Research On The Recognition And Classification Of The Writing Style Of Social Short Text

With the rapid development of the Internet,social platforms have won the love of users due to their simplicity,convenience,and fast delivery speed.Therefore,social platforms have generated massive amounts of social short text data,and many academic researchers use these massive amounts of data to mine the information contained therein,such as the publisher’s intentions,interests,and writing styles.As each person has different vocabulary habits,sentence rhythms,and different tone styles of speaking,his blog posts published on social platforms present different writing styles.For example: some people have sharp and playful styles,some like humor and simplicity,and some people publish loose and illogical blog posts.And these styles can be used as the basis for judging style.For example,Most people with low styles use the most basic language to express.At the same time,the precise recognition of writing style can also serve the task of personalized text generation and provide control attributes for personalized text generation.Therefore,the task of identifying writing style has certain research significance.This article divides the writing styles of short social texts from the perspective of rhetoric.Commonly used rhetoric includes metaphor,parallelism,and contrast.In addition,metaphor in metaphor is also a style,that is,metaphor recognition can also be regarded as a two-category style recognition task.Accurate recognition of metaphor can promote the understanding of semantics in specific tasks of natural language processing.The rhetorical style recognition is based on the sentence level,and the difference between various styles is learned through the training of the overall sentence.In addition to the whole sentence can reflect the style type,the rhetorical specific sentence structure or keywords can also indirectly reflect some style types.In order to more accurately dig out the publisher’s writing style,this article comprehensively considers the importance of text information and text keywords to the writing style,and proposes a method of embedding artificial attention into ERNIE＿BiLSTM to realize the identification and classification of writing style.First,use the ERNIE model to embed the text sequence and the keyword tag sequence words and add them,and encode the fused embedding matrix to obtain contextsensitive semantic representation;Then,use the bidirectional LSTM layer to train the word vector again to capture the long distance of the text;Finally,the Softmax classifier is used to predict the category of writing style.In order to verify the effectiveness of the method,a large amount of real Weibo data is obtained by using a Python crawler program as the experimental corpus for the task of writing style recognition.The experimental results show that the classification model based on artificial attention embedded in ERNIE＿BiLSTM can accurately recognize the writing style in short text data;and comparing the ERNIE and BERT models,the experimental results show that the ERNIE＿BiLSTM+ artificial attention model proposed in this paper can recognize writing style Best performance on the task.In addition,the model is also applicable to metaphor recognition tasks..