Automatic Analysis and Presentation of Learning Resources
Master thesis
Permanent lenke
http://hdl.handle.net/11250/2615826Utgivelsesdato
2016Metadata
Vis full innførselSamlinger
Sammendrag
This master thesis investigates how to make examples of student assignments searchable online, and hence better accessible. Dokker.no provides a platform where students can document their work, given a provided assignment. The student receives an assignment from their supervisors, then plans, implements and assesses his or her work online. Today, there is no reasonable search functionality within the platform. Students and teachers may benefit from searching existing assignments and student reports when investigating which kind of project they want to pursue.
We will enhance the information available to them at startup by providing examples relevant to their task. An example consists of an exercise plan together with educational goals, the implementation of the plan, and, lastly, the student s assessment of the over- all conducted procedure. A large proportion of the examples are connected to education programmes and curricula.
We look into how we can create an example retrieval system using existing state of the art components and tools. We also outline architecture that has the potential to scale as the user mass grows considerably. A user interface is created so we can perform user tests on the developed system.
Four different experiments will be carried out, all aiming at improving the example retrieval for the end user. We improve the categorization of the pre-assembled data set, by using an external taxonomy (GREP). A function to assess the quality of an example, together with Elasticsearch queries, is evolved using differential evolution. Lastly, we run a user test where we receive feedback on the performance of the retrieval system.
The pre-assembled data set consists of 21,790 examples, mainly written in Norwegian Bokmal. The reports were written by students in vocational schools in Norway. For our experiments, we manually categorize and assess the quality of 1000 examples we used to measure the precision of our implementations.