Research And Application Of Event Extraction Based On Deep Learning And Domain Knowledge

In the context of information explosion,how to quickly obtain the required knowledge from massive information has become one of the important issues that need to be solved urgently in today's society.Event extraction technology can effectively improve the ability of human information retrieval.At present,with the continuous development of deep learning models,event extraction technology has been widely used in various fields such as finance and music.However,in the field of instrumentation,which is one of the strategic industries of the national economy,event extraction technology has not been popularized.At the same time,the event extraction technology has the phenomenon of incomplete extraction of information.After analyzing the research status of event extraction and event relations,a joint event extraction model based on deep learning was established and implemented,and an event analysis prototype system was designed on this basis.Finally,it was combined with specific cases and applied to relevant scenarios in the instrument field.It mainly includes event extraction and event relationship analysis modules.The event extraction module is based on the BERT+CRF model and extracts events through the method of sequence annotation.In order to adapt to the professional field,a small-scale instrumentation field event extraction corpus is established,and at the same time,combined with the domain knowledge base to supplement the knowledge of the instrument and equipment in the event.In the event relationship analysis module,in order to extract the methods and methods involved in the event and the causal relationship between the events,a keyword knowledge base was designed and established;in order to solve the problem of inconsistent expression of time in the event,the relative concept of using time was proposed to restore Time to date.In order to improve the accuracy of users' understanding of similar events related to time series,a method of summarizing events using the concept dependency tree is proposed.And from the concept of the year from the abstract to the concrete process,the subordinate tree of the time concept of the year was established.The event extraction module extracts domain-related events in the text,and the domain knowledge base expands the information of the equipment in the event.The event relationship analysis module uses the keyword knowledge base to extract the causal relationship between domain events in the form of event pairs through the location information between keywords and event trigger words.After the event information is expanded and integrated through the event model,the joint model improves the integrity of the extracted event information and solves the problem of incomplete event extraction information to a certain extent.Using the corresponding relationship between the proportion of event weight and the description language,by calculating the proportion of event weight,an accurate description vocabulary can be found.The effect of event induction can reflect the actual situation and improve the accuracy of understanding.The research results are applied to the field of instrumentation,and the development of an event analysis prototype system based on deep learning and domain knowledge has been completed.In the prototype system,application examples in three scenarios,event extraction,instrument equipment failure analysis,and instrument industry competition analysis are given to demonstrate the effect of the system implementation and verify the feasibility of the system.