Cross-language attribute level sentiment classification method based on translation matching

Abstract

The invention discloses a cross-language attribute-level sentiment classification method based on translation matching, which can improve the performance of attribute-level sentiment classification of a target language with scarce corpus resources by using a source language with rich corpus resources, and comprises two parts of field classification and attribute sentiment classification, in the model construction process, model hyper-parameters are set based on model features, including the number of multi-head self-attention layers, gradient inversion hyper-parameter values and the like. The field classification utilizes adversarial training of a field discriminator and a language encoder to reduce the field deviation problem of a real language and a translation language caused by machine translation, and attribute sentiment classification performs fine-grained interaction on attribute sequence representation and sentence sequence representation to obtain attribute-level sentence representation; and a final emotion prediction result is obtained through a full connection layer and a softmax layer. The attribute-level sentiment classification model provided by the invention is low in construction cost, and a comparison test verification result shows that compared with other models, the result of the attribute-level sentiment classification model provided by the invention is optimal.

Description

technical field [0001] The invention relates to a natural language processing method, in particular to a translation-matching-based cross-language attribute-level emotion classification method. Background technique [0002] Research on coarse-grained cross-lingual text classification tasks is mainly divided into two methods: one is through bilingual dictionaries and machine translation. , and then classify the translated text using a classifier based on the source language. The second is cross-language representation learning, that is, cross-language model transfer through shared feature space to achieve cross-language feature alignment at a lower cost. In cross-lingual word embedding, words with similar meanings in different languages ​​have similar vector representations. For example, Klementiev et al. used a large number of parallel corpora to train cross-language word embeddings in "Inducing crosslingual distributed representations of words", and achieved good results ...

AI Technical Summary

Problems solved by technology

[0004] The attribute-level sentiment analysis (Aspect-level Sentiment Analysis) is a kind of fine-grained sentiment classification, which relies not on the whole sentence, but on a group of words or phrases related to specific attributes. in its infancy

Images