Mass Mining Of Short Texts And Their Visualization

With the rapid development of mobile Internet, the Internet has become an important platform for the exchange of ideas, from MSN to QQ and Twitter to micro-blog. A variety of network platforms are quietly changing the world and the traditional media of communication and marketing approach. At the same time, the short text information, which generated in social networks and contains a large number of immeasurable commercial and social value, is exploding. How to use text mining technology to process text message has become an urgent need. Current research on a variety of text mining technology is growing rapidly, opportunities and challenges coexist. Because of its unique features, it is difficult to use traditional methods to model short text messages. Its sparsity semantic features lead to undesirable results. Therefore, this paper adopts deep learning approach to study short text messages.This paper analyzes the characteristics of Chinese short text messages and microblogging messages, besides, explores the applications of deep learning in text mining. Finally, a method based on LSTM (Long Short Term Memory) was used to process short text messages. Compared to most conventional text mining based on words model, LSTM mostly differs in that LSTM considers structural information for the entire sentence rather than the frequency of a simple word or words. Besides, the paper tentatively uses deep learning approach to mine Chinese short text and, to some extent, solving the problem of Chinese characters input and proposing a word vector mapping method based on Chinese characters strokes. Through analyzing the word-formation methods and pronunciation rules of Chinese characters, a 32-dimensional vector can be mapped to input deep neural network. LSTM is a kind of feedback neural network. It can process time series data efficiently, therefore, the text data can be considered as a sequence of data to be generated into multi-layer network architecture for processing by the use of deep leaning.Finally, Weibo message was used as test data. Experiments show LSTM-RNN is feasible and effective in topic mining task. During the experiment, the model will be based on the theme of short text messages as a control to verify the accuracy of the method.