A Novel Fuzzy Logic Model for Multi-label Fine-Grained Emotion Retrieval.

The traditional opinion retrieval methods can acquire the topic-relevant and subjective documents or sentences with the issued query. However, these methods usually focus on the sentiment polarities in the retrieval results, but ignore the emotion intensities. In fact, the users may pay more attentions on the emotion similarity between the query words and retrieval results, i.e. the sentences with exactly the same fine-grained emotion labels and similar emotion intensities with the query should be ranked higher. To address the problem, we propose a new method based on fuzzy set theory. According to the theory, we build a model for multi-label fine-grained emotion retrieval, which utilizes fuzzy relation equation to calculate the value of sentiment words and then uses lattice close-degree to retrieval emotions and rank on their intensity. Extensive experiments are conducted on a well-known Chinese blog emotion corpus. Experimental results show that our proposed multi-label fine-grained emotion retrieval algorithm outperforms baseline methods by a large margin.