| With the development of the network era,the current network space is full of a large number of text data,and these data contains a lot of useful value,so it is particularly important to study these text data.Although there are many useful values in the text,there are a wide range of default and referential phenomena in the massive text,which makes the text lack of semantics at the overall level.Although these phenomena are not too difficult for humans to understand their meanings,they are very difficult and challenging for machines to understand text information.Therefore,it is particularly important to study the anaphora resolution in machine understanding,machine translation,information extraction and other tasks.This paper studies the anaphora resolution of personal pronouns in the aspect of anaphora.From the two aspects of personal pronoun referential resolution and item identification,this paper sets out to build a referential resolution model based on personal pronoun,mining semantic information of context from text level to identify and eliminate pronoun in text.It is found in the research of the personal pronoun resolution task that reasonable identification of the unsolved pronoun items can effectively improve the performance of the personal pronoun resolution model of personal pronoun.This is because proper pronoun recognition can introduce unnecessary noise and interference into the anaphora resolution process.Therefore,it is very important to study the undecipherable items of explicit personal pronouns.Most of the current researches on anaphora resolution are based on shallow machine learning model or manual extraction of semantic features of related pronouns,which cannot well mine the deep semantic information in the text and affects the performance of the anaphora resolution model.Therefore,this paper studies the anaphora resolution of personal pronouns in the following two aspects:1.For the personal pronoun anaphora resolution task,a model based on deep learning was constructed.The word vector method was combined with the attention mechanism model to construct the personal pronoun anaphora resolution model.Firstly,by analyzing and summarizing the characteristics and expression rules of personal pronouns,the corresponding features are extracted according to the characteristics of the induction.Then the attention mechanism can be used to adjust the weight and avoid the loss of information transmission between layers.Finally,DBN is used to dig the semantic information of personal pronoun at a deeper level and integrate it with the characteristics of knowledge and rules to complete the task of personal pronoun resolution of personal pronoun.The final accuracy of the experiment reached 81.14%.2.The task of identifying personal pronouns to be eliminated.This paper constructs a model framework of personal pronoun undigested item recognition based on the attention mechanism of convolutional neural network(CNN)and bidirectional short-short time memory network(Bi-LSTM).CNN network was used to extract local information of pronouns in the text,and Bi-LSTM network was used to extract contextual semantic information of personal pronouns in the text.In addition,eleven features at the level of knowledge and rules are extracted,and then the features extracted from the model are fused and input into the classifier.Then,the classifier is used to judge whether personal pronoun is a real elimination term,and finally the research on identifying the unsolved term of personal pronoun is completed. |
