Sentiment Analysis of MicroBlog Comments Based on Multi-feature Fusion
Le Chen, Yong Huang · 2021 4th International Conference on Algorithms, Computing and Artificial Intelligence · 2021
The traditional word vector model based on context cannot effectively extract emotional features from multi-category online comment texts with short length, casual grammar and a large number of emojis. Aiming at the above, a multi-feature fusion model (WOOSD-CNN) integrating word order, semantics and dictionary features was proposed. Firstly, word order features obtained from Fasttext training were fused with semantic features obtained from Word2vec training. Then, according to the characteristics of emojis, the higher the frequency of emojis appearing in comments, the Universal Frequency of Emojis (EM-TDF) was introduced, and the emotional similarity matrix (ESV) is constructed by using it to obtain dictionary features. Finally, feature vectors were spliced of fusion and applied to emotion classification. Experiments were carried out on two datasets with multi-category MicroBlog comment. The experimental results show that the proposed model can obtain relatively accurate sentiment word vector, compared with Word2vec model, the indicators of Acc, Macro-F1 and Weighted-F1 have the highest improvement of 7.62%, 6.25% and 8.00%, respectively.