Research On Event Reasoning Method Oriented To News Text

Event reasoning research is to infer subsequent events that may occur through a series of event context sequences.It is one of the research hotspots in the direction of event extraction and analysis.The research object of this thesis is a collection of events with a certain logical relationship.The research goal is to select the most likely subsequent event from candidate events based on the prior information of the logical relationship between the events.News is an important carrier of information dissemination,which contains a large number of objective and true events.Extracting events from news texts and performing inference analysis helps to quickly and accurately obtain the development context and related information between related events.To this end,this thesis proposes an event reasoning method oriented to news texts.This method extracts and clusters largescale news texts through event chains to form clusters of narrative chains with similar themes and scenarios.The event graph of the relationship completes the event reasoning task.The main work of this thesis is as follows:(1)Aiming at the problem of the single relationship between event sequences in traditional narrative event research,this thesis proposes an event chain extraction method that integrates event causality.This method uses explicit causal connection rules to identify causal event sentence pairs from the text.,And then based on natural language processing techniques such as dependency parsing,part-of-speech tagging,and reference resolution to identify event constituent elements.Taking the causal event pair as the supplementary information of the co-protagonist's narrative chain,enriches the connection relationship between events,and is the basic work of building a multirelational event evolution map.(2)Aiming at the problems of unsupervised event extraction methods such as large noise interference and failure to make full use of the correlation between event sequences,this thesis proposes a method based on narrative event chain clustering to process narrative events.The different representations of narrative chains cluster large-scale narrative chains to obtain narrative chain clusters with similar scenarios and themes.Subsequent experiments show that this method can effectively improve the accuracy of event prediction tasks.(3)Aiming at the problem that the existing event feature learning methods fail to fully explore the differential impact of different events in the event sequence on downstream events,this thesis proposes a gated graph neural network model(GGNNMHA),The model fully considers the impact of each event in the event sequence on the event to be predicted and the evolution characteristics between events.Compared with the traditional event feature learning model based on statistical probability models and recurrent neural networks,this multi-head attention mechanism The graph neural network model has a better effect in event schema learning.