Automatically Identifying Learners' Problem Solving Strategies In-Process Solving Algorithmic Problems

摘要

Learners, not only in secondary education, but also in CS1 courses at universities, often use special learning and programming environments to practice basics of programming and solving algorithmic problems. These tools give feedback to the learners without taking into account the individual problem solving strategy. The system messages are often hardly helpful because of their pure technical information. To adapt the feedback of the" software to each learner's problem solving process it is necessary to enhance the learning environment with a tool to identify the individual problem solving strategy in-process automatically.rnLike described in [3], a tracking software was developed that records learner-system interactions differentiated by certain categories during the problem solving process. The collected data were analyzed with the help of special diagnostic software by two human researchers. Four clearly identifiable patterns in the chronology of the recorded learner-system-interactions were one of their results. Independently the observers attributed problem solving strategies to these patterns [2]. Their results were nearly identical. Now, this attribution should be validated. Therefore, another instrument to identify the used problem solving strategy is necessary, but it should not depend on stepwise observing the problem solving process. A solution is the employment of questionnaires based on the theories of Ajzen and Fishbein [1]. In this context, the learner's problem solving strategy is adequate to the behavior that is predicted by the evaluation of the questionnaires.rnThe authors' hope is that the results of analyzing the questionnaires fit to the results of the human observers.rnPattern recognition methods can be used to identify the patterns in the collected data automatically. Especially automatic speech recognition [4] has many similar requirements as the research project described here. In speech recognition the output are phonemes, here the output are the different categories of learner-system-interactions. In speech recognition there is a hidden markov model (HMM) for every word, here there is a HMM for every single problem solving strategy. With the help of transition and output probabilities, single learner-system-interactions not fitting to the strategy pattern and tolerance concerning residence time at several interactions are modeled. To train the software, frequencies in the collected data of pilot studies are counted. The software counts back the most probable pattern (problem solving strategy) using the Viterbi algorithm.rnIn this way it is possible to identify learner's individual problem solving strategy in-process automatically. Another tool to find additional other patterns will be developed and implemented.