Semi-supervised Feature Selection Method for Fuzzy Clustering of Emotional States from Social Streams Messages

Capturing the text content, especially when it reflects the human emotional states and feelings, is crucial in every decision-making process: from the item purchase to the marketing campaign, the user mood is becoming an essential peculiarity to always monitoring. This work proposed a new method based on a fuzzy clustering algorithm that takes into account human suggestions for feature selection. The method exploits two fuzzy indices, namely, the feature relevance that is initially provided by the human expertise and the feature incidence on a specific cluster. The Extended Fuzzy C-Means (EFCM) clustering is used to balance the two “dueling” indexes; a t-norm operator-based feature importance index enables the appropriate feature set selection. Experimental results on social message streams show the method’s effectiveness in supporting those emotions the human considers relevant in the textual context.