Research On The Technology Of Public Opinion Base On Deep Lexical Network Learning

With fast development of the internet and communication technology, new media based on the internet has gradually replaced traditional media and has become main body for people to obtain and spread public opinion information. Due to real-time and high efficiency of the internet, public opinion information increased in explosion mode, traditional monitoring and analysis technologies cannot meet current requirement; therefore, researching monitoring technologies has become current research hot point. This paper researches key technologies of public opinion monitoring and analysis, and the main content includes the following:(1)Research feature extraction technology, and put forward a kind of text feature extraction method based on vocabulary network. In this method, the text data is expressed in the form of map by extracting the structure information such as text frequency, text features relevance and semantic similarity, adopts network key nodes discovery technology to extract key node in the map as text feature. Experiment show that the text feature vector obtained by this method perform better in clustering.(2) Research feature coding technology, put forward a kind of dimensionality reduction method based on Sparse Group deep learning network to deal with high dimensional sparse data. In addition, this paper impoved Single-Pass incremental clustering, used coverage to calculate the similarity and introduced candidate feature vector. In order to used this incremental clustering to the result of deep learning network.(3) Design and implement public opinion monitoring system based on the preceding research result. This system has following function:discover and track network hot topic, analyze topic participator, and realize topic spread early warning.By researching key technologies of public opinion monitoring. Aiming at inaccuracy of feature extraction caused by text non-structure and highly complicated problem caused by high dimensional sparse data, this paper puts forward vocabulary network feature extraction and incremental clustering based on deep learning network dimensionality reduction. In addition, this paper designed and implemented a public opinion system.