Process-extraction-based text similarity measure for emergency response plans

Abstract Measuring text similarity is an important challenge for Chinese emergency response plans, which describe complex response processes involving four level responses and collaborative interactions among multiple departments and roles. Most existing studies focus on lexical semantic features and their results are not applicable to Chinese emergency response plans. In order to reflect process semantics of Chinese emergency response plans, a novel text similarity measure is proposed. First, an emergency response process including four sub-processes is extracted from an emergency response plan, and each emergency response sub-process is represented by three types of response tasks, i.e. regular task, message sending task and message receiving task. Next, vector representations of process elements including task statements, roles and departments are generated. Then, response task vectors that are represented by a combination of process element vectors are generated and three response task set vectors are used for emergency response sub-process representation. Finally, the similarity between sub-processes is computed as the average cosine similarity of three pairs of response task set vectors and the text similarity is the average similarity of four sub-processes. A real-world data set is collected for experimental evaluation and the results illustrate that the proposed approach performs well in distinguishing topics and levels of emergency response plans.