[Objective]In order to solve the problem of using affective knowledge to enhance syntactic dependency graphs in existing aspect-based sentiment analysis, which ignores the syntactic reachable and positional relationships between words, and insufficient extraction of semantic information, this paper proposes an aspect-based sentiment analysis model based on position-weight reachability matrix and multi-space semantic information extraction.

[Methods]Using reachability matrix to add syntactic reachability relationships of words to syntactic dependency graph, modifying the reachability matrix based on position-weight mechanism to enhance contextual feature extraction, and then fusing it with the knowledge enhanced dependency graph to extract aspect word features. Simultaneously using multi-head self attention mechanism to learn contextual semantic information from multiple feature spaces, and finally fusing feature vectors containing positional information, syntactic information, affective knowledge, and semantic information for sentiment polarity classification.

[Results]When using the GloVe pretrained corpus, the accuracy of the model on the Lap14, Rest14, and Rest15 datasets was improved by 1.00%, 1.25%, and 1.11% compared to the better model. When using BERT, the accuracy on the Lap14, Rest14, Rest15, and Rest16 datasets was increased by 0.76%, 0.22%, 1.98%, and 0.31% compared to the better model respectively.

[Limitations]The model was only tested on public datasets and not on other datasets such as Chinese.

[Conclusions]The model improves the aggregation effect of graph convolution network, enhances contextual feature extraction, enhances semantic learning effectiveness, improves the accuracy of aspect-based sentiment analysis.