Abstract:

Self-assessment is central to the learning process, especially in distance learning. Self-assessment allows students to advance autonomously and at the pace required by their personal circumstances. An important aspect of self-assessment and personalized learning is to provide learners with formative feedback as they work through the course. Formative feedback is provided to students after answering an assessment question and is a key element in formative assessment systems. It provides students with the information needed to bridge the gap between their current and desired performance.

Our hypothesis is that, based on the exam questions collected in previous courses, we can perform analysis to generate in a semi-automatic way explanations to help the students in the learning process. The subject considered for this study is taught in the second year of a Computer Science degree in the National University of Distance Education of Spain (UNED). This subject has an advanced level and requires prior knowledge of mathematics and programming, and according to the students results is one of the most difficult subjects in the degree. Specifically, it refers to advanced data structures of computer science, and to algorithmic schemes, which represent general principles to address a problem with. Among the data structures studied are hash tables, graphs and heaps. Among the schemes studied is the greedy one, which is applicable to problems in which there is a criterion that allows to build the solutions directly, without having to undo decisions already taken. However, this approach is not applicable to many problems. Another scheme studied in the subject is divide and conquer, in which the problems are divided in other simpler ones, and the backtracking scheme in which all potential solutions are explored, while it is not proven that they can not be a valid solution. For teaching each topic the general case is presented and exemplified in a particular problem. Then, other classic problems of application of the structure or the scheme are shown.

The generation of explanations will depend on the type of issue under consideration: theoretical, practical, computational cost, choice of the best algorithmic scheme, etc. We are performing an analysis of each type of question and proposing specific solutions for each case. In order to automate the process as much as possible, we will incorporate natural language processing and artificial intelligence techniques to each of the cases. These include advanced techniques of semantic similarity, as well as domain ontologies.

For the evaluation at the current level of development, we will rely on the analysis of a group of professors with extensive experience in the subject under consideration. In the future, when the system is completed with explanations for all types of questions, we will move on to a student evaluation.

Keywords:

Self-assessment, formative feedback, computer science, artificial intelligence, deep learning.