Research On Chinese Punctuation Prediction In Plain Text Scenarios

With the continuous development of computer technology and the Internet,the social and human-computer interaction between human beings has gradually changed,from a single way of communication of words to a variety of ways of words,audio and video.The change of interaction mode brings more convenience to people’s daily life,and also challenges the update and iteration of human technology.For example,in the current social software,the voice social mode usually transforms the voice signal into the text information,and then presents the text as the carrier of information transmission in front of people.However,in most scenarios,the text information from audio transcription does not contain punctuation,which plays an important role in the expression of human emotions and semantics.The same sentence is given different punctuation will also transmit different emotional information.So it is very important for the development of this kind of technology to give the correct punctuation to the text,which makes the communication between people and machines more fluent.In the early stage of researchers’ research on punctuation prediction,most of the research objects are based on audio information and text information,and even rely too much on audio information.With the development of natural language processing technology,the task of punctuation for plain text also has a more important application scenario,but the features of plain text information in the task of punctuation prediction have not been fully mined,so the task of punctuation prediction in the scene of plain text has great research value.In the development of punctuation prediction task,the prediction method of machine learning language model has been widely used.However,due to the poor ability of data feature extraction and excessive dependence on audio information and other factors,the model can complete the punctuation prediction task to a certain extent,but the overall accuracy is not high and the effect is not satisfactory.With the further increase of data volume and the extensive application of deep learning,the punctuation prediction model based on deep learning has attracted the attention of researchers,but at present,there is a further exploration space for the research of the features contained in this information.In order to solve these problems and difficulties,this paper proposes a deep learning method to solve the problem,and has achieved good results.The main innovations of this paper are as follows:1.This paper analyzes the current research situation of punctuation prediction task and points out the shortcomings of pure text punctuation prediction in the current research,combined with the current research results and the research progress of related sequence annotation task,gives the improvement direction.2.Propose a semantic enhancement module for punctuation.In this paper,attention mechanism is used to implement this module.Through empirical analysis,it is proved that the introduction of this module can help the pure text punctuation prediction.It shows that there is still a rising space in the model of text semantic representation only by word vector.3.Multi task learning mode is introduced to enhance the prediction effect of the model.Through the introduction of multi task training mode,the feasibility of multi task learning mode in punctuation prediction task is proved.In this paper,part of speech information is introduced as an auxiliary task to improve the original model,and the experimental results are improved to some extent.It is proved that there is a certain correlation between the part of speech information prediction task and the punctuation prediction task.The introduction of other information in disguised form enhances the prediction ability of the model to the punctuation point,which indicates that such information should be fully used in practical application.To sum up,this paper proposes a set of punctuation prediction algorithm based on long and short memory neural network.This algorithm makes full use of word meaning information,part of speech information and punctuation similar information,and obtains considerable results,which provides a new research direction for the future research.