LasallEats

In our user study, 85.7% of the students who tried LasallEats said it read their mood correctly, and they rated its recommendations 4.36 out of 5. We presented the paper virtually at IEEE FMLDS 2025 in Los Angeles, and it's published in IEEE Xplore.

LasallEats recommends where to eat near De La Salle University Manila based on how you feel and what you won't eat. I came up with the concept, designed the flow and the architecture, and built the app myself. My groupmates grounded the idea in research, checked my decisions against the literature and ran the study with me.

Why mood belongs in the recommendation

Before building anything, we surveyed 21 DLSU students about how they pick a place to eat. A lot of them said they just walk around and decide. Manila traffic often keeps students close to campus, so the choice is between the places nearby, and it's usually made on the spot. Research on eating behavior says mood shapes that kind of choice, yet the food apps students already use don't ask about it. 70% of the students we surveyed said they'd use an app that did.

Mood in, one restaurant out

The app has three steps. First, you tap the one image out of six that matches how you feel. Each image stands for one of Ekman's six basic emotions: anger, disgust, fear, joy, sadness and surprise. They're a mix of memes, abstract images and photos, drawn at random from a pool of at least 10 per emotion. None of them show food, so the picture can't nudge what you feel like eating. We picked and labeled every image ourselves, by vote.

Second, you tick the cuisines and allergens you want to avoid. Third, Gemini 2.5 Pro picks the restaurant.

I didn't ask Gemini for a recommendation in general. The prompt gives it your emotion, your exclusions and a list of restaurants near campus that we put together, with strict rules: choose only from the list, never break an exclusion, return exactly one place and explain why it fits your mood. The closed list means every answer is a real place a student can walk to, and the explanation lets you judge the pick against how you actually feel.

That's the part I'd still build the same way. A model is only as useful as the context and the constraints you give it, and that's most of what I work on now.

What the misses taught us

Fear came up more often in the app's readings than students reported feeling it. That points back at the images: we labeled them ourselves, so the dataset carries our reading of each picture. If I built it again, I'd start with a larger image set labeled by more people, with Philippine memes and art that students here actually recognize.

One participant suggested reading mood from the camera with facial recognition. We considered it and dropped it for privacy and feasibility. Picking an image keeps the user in control, and nobody has to hand their face to a restaurant app.

Thanks to Katipunan Miguel Delos Reyes, Rafael Enrico Hugo, Vance Gyan Robles and Dylan Andrei Rodriguez for the research and the study work behind this, and to our adviser, Norshuhani Zamin, for helping us shape the paper.

LasallEats was a three-month class project, evaluated with a small group of students, so the results are early signals from a pilot.