UCM Recommender System
Cognitive computing or the ability of machines to mimic human reasoning is a promising technology that is developing very fast. Cognitive computers are able to learn from information they are provided with. Recent examples of such technology are IBM Watson or Google DeepMind. These two tools are technology platforms that consist of hardware/software that have the capability to learn, reason, and make inferences from an unstructured and large amount of information (Chen, Argentinis, & Weber, 2016). This technology can also be applied in universities and function as a valuable tool for self-advising.
In a university setting, the purpose is to develop a cognitive computing system, a recommender system, that is able to facilitate students' navigation through their study program. For this case, the input information for a cognitive computer can be semi-structured data like manuals and user guides but it can also be unstructured data like blogs and other social media sources. This is what University College Maastricht (UCM) is currently developing.
UCM provides students with the opportunity to design their own curriculum. In this process, there are roughly two areas that need to be addressed because they are decisive for the outcome of a student’s curriculum and its quality. The first is what you do. The answer to this question consists of the courses, skills and projects that students have completed in order to fulfill the graduation requirements of the program. The second is how they do it. The answer to this question consists of the actual program that students have followed in order to arrive at the graduation requirements: the order in which a student took courses and how they navigated from one course to the other.
In this process of deciding what students want to do and how they want to do it there are several stakeholders that will influence what the actual program is going to look like. Examples of stakeholders are the students themselves, faculty members that advise students, and scheduling officers. Considering that students are immediately involved in designing, understanding and evaluating their progress through their degree it is crucial for them to have resources that inform their decisions. AI is able to help students with this process to ensure that they do have all or most of the information available to them to make decisions during their studies.
A Recommender System is able to assist students in achieving different learning objectives and competencies. Besides being critical about the academic material that they engage with, students should also think critically about their path at university. In order to do so, they need to have access to as much information as possible in a format that enables them to think about what they have done, what they need to do and what they can choose from. The Recommender System improves students' information position, as it is able to compile and synthesize information in seconds in a way that would take a long time and effort for a person to do. For instance, by feeding the Recommender System with the descriptions of all courses available within a degree and at the whole university, and creating keywords (e.g., personality, algorithm, debate, Europe) based on these descriptions, a student is able to select keywords of interest and have the system provide course possibilities for those interests. Selecting keywords and thinking about their combinations is an aspect that invites students to think critically and systematically evaluate their interests to then have access to recommendations.
Another aspect of the Recommender System that invites students to think critically and actively reflect on their development at university is to assess the skills they have or wish to acquire. Based on the type of assessment that is included in a course description, the system is able to provide a visualization of the skills taught in different courses. For example, if a student has taken 10 courses out of which 9 included essays as examination and one included a presentation, the system provides an overview of the skills gathered in these courses and shows the user that in terms of writing essays, s/he has gathered a substantial amount of knowledge. If this student wants to improve her/his presentation skills, s/he can now select courses that have presentations as one form of assessment. Additionally, it is possible to select some courses and see the skills that one would obtain by taking those options. By employing these functionalities, students can move beyond the mere description of courses provided in an academic transcript. Engaging with the functionalities of the tool requires students to be active in self-inquiry as they build their narratives and consider which skills they would like to develop. Analogously, the system can provide visualizations of the Aims of the Degree that are met through different courses.
In a context similar to UCM, self-advising is fundamental to understand the link between personal interests and the courses available at the university, and the recommender system is able to help with this task. Although course catalogs are available, looking through all the courses is time-consuming and is an activity that might deter students from looking beyond one academic department. Furthermore, courses that are not traditionally related to the discipline that a student is interested in might be disregarded without a more critical understanding of the course's content. The Recommender System can address these two problems and guide students in their self-advising while looking for courses that match their academic goals. Firstly, the system is able to store all the information about the courses available and provide keywords that students can choose to then get suggestions. For example, if 'sociology' is a keyword and it is selected, the system will give suggestions about courses that fall within that category. Secondly, other keywords that are not strictly bound to one single discipline can be provided which broadens the possibility of courses that would otherwise be ignored. Poverty, for example, can be studied from different disciplines, and perhaps a student would find it engaging to study it from both a psychological and economic perspective. By selecting the keyword 'poverty' the system presents all courses that are linked to this word, and invites students to think further about their aims. With more structured information, a student is able to make better choices that take many factors into consideration.
Most of the resources needed to implement a Recommender System are already available at universities. This is due to the input data being course descriptions, course catalogs, students' transcripts, aims of the course, and aims of the degree. Course descriptions aid the system in structuring information that is then presented to students. Besides mere content descriptions, courses need to be linked to specific learning outcomes and/or competencies that a particular course teaches. By having these details, the system is able to provide a visualization of skills acquired.
Transcripts provide insightful data on the progress made by students. This information can be completely anonymized as the system does not need to know the name of the student or even their real student ID. What the system needs is information related to how an individual navigated the program (i.e. which courses were taken, in what order, with what result). This is important to make suggestions for course selection and for the system to understand different profiles. Students with similar fields of interest, for example, might benefit from understanding how a peer navigated the program and the courses they selected. However, the system does not need to know who the student is so a completely random number can be used as an identifier for the student profile.