A Research On Script Event Prediction Bases On Deep Learning

Understanding the events described in the text is critical for many artificial intelligence(AI)applications,such as utterance understanding,intent recognition,and dialog generation.Among them,script event prediction is the most challenging task in this work.Script event prediction can also be called script event prediction,which is a subtask from a manual coding task.The manual coding task originated in 1970-80 and was first used as a textual knowledge backbone.It makes it possible to reason and other NLP tasks that need to use deep semantic knowledge information in the text,such as story generation,dialogue understanding,text speculation and other tasks.Provides the foundation.The main work of completing the script event prediction can be divided into three parts:the first part is to extract the script event from the text according to the rules;the second step is to change the script event into a computer-understandable expression;the third part is to specify the script.Event prediction models and assessment methods.In terms of refinement into technology,the main steps of completing the script event prediction task are divided into seven parts:data cleaning,dependency analysis,entity analysis,extraction of events,acquisition of event chains,construction of prediction models,and evaluation of models.Because the quality and quality of the construction of the prediction model determines whether the whole process can fully grasp the semantic information of the events in the text,the focus of this paper is on the sixth part of the construction of the prediction model and the seventh part of the evaluation model.The main content of this paper is to carry out more in-depth work on the event expression pre-processing event in the script event prediction task.The main contribution is the vector representation processing of the event.For the two-language problem of the meaning of the polysemy in the event tuple and the inability to contain the dense information between the events,a vector form for generating script events using context events is proposed.There is a certain increase in accuracy.Finally,based on the work of the predecessors,the accuracy of the model is compared.The accuracy of the model is compared with the PMI model,the Bigram model,and the long-short memory network +Attention mechanism model.In addition,a horizontal comparison experiment was performed to verify that the number of elements contained in the event tuple is not important for the experimental results.In summary,according to the experiment,it is found that the final result of the experiment is the verb or verb phrase in the event,and the dependent elements existing in the event have less influence on the result.In addition,deeper event expression vectors replace event vectors in neural networks that process time series,improving the accuracy of the results.