Text handling method and system

Abstract

The invention discloses a method for handling text,which is used for text feature extraction technology. The method includes it classifies the training text based on the training model parameters and deletes the wrong classified training text in order to concentrate new training text and retain the correctly classified training text, and it sets new version of feature according to the correctly classified training text. And it trains the model parameters based on the above new training set and the new version of the feature. The invention also provides a text-handling system.

Description

technical field [0001] The invention relates to the technical field of intelligent text information processing, in particular to a text processing method and a text processing system. Background technique [0002] Mobile phone text messages have great potential and prospects as a way of advertising, but judging from the current situation, spam text messages in mobile phone text messages have caused serious nuisance problems. In order to solve this problem, advertisement publishers need to adopt effective methods to obtain relevant information of advertisement audiences, so as to deliver targeted and responsive SMS advertisements. [0003] In order to obtain the relevant information of the advertising audience, it is necessary to mine the user's interest points from a large number of user text messages. How to quickly and effectively obtain users' points of interest from a large number of user text messages is a current problem, and text mining of text messages is just a met...

AI Technical Summary

Problems solved by technology

[0005] In the above prior art, in the traditional text feature extraction method, each document in the training document set has a strong correlation with the corresponding category

For SMS texts, a large amount of SMS texts are needed as training data sets to train the model during training, but due to the huge amount of training texts, it is impossible to manually classify each piece of training data accurately, resulting in the training text set itself including A large amount of noise data has low reliability. Using the traditional feature extraction method to extract SMS features based on the training set will cause the extracted feature set to contain more noise features, reducing the training features extracted from the training text. The reliability of the set also further reduces the accuracy of the classification

[0006] In addition, because SMS texts are different from traditional document texts, they often include some variant and irregular texts, called singular words, such as: QQ, SG, MM, and often have many separators or use different separators, It is called a singular symbol, so too much interference information is mixed in the SMS text, which leads to the extraction of a large number of wrong features, or noise features, in the text feature extraction and feature selection, and further reduces the training time extracted from the training text. The reliability of the feature set and the classification ability of the classification system

Images