Research On Key Technologies Of Network Public Opinion Information Identification And Analysis

With the rapid development of Internet technology in China,people are gradually using electronic devices to do daily work and to communicate through network channels.The netizens have become the main media for the spread of public opinion information on the Internet,and public opinion information on the Internet has exploded.The lengthy public opinion data not only seriously wastes the time and energy of the analysts of public opinion information,but also brings impact in the negative content to social stability.In addition,there are a lot of public opinion information which is valuable to relevant departments in the mass mixed network data,so how to obtain and efficiently analyze these data to help them better understand social conditions and public opinion is an urgent problem to be solved.Based on the above problems,related research & exploration on text summary technology and text classification technology involved in the identification and analysis of Internet public opinion information have been conducted in this paper.The main research contents of this paper are as follows:1)Aiming at the problem that the public opinion information is too long and contains subjective emotional content,a text summary model is implemented based on the Seq2 Seq model with attention mechanism.With this model,a brief summary information is generated for each piece of public opinion information,and the information is simplified.In addition,the Coverage mechanism is used to solve the problem of generating too many repeated words of the model;2)Aiming at the problem that the current abstractive summary models make less use of text subject information,a supervised algorithm is used to extract the keyword information of the text,and this information is used to improve the attention mechanism in the model so that the model is more sensitive to the text subject information and the effect of the model is improved;3)Aiming at the problem that most abstractive summary models encode text at the word level or character level,a double-encoder text summary model is proposed.The text is encoded at the word level and clause level in this model,which makes the context vector uesd by the decoder contain more information and the generated summary is more accurate;4)Aiming at the problem that the input sequence of the text summary modelencoder is too long,and the RNN encoder gradually loses the previous position encoding information in the text during the encoding process,a two-stage text summary model is proposed in this paper.Firstly,a supervised algorithm is used to select clauses in the original text that are more similar to the text topic and more text keywords are contained,and then input these clauses into the abstractive summary model for second-stage training.The network space and time overhead is reduced while the model performance is improved;5)Aiming at the problem that public opinion information is too fragmented and cluttered,and it is difficult to obtain high-level valuable information,the information is structured using text classification technology in this paper.Taking the police information of Chengdu Public Security Bureau as an example,according to its data characteristics,a novel text classification model SPCNN was designed and implemented based on CNN,and its classification effect is better than other comparison model's.