Researchers at the Faculty of Engineering of Universitas Negeri Malang are studying whether pairing ChatGPT with content-based filtering can work as a learning assistant for Engineering Physics, according to a user-published article on Kompasiana that does not represent the official views of the Kompas newsroom.
The study’s 2026 progress report records data collection through a survey of roughly 100 students from the class of 2024 across four classes. The research team set out to answer three questions: whether students feel helped when learning with the tool, how relevant the recommended materials are to their actual needs, and what constraints they face when using the technology.
Why Engineering Physics Needs a Different Approach
Engineering Physics covers mechanics, dynamics, energy and thermodynamics, and it requires students to grasp concepts while also working with symbols, units, diagrams and formulas. In the Mechanical Engineering Education study program at Universitas Negeri Malang, the course carries three credits under the 2023 curriculum.
The article gives the example of two students studying motion on an inclined plane. One is unsure how to draw the direction of forces; the other understands the diagram but still errs when breaking a force into components. Both type the same topic into ChatGPT, even though their learning needs differ.
How Content-Based Filtering Fits In
Content-based filtering, or CBF, is a recommendation approach that matches the characteristics of learning material — topic, difficulty level, prerequisites or presentation format — against a user’s profile. A student who has not yet mastered free-body diagrams could be prioritized for a concept summary, a visualization video and basic exercises before moving on to more complex applications of Newton’s laws. ChatGPT would handle the conversational explanation of concepts, while CBF supplies the logic for selecting the material.
The article is careful not to claim the system is proven: demonstrating effectiveness still requires processed data and appropriate testing evidence.
What a 2025 Survey at the Same Faculty Found
A separate 2025 study at the same faculty on the ethics and implementation of generative AI surveyed students in Mechanical Engineering Education and Mechanical Engineering. Of 102 respondents, 100 said they use AI to help with learning, and 95 chose ChatGPT as their main AI tool. That earlier study also flagged concerns about inaccurate information, dependence on the tools and the ethical limits of using AI — a reminder that answers about formulas, units and assumptions need to be checked against trusted course materials rather than copied straight into assignments.
What the Research Aims to Establish Next
The 2026 study is still in the data-processing stage, so its findings on whether students feel helped and how relevant the recommendations are have not yet been reported. The researchers’ stated position is that a good learning assistant should show why a formula is used, introduce prerequisites before advanced material and push students to attempt problems on their own — with lecturers still responsible for checking accuracy and assessing whether students genuinely understand the problem-solving process.
Questions Readers Still Ask
What exactly is content-based filtering in this study? It is a recommendation method that matches material characteristics such as topic, difficulty, prerequisites and format to a student’s profile, so a student lacking free-body diagram skills gets foundational material first.
How many students are in the 2026 study? About 100 students from the class of 2024, surveyed across four classes, according to the progress report.
Is the system already proven to improve grades? No. The article states that declaring effectiveness requires processed data and proper testing evidence, which are still pending.
How common is AI use among these students? In the separate 2025 faculty survey, 100 of 102 respondents reported using AI for learning, and 95 named ChatGPT as their main tool.
The 2025 survey found nearly every respondent already using AI for study help, while the 2026 research examines whether the materials recommended to students actually match what they still need to learn.
Worth a look