Based on the advantages of deep memory network in aspect-level sentiment analysis and the insufficiency of ignoring the location information between context words and aspects in related work,an aspect-level sentiment analysis method based on con-textual aspect memory network was proposed.Different memory network layers were used to focus on different parts of context memory and rich context-aware context information was obtained.To make full use of the inter-aspect information,an aspect memory network update module was designed to generate semantic and relational information of adjacent aspects for the desired aspects,and the input of the multi-head attention mechanism adopted two strategies to calculate the context and aspect word cor-relations.Experimental results on three benchmark datasets show that the proposed model has certain improvement in the accu-racy of performance evaluation indicators and Macro-Fl-score compared with related works.
维普
万方数据