Topic Discovery Research Of Accident Investigation Report Based On LDA

In recent years,China’s major coal mine accidents occur frequently,resulting in a large number of deaths and injuries,also with large economic losses and extremely bad impacts,which have aroused the close attention of the general public.After the accident,the relevant departments led the formation of an investigation team,and invited relevant experts to investigate the accident and issue a professional accident investigation report.The report contains a large number of information about the hidden dangers of the accident and suggestions for corrective actions.Based on the LDA latent topic model,this paper analyses 44 major coal mine accidents investigation reports from 2012 to 2018,which published in the State Administration of Coal Mine Safety Supervision.Firstly,the Latent Dirichlet model is introduced,and its difference from the traditional text mining method is explained.Then the mining process is explained: data cleaning,word segmentation,filtering stop words,and establishing a “Document-Topic”matrix.The basic information of the accident report is described by extracting the keywords of the high-frequency vocabulary and the probability of the inverse document,and the visual display is carried out.As for the hidden information of the accident report is extracted by the Latent Dirichlet topic model.There are three topics of the accident investigation report mainly contains,which are hidden danger topic,punishment topic and the corrective measures topic.After the machine learning,the hidden danger topic,which through the intensity comparison,can analyze which hidden subject needs special attention of the coal miner,managers and supervisors.It also can tell us which hidden dangers should be solved first when eliminating hidden dangers;For the punishment topic,the punishment information of the relevant responsible personnel includes the records of unsafe behavior in the investigation report.It’s actually a supplement to the hidden danger topic that analyze the punishment information;For the topic of rectification measures,the key point is that the coal mine should improve the management system and strengthen the management.The supervision department should implement laws and regulations,as well as strictly control the resumption of production of coal mining enterprises.Finally,the above three topics arevisually displayed,which can facilitate a more intuitive way to understand each topic.The LDA topic model essentially applies the machine learning and probabilistic models to the survey report.For the analysis unit of the article “words”,the subject is extracted and analyzed in a semi-autonomous and unsupervised manner.Faced with a large amount of data,it can provide relevant personnel with a new way of thinking about the investigation of coal mines,finding key hidden dangers and improving the safety of coal mines.