This is a nice paper that addresses in a comprehensive way a problem in automated quiz generation systems: "An inherent problem with generating tests by randomly selecting questions within a question bank is that there will typically be several questions that are designed to assess the learner's knowledge on a single item." We can think of various labour-intensive ways of solving the problem - question metadata, for example - but that makes the question bank more effort than it's worth. What is needed is a conflict-detection algorithm. The results from the algorithms proposed were very good: "PARES successfully identified 93 out of the 103 questions, which corresponds to a success rate of 90.29%." This is a solid paper, probably the best in the newest issue of IRRODL.
Today: 2 Total: 1750 [Direct link] [Share]
