Emotion data analysis method and apparatus

Abstract

The invention discloses an emotion data analysis method. The method comprises the steps of obtaining to-be-analyzed data; performing word segmentation processing on the to-be-analyzed data to obtain word segmentation feature data; generating distributed word vectors used for determining semantic relationships among words in the word segmentation feature data according to the word segmentation feature data; according to the feature data of the words in the distributed word vectors, obtaining complete data vectors of the to-be-analyzed data; and performing classified calculation on the complete data vectors according to a learning model to obtain emotion attributes used for determining the to-be-analyzed data. Meanwhile, the invention furthermore discloses an emotion data analysis apparatus.

Description

technical field [0001] The invention relates to data analysis technology, in particular to a method and device for analyzing emotional data. Background technique [0002] With the rapid development of the mobile Internet, people's behaviors in all aspects of life, work, and entertainment have also changed. For example, for the products, content or services provided by major e-commerce, social networking, reading and other platforms, the content generated by users' spontaneous comments and sharing has shown explosive growth. [0003] For example, on the book reading platform, there are tens of millions of book review texts generated every day, and these tens of millions of book review texts include readers' evaluations of the relevant content of each book, evaluations of authors, and performance and services of reading products. If the book reading platform can determine the reader's emotional attributes (good or bad reviews) of the book based on these book review texts, the...

AI Technical Summary

Problems solved by technology

Specifically, because Bayesian requires that the text feature attributes are independent and irrelevant, the semantic connection between words is less considered, while the feature words in text sentiment analysis are greatly affected by the context, and the emotional polarity is related to each word. Therefore, the word segmentation bias directly affects the calculation of the probability distribution of feature words, which leads to poor polarity classification results

[0013] Aiming at the maximum entropy algorithm: Although context-rich semantic information is considered in text sentiment analysis, the method of probabilistic statistics between words in long texts undoubtedly requires a lot of training time and space, and the calculation of its semantic connection is huge. cost of resources

However, by reducing the computational complexity through dimensionality reduction or feature selection methods, the number of features obtained does not significantly reduce the vector representation dimension

For example, there are as many as hundreds or even thousands of feature words for a long comment selected through the commonly used weighting technology of information retrieval and data mining (TF-IDF, Term Frequency-Inverse Document Frequency), and the clustering of feature words The subject of word clusters obtained by dimensionality reduction is mainly content words, but the feature words that reflect emotions cannot represent the entire comment, and the individual emotional word fragments have the problem of lack of semantics

Images