5/ Findings on SLS:
- As expected, experiential topics revealed Strobe-induced simple visual hallucinations (geometric and colorful patterns)
- More surprisingly, we also found complex hallucinations (scenes, objects, faces) and reports of altered states of consciousness (ASCs).
4/ This approach enabled structured analysis of phenomenological reports while reducing manual biases in labelling.
3/ Our goals were threefold:
1. Identify experiential topics in open-ended text-based reports of the Dreamachine (multi-sensorial) experience
2. Automatically generate meaningful labels using LLM
3. Provide a reusable computational framework for phenomenological research
2/ We applied MOSAIC to reports from participants in the Dreamachine (dreamachine.world), a collective experience using Stroboscopic Light Stimulations (SLS) and music
Huge thanks to everyone who contributed!
I would love to hear thoughts on where MOSAIC could be applied next, and which aspects could be improved / added into the pipeline !
Our MOSAIC paper is out in @nconsc!
Turning first-person reports into interpretable phenomenological clusters using topic modelling and LLMs.
8/ Beyond this study, MOSAIC is a general-purpose tool that can be applied to a wide range of phenomenological states, and aimed to keep being developed in the future.
7/ Key experiential topics included:
- Geometric and kaleidoscopic visuals
- Cosmic & dream-like journeys
- Meditative & peaceful states
- Multisensory integration with music
- Dissociation & out-of-body experiences
1/ In this paper, we introduce MOSAIC, a generalisable NLP pipeline for analysing phenomenological reports. It will be fully available on my GitHub shortly! If you can’t wait to check on the code, please contact me ([email protected])
My first PhD paper is finally out as a preprint!
Mapping of Subjective Accounts into Interpreted Clusters (MOSAIC): Topic Modelling and LLM applied to Stroboscopic Phenomenology
đź”— arxiv.org/abs/2502.18318