EMSEIF Provides an Explainable AI Framework for Multi Stakeholder Curriculum Feedback and Quality Assurance in Higher Education

Universities are vast, data-rich places that often struggle to turn the sheer volume of information they collect into meaningful action. Every year, students, teachers, alumni, parents, and employers fill out forms, write comments, and sit on committees to discuss how well a curriculum is working. This feedback is the lifeblood of quality assurance, yet in many institutions, it gets lost in a fog of manual reading, simple percentage charts, and committee minutes that are hard to trace back to specific decisions. The core challenge is not a lack of information, but a lack of clarity: how do you take thousands of individual voices, each with a different perspective and a different way of speaking, and translate them into a clear, fair, and actionable plan for improvement?

This is where the field of educational data mining steps in, attempting to use computers to find patterns in human feedback. However, simply asking a computer to say whether a comment is "good" or "bad" is often too blunt an instrument. A student might praise a teacher's energy while simultaneously complaining that the course lacks real-world projects. To be useful, a system needs to understand these nuances, identify exactly which part of the curriculum is being discussed, and explain why it reached that conclusion. Furthermore, because these decisions affect real people's education and careers, the system cannot be a black box; it must be transparent enough for human leaders to trust its suggestions and verify its fairness.

Researchers at MIT Art, Design & Technology University in India have developed a new framework called EMSEIF to solve this problem. Rather than just building another tool that predicts outcomes, they created a complete system that connects raw feedback to concrete curriculum changes while keeping a human in charge of the final decision. The team tested their system using a massive, real-world dataset: the 2023–24 annual quality-assurance records from their own university. This archive was a 1,542-page document containing thousands of feedback forms, rating tables, and official letters. The researchers converted this entire archive into a structured digital library of 8,742 individual feedback units, ensuring that every piece of data was tagged with who said it, what department it came from, and what specific topic it addressed.

The heart of the system is its ability to listen to different voices without flattening them. In a typical university, a student might ask for more "hands-on practice," while an employer might ask for "industry exposure." These sound different, but they are asking for the same thing. The researchers taught their system to recognize these different ways of speaking and group them under a common theme, such as "practical exposure." They also built a digital map, known as a knowledge graph, that links these themes to specific courses and learning goals. This allows the system to see that a request for "more projects" in one department is connected to a request for "better equipment" in another, creating a unified picture of what needs to change.

To ensure the system is trustworthy, the researchers designed it to be explainable. When the computer suggests a change, it does not just give a yes or no answer. Instead, it produces an "evidence card" that shows exactly which words in the feedback led to the suggestion, how confident the system is, and whether people from different groups—like students and employers—agree on the issue. If the system detects that it is treating one group of people unfairly, it flags this for human review before any recommendation is made. This "human-in-the-loop" approach means that the computer handles the heavy lifting of sorting and analyzing data, but a human academic leader must always approve the final action.

The results of the study showed that this approach works significantly better than older methods. When tested against standard computer models, the new system was much more accurate at identifying specific curriculum issues and predicting which feedback was actionable. It successfully identified three major areas where all stakeholders agreed on the need for change: the need for more practical, hands-on learning; the need to update skills in artificial intelligence and digital tools; and the need to better prepare students for the job market. For example, the system found that while students in an aerospace program were generally satisfied, they specifically wanted more access to aircraft models and field visits, a detail that might have been missed in a simple overall satisfaction score.

Crucially, the study demonstrated that the system could maintain fairness across different groups. Before adjustments, the system showed small biases where it might have been less accurate for certain departments or stakeholder roles. By applying specific corrections, the researchers brought these gaps down to a level that met strict governance standards, ensuring that no single voice was systematically ignored. The system also proved that it could align its suggestions with actual actions taken by the university. When the researchers compared the computer's recommendations to the official "action-taken" letters and committee minutes from the university, they found a strong match, proving that the system was not just generating ideas, but identifying the very changes the institution was already trying to make.

The researchers are careful to note that this is not a magic wand that solves all educational problems automatically. The system is a tool for support, not a replacement for human judgment. It cannot decide what the best curriculum is; it can only organize the evidence so that the people who make those decisions can see the full picture clearly. The study also highlights that while the system worked well within this specific university, applying it elsewhere would require local adjustments to fit different cultures and rules. Nevertheless, the work provides a clear, reproducible blueprint for how universities can move from collecting feedback to actually using it, closing the loop between what people say and what institutions do.

In the end, the study offers a new way to think about quality assurance. It suggests that the value of feedback lies not in the final score or the average rating, but in the specific, actionable evidence hidden within the comments. By combining advanced computer analysis with human oversight and a commitment to fairness, the researchers have shown that it is possible to turn a chaotic mountain of paperwork into a clear, navigable path for improvement. The system does not just tell universities what is wrong; it shows them exactly where to look, why it matters, and how to fix it, all while keeping the human voice at the center of the process.