of emergency plans.Finally,experiment analysis on national,provincial,municipal,and district emergency plan response texts,and qualitative and quantitative evaluation results demonstrate that the proposed approach is effective for reviewing and improving emergency plan response texts.(4)Similarity measure and quality revision approaches for emergency plan response text are proposed based on emergency response process model extraction and vector representation.First,vector representation about process elements,response tasks,response sub-processes and emergency response process are given,respectively.Second,process description semantic similarity of emergency plan response texts is calculated by generated emergency response process vectors.Third,combining the extracted emergency response process model,vector representation and similarity computing results,the emergency plan response text with quality problems is automatically revised.Finally,a real-world data set about three topic and three levels of emergency plans is collected for experimental evaluation,and the results illustrate that the proposed text similarity computing approach performs well in distinguishing topics and levels of emergency plan response text.In addition,a case study of emergency plan response text quality revision illustrates that the proposed approach is feasible.