Quantum-like implicit sentiment analysis with sememes knowledge

Sentiment analysis, which can discriminate the sentiment tendency of subjective texts, is one of the important research works in the field of natural language processing. Sentiment analysis can be divided into explicit sentiment analysis and implicit sentiment analysis according to whether the text contains explicit sentiment words. Most of the current research work focuses on explicit sentiment analysis, while implicit sentiment analysis has become one of the most challenging research tasks because the sentiment characteristics are too implicit. In this paper, “Fused with Sememe Knowledge Quantum-like Chinese Implicit Sentiment Analysis (FSKQ)” is proposed, which introduces the density matrix in quantum theory and takes sememe, the smallest common sense semantic unit in natural language, as an external knowledge base to build a sememe-based density matrix. The matrix can be regarded as a complete knowledge system with strong generalization, which models the global information of the most fine-grained semantic knowledge. Its incorporation into the text vector results in a high quality of text representation, which effectively improves the performance of the model in Chinese implicit sentiment analysis. Ablation experiments and comparison experiments are conducted in the SMP ECISA2019 dataset, and the results show that the F1 score of the model is improved by 2.6% compared with the best model, which proves the effectiveness and superiority of the idea. In addition, in order to verify the performance of the proposed method in terms of text representation quality, it is also applied to existing models in the aspect-level sentiment analysis and event detection, and it is compared with the original model without using the idea and the baseline model on Twitter, Lap14, Rest14/15/16 and ACE2005 datasets. The results show that compared with the original model and the baseline model in this field, the model combined with the idea improves the accuracy and F1 score, which further proves the effectiveness, superiority and generalization of the FSKQ model.