Text sentiment analysis method based on dynamic threshold and multiple classifiers

Abstract

The invention relates to a semi-supervised text sentiment analysis method based on dynamic threshold and multiple classifiers. The method herein includes: marking sentiment of a part of data with L, and data that are not subjected to sentiment marking as U; performing word pre-segmentation on the data; using word2vec tool to convert words into numeric vectors; using a sampling method with a release function to sample L repeatedly T times so as to obtain T sample sequence Lt and T non-selected sample sequences OOBt; selecting one sample sequence Lt and data with no sentiment marking, and marking them as Ut; using Lt to train Support Vector Machine (SVM) classifier Ct, predicting Ut with Ct, adding samples l having reliability reaching a threshold into Lt, and deleting from Ut; updating thethresholds; using OOBt to calculate reliability Pt of the classifier Ct to obtain T base classifiers and their corresponding reliabilities P; using T BCs to predict prediction text s.

Description

technical field [0001] The invention belongs to the technical field of text classification based on semi-supervised learning, and in particular relates to a text sentiment analysis method based on a dynamic threshold and multiple classifiers. Background technique [0002] With the rapid development of social networks, more and more people tend to express their opinions and opinions on online platforms, such as the currently popular Weibo, WeChat Moments, Douban, Zhihu and so on. In order to better understand and utilize these comments, sentiment analysis for social media user comments has become a current research hotspot. This type of sentiment analysis can be applied to different scenarios, which is helpful for various departments to make network decisions, financial forecasts, policy formulation and public opinion analysis, etc. For example, Douban users can analyze the emotional polarity contained in other users' reviews of a certain movie to determine whether to choose...

AI Technical Summary

Problems solved by technology

The typical unsupervised learning method is based on the emotional lexicon, but the network culture is changing with each passing day, and the word update speed is extremely fast, so it is difficult to obtain a suitable and real-time updated emotional lexicon

Another type of unsupervised learning is a method based on clustering ideas, but this method can only classify texts with different emotions, but cannot obtain specific emotion types

Images