A hybrid approach for composing groups in collaborative learning contexts

Considering that there are many alternatives in the literature for composing groups in collaborative learning contexts, we present a proposal that exhibits several features. First, and from the operational point of view, our proposal is highly flexible because i) it allows for several group sizes and an arbitrary array of grouping attributes, and ii) it may be easily adapted to consider several homogeneity/heterogeneity criteria. Second, and from the algorithmic point of view, it combines the best of two apparently opposite worlds: it uses a local brute-force search within an iterative process guided by a randomized heuristic criterion. Thus, this approach is still Non-Polynomic (NP) but in terms of the size of the groups, whereas is Polynomic (P) in terms of the number of students. Third, the experiments with several datasets, with student numbers varying from 20 to 3500, demonstrate reasonable performance and running times for this approach. We contrasted these times with those reported in 19 related works and, first taking into account certain considerations, we found that ours were lower in most cases. Nevertheless, and as the fourth feature, we make available both the datasets and the source code to allow for more objective comparisons of approaches, including our own.