Abstract: 【Purpose/significance】Sentiment classification/analysis has become an important issue in information sciences, with sub⁃stantial light shed on one overall sentiment for one textual target. The present research seeks to explore a deep learning basedmulti-aspect sentiment classification approach.【Method/process】An Attention-BiGRU-CNN model is contrived. The BiGRU and CNN layers aim at effectively capturing contextual information and local features, while the attention mechanism is incorporated to op⁃timize the weights of hidden layers. The model is expected to essentially delve into the deep semantics of text for multi-sentiment knowledge.【Result/conclusion】Experiments on a Chinese e-commerce review corpus evidence the superiority of the proposed method over other neural networks based models, with effectively improved accuracy in multi-aspect sentiment classification. 【Innovation/limi⁃tation】The research adds to literature of multi-aspect sentiment classification/analysis with methodological and practical implications
for e-commerce product improvement and marketing optimization. Further effort could be made to investigate a more scalable multi-aspect sentiment classification approach for large and cross-domain corpus whose aspects is implicitly complex and should be extracted by dedicated deep learning approaches.