Multi-labeling strategy based on a heuristic text segmentation for educational content

In this study, it is aimed to extract the learning concepts from the educational contents, to segment the context into some overlapped text blocks that have semantic integrity, and to label the paragraphs with in the text blocks with multiple learning concepts. The study uses the History of Art book taught in schools affiliated to the Republic of Turkey Ministry of National Education. Natural language processing and heuristic clustering techniques are applied on the book and it is aimed to determine which learning concepts are associated with each paragraph of the document. For this purpose, feature vectors representing the parsed text blocks are extracted and the Particle Swarm Optimization clustering technique is applied after applying Principal Component Analysis on the feature vectors. In addition, the segmented text blocks matched with the learning concepts presented by an expert in the book to make a performance analysis of the proposed system. Then, the weighted mean squared error is calculated by comparing expert opinions and system outputs. The obtained results give hope about educational content can be decomposed into text blocks labeled with more than one learning concept.