Towards automatic scaffolding of on-line discussions in engineering courses

On-line discussion boards play an important role in distance education and web enhanced courses. Recent studies have pointed to on-line discussion board as a promising strategy for promoting collaborative problem solving courses and discovery-oriented activities [Scardamalia & Bereiter, 1996]. However, other research indicates that existing systems for on-line discussion may not always be fully effective in promoting learning in undergraduate courses. For example, some analyses of collaborative on-line learning indicate that student participation is low or weak, even when students are encouraged to participate [Palloff & Pratt 1999; Kim & Beal, 2006]. As course enrollments increase, with some introductory courses enrolling several hundred students, the heavier on-line interaction can place a considerable burden on instructors and teaching assistants. Discussion threads are often very short, many consisting of only one or two messages, and students do not fully exploit the collaborative problem solving environment where they could discuss relevant technical issues with one another. We are developing instructional software tools that can automatically assess student participation and promote interactions by sending responses to student messages. In this paper, we present a novel software tool that applies data mining and information retrieval techniques for guiding student discussions. Given a discussion thread, the tool retrieves useful information from past student discussions in related courses and automatically presents the retrieved information on the discussion board. Our hypothesis is that we can scaffold discussions by sending messages from past students who had similar assignments or problems regarding course topics. We expect when a discussion thread is short or stagnant, such intervention may promote further participation from students. We first semi-automatically extract domain terms from textbooks specified for the courses and task related terms from discussion corpus, and use them in modelling individual messages with term vectors. We apply LSA (Latent Semantic Analysis) [Landauer and Dumais, 1997] and TF*IDF (term frequency and inverse document frequency) [Salton, 1989] in finding useful information from past student discussions. The tool exploits the discussions from the same undergraduate course in past eight semesters and related graduate courses in two semesters. We performed an initial analysis of retrieved messages. The analysis focused on the degree of relevancy to the current student question. The preliminary results indicate that the tool can retrieve moderately relevant information from past discussions.