Research On Semantic Coherence Analysis Model Of English Text

In recent years,with the rapid development of artificial intelligence,has brought unprecedented technological changes,which not only improvs people’s working style and life quality,but also increases the intelligent level of human-computer interaction services.Among these changes,automatic text generation and quality assessment and other research fields have attracted much attention,and many universities have also applied relevant research into teaching AIDS.Text coherence is one of the key indicators to evaluate the quality of English composition,but the existing automatic English composition correcting system often ignores this evaluation parameter.In addition,the online world is complex.By analyzing the semantic coherence of text can help screen out fake comments and detect intrusive statements in spam,which can help to create a more secure online environment.Therefore,the demand for analyzing text coherence is becoming more and more widespread.This study considers the influence of several key elements,such as discourse relations and semantic understanding on semantic coherence research,and designs and implements an English text semantic coherence analysis model to meet the actual needs.Experiments show that the average absolute error between the correction results and manual correction results is 3.1921,and the Pearson correlation coefficient is 0.6729.Compared with the previous coherent analysis model,the model obtains a higher evaluation index and has better practical value.The main contents of this research are as follows:1.Word errors are one of the factors affecting the semantic coherence of the text.Adopting the real word confusion set combined with the N-element model to correct word errors,and analyzes the semantic coherence of the text after the text is pre-processed.In addition,discourse relations help to better understand semantic information and ensure the coherence and integrity of the text.Therefore,this study builds a discourse parser based on the Rhetorical Structure Theory,which analyzes the text structure and semantic information in clauses and between sentences,and extracts the discourse relation information.2.This study designs a Tree LSTM network model based on discourse analysis,inputting discourse relation information into the Tree LSTM network,using the memory unit and gating mechanism of the network to acquire semantic structure features for a long time and a long distance,focusing on the core semantic unit and analyzing the interrelation between satellite semantic unit and its semantic content.3.In order to better detect the global coherence of text,a hierarchical network model based on attention mechanism is designed.In the three levels of sentence,paragraph and document,Bi-LSTM network is used to extract the global dependent information of text,and attention mechanism is added to enable it to select important features related to semantic coherence adaptively.Finally,the semantic coherence analysis model of English text is generated by combining the Tree LSTM network based on discourse parsing with the hierarchical network based on attention mechanism.