Research On The Construction Method Of The Cause-and-effect Graph Of Power Complaint Event

Electric power industry is an important part related to the national economy and people’s livelihood.Power complaint,as a crucial channel for electric power departments to communicate with users and find problems,have attracted much attention of relevant departments.Event evolutionary graph can dig out the logical relationship of power complaint events,help the power department acquire the knowledge and experience of the causal and sequential relationship,and find the critical paths and nodes.So as to provide decision support to the power departments and improve the power service and stability,and ultimately improve people’s living standards.However,there are still many problems to be solved in the the process of constructing the event evolutionary graph of power complaint due to the related research is rare in this field and the strong field characteristics.Based on the text characteristics of the power complaint from 12398 energy regulatory hotline,this paper is organized by the key technologies of constructing the event evolutionary graph like event classification,event relation extraction and event generalization,and finally builds the event evolutionary graph of power complaints for different types.The research contents and achievements of this paper are as follows:(1)In response to the characteristics of small sample,multi-label,unbalanced sample and domain characteristics of power complaint text,this paper proposes keyword-based non-monitoring simple Bayesian text classification methods and extracts text keywords to update the initial keywords bank based on the algorithm.The classification accuracy increased by 2.4%.In addition,compared with the supervised text classification method based on the improved Text RNN model,the accuracy P,recall rate R and F1 value indexes of macro average are significantly better,indicating that the problem of small and unbalanced samples is better solved.In addition,the unsupervised text classification takes less time and does not require a lot of manual annotation,which saves human and material resources.(2)The method of relationship extraction based on pattern matching is adopted for event relation extraction.Firstly,based on the existing explicit causal syntax model research and the text characteristics of power complaint text,a causal relationship extraction pattern suitable for the power complaint text is summarized.Then,an innovative extraction pattern of power complaint sequential relationship is proposed.Finally,events are extracted by using dependency parsing and semantic role labeling methods.The experimental result shows that the extraction accuracy of causal events is 93.25%.(3)For event generalization,Word2 vec vector is used to improve the agglomerative hierarchical clustering algorithm to cluster the short texts.The experimental result shows that the algorithm is accurate and suitable for short texts.Then the Newton cooling law which can fusing the time factor is introduced to optimize the event transfer probability algorithm.Finally,taking the event evolutionary graph of electricity safety events as an example and introducing the quality management theory to analyze the event evolutionary graph,then putting forward the corresponding suggestions on power optimization management.