Research On Automatic Identification Of Attributes In Items For Cognitive Diagnosis

Cognitive diagnosis is included in pattern recognition. Two problems might be encountered in Cognitive Diagnosis in the implementation process: 1. Experts have the heavy workload, easy fatigue in attribute recognizing work. 2. Q-Matrix identified by experts might not exactly reflect students'Knowledge Structure.To solve the above two issues, the attributes in an item should be recognized based on the students'responses.Identification of attributes in items might be divided to three cases for discussion under the condition that the score matrix is given:Case 1: Known attributes hierarchical structure and some items attributes are known and it corresponds to the columns in reachability matrix R.Case 2: Known attributes hierarchical structure and some items attributes are known and it corresponds to the columns in Qs matrix.Case 3: Attributes hierarchical structure is unknown, it hopes to analyze the attributes of the item according to the scoring matrix.Attribute hierarchical model (AHM) has fully considered the characteristics of human cognitive learning and knowledge structure has a hierarchical feature .And from this perspective, there is an implication relationship between items. A method based on association rules (ARs) in data mining is proposed to probe the implication relationship. In other words, an item is what constitutes sub items. Attributers in items can be obtained from its sub items.For case 2, the item categories is all known, a variety of pattern recognition methods can be considered, here probabilistic neural network (PNN) is used for classification.Monte Carlo simulation results show that both methods have a higher Pattern match ratio (PMR) and Marginal match ratio (MMR), which would at least give some proposal to experts for attributers identification .For case 3, the situation is more complicated, the discussion in details is made.