基于TI-LSTM的文本自动分类算法及应用

Research on Automatic Text Classification Based on TI-LSTM

为了有效解决中文文本分类问题,提高文本分类的准确性,提出一种基于TF-IDF和神经网络相结合的文本自动分类算法——TI-LSTM算法。算法根据语义情景提取相应特征,进行量化,通过长短期神经网络(LSTM)对量化后的特征进行训练并赋予权重,最后以特征权重为依据对中文文本信息进行评价。使用TI-LSTM算法可以在保留原文语义的情况下准确提取特征。将该算法应用到长春理工大学贫困生等级分类研究中。与传统的KNN、逻辑回归、朴素贝叶斯和LSTM分类方法进行了比较,训练和测试的准确率都有了较大的提升,准确率达到了86%以上。

In order to solve the problem of Chinese text classification and improve the accuracy,a text automatic classification algorithm based on TF-IDF and neural network is proposed named by TI-LSTM algorithm in this paper.Firstly,the corresponding features are extracted and quantified in the algorithm according to the semantic situation.Then the quantified features are trained and weighted with the long-short term neural network(LSTM).Finally,Chinese text information is evaluated based on feature weight.This method has been successfully applied to the classification of poverty-stricken students in our school.Compared with traditional KNN,logistic regression,naive Bayes and LSTM classification methods,the accuracy of training and testing has been greatly improved.The automatic text classification algorithm in TI-LSTM algorithm can extract features accurately with the original text semantic,and the average accuracy rate is over 86%.