An Improved Method for Fine-grained Sentiment Analysis Based on Sentiment Word Embedding

Abstract

The invention discloses an improved fine-grained sentiment analysis method based on sentiment word embedding, which includes: acquisition of text data sets, manual labeling and preprocessing. Compute the semantic word vector corresponding to each word in the text dataset. Get a collection of sentiment words. Compute a set of sentiment phrases for each word in the training dataset. Calculate the sentiment word vector corresponding to each word in the training dataset. Construct sentiment embeddings for each word in the training dataset. Train a classifier to get a fine-grained sentiment analysis model. The present invention utilizes prior emotion knowledge combined with word embedding model, and constructs emotional word embedding suitable for fine-grained emotion analysis for emotion analysis. The present invention can better identify emotional information at the word level, describe users' emotions more accurately and fine-grained, and can be used for fine-grained emotional analysis tasks, such as: analysis of user consumption habits, analysis of user comments on products, etc.

Description

technical field [0001] The invention belongs to the technical field of Chinese text sentiment analysis, and in particular relates to an improved fine-grained sentiment analysis method based on sentiment word embedding for Chinese short texts such as microblogs and network comments. Background technique [0002] With the rapid development of the Internet, millions of people use social networks every day, such as Weibo, Tieba and other online platforms to express their views on products, services, news, events, etc. Analyzing the opinions or opinions expressed by users is very important for marketing professionals and researchers. Sentiment analysis of Weibo texts has become increasingly important due to the popularity of Weibo in Chinese society. Since 2013, the China Computer Federation (CCF for short) has specially set up the Chinese Weibo sentiment classification evaluation task in the second Natural Language Processing and Chinese Computing Conference (NLPCC for short), ...

AI Technical Summary

Problems solved by technology

However, these methods have certain limitations in sentiment analysis tasks: word embedding implements many NLP tasks by learning low-dimensional continuous value vector representations of words, but the word representation obtained by traditional word embedding methods only contains semantic information in the text corpus , without considering the emotional information in the text corpus

Images