Research On The Construction And Application Of Event-Oriented Text Representation Model

Text representation is not only the basic problem of natural language research,but also the key to help us deal with text information efficiently.Most of the traditional text representation models are based on words or phrases,which is easy to lose semantic information.In recent years,the concept of event has attracted great attention of scholars.Narrative texts contain a large number of events.Event oriented text representation model is studied.Taking events or event elements as semantic units,the deep semantic information in texts can be well preserved.In this paper,firstly,the default object elements,time elements and location elements in text are studied;secondly,the event oriented text representation model event semantic network is constructed by taking event as basic semantic unit and combining with the idea of graph structure model;finally,the application of automatic summarization is realized based on event semantic network.The main content of this paper is divided into the following three parts:(1)Default fill up for event elements.In order to make the narration more coherent and fluent,the default of object elements,time elements and place elements often appear in a narrative text.It is an important problem to make up the missing event elements.This paper mainly studies the default time elements,environment elements(location elements)and object elements.The time and place elements are the same attributes of events,and the method of combining the context structure rules with the same event trigger vocabulary is used to complete the content;the object elements are complemented by the combination of the context structure rules and the common occurrence rules of events in the sentence,and the semantic association of events with multiple object elements in the text is calculated.The specific steps of the complement and the corresponding examples are given,and the experimental results of the default complement of event elements are given.(2)Event oriented text representation model event semantic network.In this paper,the event is taken as the research object,and the trigger word(action element)in the event is used to replace the event as the feature item.At the same time,combined with the idea of graph structure model,the semantic association between events is deeply mined:firstly,according to the characteristics of different event elements,different methods are used to calculate the association degree between event elements,and then different weights are given to different event elements,and the correlation degree between events is calculated by using the association degree of event elements through weighted formula,and according to the event correlation degree Finally,a new text representation model event semantic network is constructed,which can preserve the structure and semantic information of the text.(3)Automatic summarization based on event semantic network.The weight of each feature item in the event semantic network is calculated to get the weight of each node;then the weight of each node is calculated by the improved PageRank method to get the importance of each node in the text;then the weight of the corresponding event sentence in the text is obtained by the updated node importance,and the sentences are arranged in descending order according to the weight,The most important part of the sentence is selected as the summary sentence of the text;finally,the redundant sentences are output as the summary of the text according to the order in which they appear in the original text.Experiments show that the automatic summarization method based on event semantic network achieves good results.Figure[9]table[11]reference[61]...