Research On Short Text Classification Of Social Media Based On Deep Learning

With the increase of active people on social network and the occupation of traditional paper media by electronic media,the social crowd is used to get the current news through short Web News,and the network information becomes fragmentary and huge.In the face of the rapid expansion of information in society,because of the high-speed social rhythm,people urgently need accurate information positioning.In addition,in some challenging social times,the faster the orientation of the information category,the faster the access to the field,the Split Second response to disaster avoidance,access to timely assistance,implementation of assistance,etc.,provide technology-enabled solutions.At this time in the massive influx of text information,efficient carding to extract the need for help,early warning,urgent information has become an important requirement.In the classification based on traditional algorithms,the frequency of subject words is used as the basis of classification from the statistical perspective,which has many disadvantages that can't be ignored The traditional machine learning method needs manual processing and labeling of text features,and the accuracy of this labeling can't be guaranteed;The time and space cost of the partitioning process is too high,and the timeliness is lost in the processing of social media texts characterized by rapid updating.In this paper,we choose news text as experimental data,and apply the neural network of Statistics,word vector subject model,machine learning and deep learning to short text classification,in order to enhance the extensibility of various algorithms for text processing,to improve the efficiency of feature extraction in text classification,to better infer the category of text,and to build an efficient text classifier by building a good feature representation system.The work is as follows: the text classification process is constructed in the context of data algorithm updating.Using deep learning neural networks that can process or predict sequence data,mining the influence of context information on the meaning of a word in a short text,and expressing the existing semantic relation by using the sequence,to complete the construction of an efficient text classifier.In the course of training,after bottom-up layer-by-layer training and parameter tuning,the training process inputted unlabeled vector data from the bottom layer,and built the network layer by layer to the next layer,and finished the adjustment of the parameters in the model in the opposite direction,to optimize the classification model.And by comparing with the results of the text classification algorithm applied in industry,the applicability of the Complex Algorithm is considered.Finally,the proposed neural network based on CNN and GRU is validated to improve the performance of short text classification.At the same time,the trained model can predict the text type of any input.