Text classification algorithm based on hybrid multinomial distribution

Abstract

The invention provides a text classification algorithm based on hybrid multinomial distribution. The algorithm comprises the following steps: S1, inputting a training set text; S2, calculating and storing probability distribution of all text classifications C; S3, initializing parameter values [theta], [pi]k, and the number K of components of the hybrid multinomial distribution; S4, using the current parameter values [theta] and [pi]k to calculate expectation of a log-likelihood function of complete data with respect to hidden variable posterior probability distribution; S5, using an EM algorithm to train the parameter values [theta] and [pi]k of the hybrid multinomial distribution; S6, for different number K of components, respectively drawing figures lines of forecast errors on the testing set and the training set by the model, selecting a K value whose forecast error is minimum; S7, outputting a result. Beneficial effects of the algorithm are that the naive Bayes algorithm is combined with hybrid multinomial distribution, the EM algorithm is used to estimate parameters of a hybrid model, to improve classification precision of the model.

Description

technical field [0001] The invention relates to a text classification algorithm, in particular to a text classification algorithm based on mixed multinomial distribution. Background technique [0002] With the emergence of a large number of online texts and the rise of machine learning, large-scale text classification and retrieval have aroused the interest of researchers. A large number of results show that the method based on statistical learning has high text classification accuracy and can be applied to any field of learning, making it the mainstream method of text classification. [0003] The traditional text classification algorithm based on Naive Bayes assumes that under a given text category, each text feature vector attribute is independent and identically distributed. Due to the simple assumption, the text classification algorithm based on Naive Bayes has a small computational complexity. In some cases, better classification results can also be achieved. However, ...

AI Technical Summary

Problems solved by technology

[0003] The traditional text classification algorithm based on Naive Bayes assumes that under a given text category, each text feature vector attribute is independent and identically distributed. Due to the simple assumption, the text classification algorithm based on Naive Bayes has a small computational complexity. In some cases, better classification results can also be achieved. However, in actual tasks, this conditional independence assumption is difficult to establish and has great limitations. It is necessary to consider relaxing the attribute conditional independence assumption to a certain extent.

Images