**4. Conclusions and Future Work**

In this paper, we proposed the sentiment-feature-enhanced deep neural network for text sentiment classification tasks by integrating sentiment linguistic knowledge into the deep neural network via a sentiment attention mechanism. We implemented the novel sentiment attention mechanism by combining a traditional sentiment lexicon with an attention mechanism to learn sentiment-feature-enhanced word representation, bridging the knowledge gap between conventional sentiment linguistic knowledge and the deep learning method. In addition, we also designed a deep neural network to effectively combine local features within the sentence captured by the convolutional neural network and the sequence information across a sentence generated by a bidirectional GRU network, which can further improve the quality of text representation for achieving greater promotion of text sentiment classification tasks. Extensive experiments were conducted on real-world datasets with a binary-sentiment-label and a multi-sentiment-label. The experimental results showed that the proposed SDNN significantly outperformed the state-of-the-art methods for text sentiment classification tasks.

In the future, we will try to use this model to analyze the actual emotions of specific users, such as sadness, happiness or depression, which can lay a good foundation for providing better specific users' experience in terms of the marketing service industry. Furthermore, it can be seen from experimental results that the proposed model still has much room for improvement on the dataset with fine-grained sentiment labels. Therefore, the reinforcement learning method to build structured text representation without explicit linguistic structure annotations will be tried in our experiment to see if it will achieve better performance.

**Author Contributions:** W.L. (Wenkuan Li), P.L., and Q.Z. conceived and designed the study. W.L. (Wenkuan Li), Q.Z., and W.L. (Wenfeng Liu) performed the experiments. P.L. provided the data. W.L. (Wenkuan Li) and Q.Z. analyzed the data. W.L. (Wenkuan Li) and W.L. (Wenfeng Liu) proofed the algorithm. W.L. (Wenkuan Li) wrote the paper. All authors read and approved the paper.

**Acknowledgments:** This work was supported by the National Natural Science Foundation of China (Grant No. 61373148), the Science Foundation of the Ministry of Education of China (No. 14YJC860042), and the Shandong Provincial Social Science Planning Project (No. 18CXWJ01/18BJYJ04/17CHLJ33/17CHLJ30/17CHLJ18).

**Conflicts of Interest:** The authors declare no conflict of interest.
