I've been thinking about ways to improve emotional self-awareness without relying on mood logs, which often feel forced or artificial. Instead of asking "how are you feeling?", what if we looked at what someone did during the day — and derived their emotional state from that?

I tried implementing this using a BERT-based model trained on Google’s GoEmotions dataset (which covers 28 nuanced emotions like remorse, joy, optimism, etc.). The idea is to let users reflect on their good or bad deeds and then use NLP to classify the underlying emotions.

Some patterns I’ve noticed:

Visualizing daily dominant emotions on a color-coded calendar gives a quick emotional overview.

Comparing emotions across days offers insights into how our actions shape our emotional trends.

Action-first reflection seems to reduce the pressure to “perform” emotionally compared to mood logs.

This is all part of an Android experiment I’m working on. Curious if others here have explored similar approaches — using AI to model emotion based on actions rather than self-reporting. Would love to hear your thoughts or feedback.