A document-level sentiment analysis method based on domain-specific sentiment words

Abstract

The present invention provides a document-level sentiment analysis method based on sentiment words in a specific field, which is carried out in the following order: collect document data sets, use the Skip-gram word vector model to train the collection of prototype words, and obtain the word vector corresponding to each prototype word , and use the attention mechanism to reorganize the word vector to capture the relationship between non-continuous words in the word vector; use the asymmetric convolutional neural network and the two-way gate recurrent neural network based on the attention mechanism to synthesize words and sentences, Form document vector features; use the domain sentiment dictionary of the Skip-gram word vector model to generate emotional feature vectors; finally, use the linear combination layer to combine document vector features and emotional feature vectors to form document features that are beneficial to document classification. Sentiment analysis is widely used in product analysis, product recommendation, stock price trend prediction, etc. The method provided by the invention can accurately and efficiently perform sentiment analysis on documents, and has great commercial value.

Description

technical field [0001] The invention relates to the technical field of natural language processing, in particular to a document-level sentiment analysis method based on sentiment words in a specific field. Background technique [0002] Sentiment analysis, also known as opinion mining or opinion mining, is a fundamental task in natural language processing and statistical linguistics. Sentiment analysis is very important to understand the opinion information generated by users on social networks or product reviews, and can provide decision support for merchants and other users; in public opinion monitoring, it can keep abreast of people's attitudes towards emergencies and guide public opinion trends, etc. It has attracted extensive attention from industry and academia. Sentiment analysis is divided into word level, sentence level and document level according to granularity. Compared with word-level and sentence-level, document-level sentiment analysis needs to consider the o...

AI Technical Summary

Problems solved by technology

Compared with word-level and sentence-level, document-level sentiment analysis needs to consider the overall structure of the document (the document is composed of sentences, and the sentence is composed of words), and extracting complex features of the document makes this research complicated.

Existing research ignores the relationship between non-consecutive words, including syntactic features (phrase structures with a certain distance) and semantic features (the object referred to by “its”); some studies do not make full use of prior knowledge such as sentiment lexicons , to enrich the emotional features of documents, however, sentiment dictionaries play an important role in sentiment analysis tasks. Sentiment words are an important basis for sentiment analysis. It is necessary to establish an accurate and high-coverage sentiment dictionary, but compared with domain-specific , these sentiment dictionaries have poor performance for sentiment analysis, because domain-specific sentiment terms may not appear in general dictionaries, and the same term has different meanings in different situations in specific domains and general dictionaries, therefore, It is necessary to construct a domain-specific sentiment dictionary

Summarizing previous studies, it is found that when modeling documents, usually only the document representation or sentiment dictionary is considered, and the two are not combined, making the extracted document features relatively simple

Images