.png){kind=icon}
.png){kind=icon}
Return
A Local context focus learning model for joint multi-task using syntactic dependency relative distance
.png){kind=icon}
PRE
AI
DOI：10.1007/s10489-022-03684-0.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Aspect-based sentiment analysis (ABSA) is a significant task in natural language processing. Although many ABSA systems have been proposed, the correlation between the aspect's sentiment polarity and local context semantic information was not a point of focus. Moreover, aspect term extraction and aspect sentiment classification are fundamental tasks of aspect-based sentiment analysis. However, most existing systems have failed to recognize the natural relation between these two tasks and therefore treat them as relatively independent tasks. In this work, a local context focus method is proposed. It represents semantic distance using syntactic dependency relative distance which is calculated on the basis of an undirected dependency graph. We introduced this method into a multi-task learning framework with a multi-head attention mechanism for aspect term extraction and aspect sentiment classification joint task. Compared with existing models, the proposed local context focus method measures the semantic distance more precisely and helps our model capture more effective local semantic information. In addition, a multi-head attention mechanism is employed to further enhance local semantic representation. Furthermore, the proposed model makes full use of aspect terminology information and aspect sentiment information provided by the two subtasks, thereby improving the overall performance. The experimental results on four datasets show that the proposed model outperforms single task and multi-task models on the aspect term extraction and aspect sentiment classification tasks.
Keywords:
Sentiment analysis
Aspect
Multi-task
Multi-head attention
Local context focus
Journal
IF:
3.5
Papers:
7.5K
Citations:
1.7W
