A Sentiment Classification Method Based on Parts of Speech Combination and Feature Selection

Abstract

The emotion classification method based on part-of-speech combination and feature selection of the present invention comprises the steps of: first initializing the word-part-of-speech Word2vec model; secondly the data is preprocessed, and based on the sentiment lexicon from the preprocessed data, select the one with emotional information Feature words; then each feature word and part of speech of text is combined, and text is converted into word part-of-speech pair sequence text; Then obtain the vector of word part-of-speech pair sequence text each feature word by word-part-of-speech Word2vec model, and for each A piece of text adds the word vectors according to the dimensions and takes the average value to represent the text, so as to obtain the feature vector of the text; finally, the SVM classifier is used to obtain the sentiment classification model. The beneficial effect is: using the sentiment dictionary to extract feature words, highlighting the feature words with single-sentiment information; on the other hand, based on phrase structure optimization, word segmentation extracts phrase structures with emotional tendencies, and combines words and parts of speech to solve the problem of polysemy .

Description

technical field [0001] The invention relates to the field of computer science, in particular to an emotion classification method based on part-of-speech combination and feature selection. Background technique [0002] With the rapid development of social networking platforms, especially Weibo, a large number of netizens can express their opinions and emotions on social events more conveniently, resulting in a large amount of Weibo comment data, which contains rich opinions and views. Emotional information, how to deeply analyze and mine the emotional tendency of the massive data of Weibo texts has become a hot research direction. Traditional sentiment classification methods only focus on lexical features and syntactic features, ignoring the semantic features between words. [0003] Although the word vector model trained by traditional Word2vec can reflect the potential semantic relationship between words, there are often some problems when training the model. First, the Wor...

AI Technical Summary

Problems solved by technology

[0003] Although the word vector model trained by traditional Word2vec can reflect the potential semantic relationship between words, there are often some problems when training the model. First, the Word2vec tool cannot directly extract the phrase structure that better reflects the emotional tendency of the text. For example, "unhappy" is divided into "no" and "happy". Word2vec learns the contextual semantics of the words "no" and "happy" during training, and cannot directly learn the vector of the phrase "unhappy".

The second is that it is impossible to distinguish the semantics of the same word under different parts of speech. For example, "Xiao Ming bought a bundle of incense and used it for sacrifices, but the incense I bought this time is too rubbish" and "The rice cooked by Xiao Ming is really fragrant", in the previous sentence "Xiang" in "Xiang" is a noun, which refers to the thin strips made of sawdust mixed with spices used in worshiping ancestors or worshiping gods. It has no emotional color and is a neutral word; Smell is a compliment

[0004] Traditional data storage and processing methods greatly waste computer resources and time

Moreover, due to its step-by-step processing mechanism, the traditional Hadoop cluster limits its performance efficiency, and the I / O overhead for the disk is extremely high.

Images