Research And Implementation Of ICT Oriented Fault Knowledge Extraction And Root Cause Analysis System

Nowadays,information and communication technology has become an important part of people’s lives,the importance of handling related faults is self-evident.When the hardware and software related to communication domain are faulty,it is usually rather difficult for relevant personnel to identify the root cause of faults and perform fault recovery in a short period of time.Issues above makes locating faults automatically and quickly a major difficulty in the field of operation and maintenance.Traditional methods of root cause analysis often do not apply to unstructured fault data adequately and ignores the causal propagation relationship between potential faults in the text,so they usually fail to seek deeper errors.Considering that failures are usually driven by events such as software and hardware failures or related processing procedure,this paper deeply studies the knowledge extraction algorithm of events of ICT domain failures and constructs event-centered fault propagation graphs on related fault data by event extraction and event relationship extraction in order to in implement root cause analysis of faults.The main work in this paper is as follows:1)Event extraction algorithm and event relationship extraction algorithm are implemented on semi-structured data of faults in the domain.Among the task of event extraction we combined pre-trained model,temporal convolutional network and cascade structure,captured dependency information in text effectively and improved performance of label decoding.Conditional layer normalization is used to enhance the feature of event trigger in the task of role extraction,which improved the performance of role extraction.Among the task of event relationship extraction,the event type information is enhanced by adding special tags for different events during the preprocessing of the pre-trained model,we also used temporal convolutional network to capture sequence features among texts.The performance of our model is measured with previous extraction algorithms to ensure its advantages.2)Based on the knowledge extraction algorithms above,design and implement a root cause analysis system.By constructing a fault graph on related data,the fault propagation path was visually described and analyzed in a more reasonable way.We provide system functions such as the management of data,training,tuning and testing of relevant algorithm models,automated construction,maintenance and storage of fault graphs,and also the root cause of failure retrieval.Our system can help users locate problematic resource objects quickly among a large number of fault events,and can automatically construct graphs on given data,hence it has both scalability and practicality.