Itembank redundancy checking based on multi-instance learning

A method based on multi-instance learning to improve the itembank redundancy checking algorithm is proposed. Redundancy checking for items with multiple questions is addressed through transforming it into a multi-instance learning problem. High-frequency words extracting algorithm based on suffix tree is used to extract content features of items and the use of thesaurus can be avoided. Combined with metadata features of items, a method to compute item similarity is proposed. Experiments on the realworld itembank dataset show that the proposed method is an effective and feasible solution to the itembank redundancy checking problem, and achieves 91.3% precision and 92.3% recall. It laid groundwork for future work on the integration of itembank systems.