0 reactions
7,800 concepts embedded and projected into 2D — visualising a universal semantic space
This is a follow-up to a post I shared here a few days ago, after refining the dataset and projection.
Each point represents a distinct concept (objects, ideas, foods, biological entities, social constructs, technologies, etc.).
Process (high level):
* Each concept is first encoded into a compact, structured semantic representation (a fixed-width trait code).
* Those codes are embedded into a high-dimensional vector space.
* The vectors are projected into 2D using 'PacMAP' for visualisation.
Colours indicate top-level categories (Physical, Functional, Abstract, Social).
What I find interesting is that:
* Clear semantic clusters emerge *without* any hard-coded ontology.
* Some domains form tight islands (e.g. biological taxa, culinary items), while others stretch into gradients.
* A small number of concepts act as bridges between otherwise distant regions.
* Wikidata includes a lot of Apples
This isn’t intended particularly as a “map of knowledge”, but as a visual exploration of how structural similarity and semantic similarity interact at scale.
Source: https://factory.universalhex.org/explorer (select UHT-PACMAP for this specific visualisation)
Data is mostly from wikidata, with some recent 'community' additions.
Happy to go into detail on any aspect, if anyone is interested!
#education
source
This is a follow-up to a post I shared here a few days ago, after refining the dataset and projection.
Each point represents a distinct concept (objects, ideas, foods, biological entities, social constructs, technologies, etc.).
Process (high level):
* Each concept is first encoded into a compact, structured semantic representation (a fixed-width trait code).
* Those codes are embedded into a high-dimensional vector space.
* The vectors are projected into 2D using 'PacMAP' for visualisation.
Colours indicate top-level categories (Physical, Functional, Abstract, Social).
What I find interesting is that:
* Clear semantic clusters emerge *without* any hard-coded ontology.
* Some domains form tight islands (e.g. biological taxa, culinary items), while others stretch into gradients.
* A small number of concepts act as bridges between otherwise distant regions.
* Wikidata includes a lot of Apples
This isn’t intended particularly as a “map of knowledge”, but as a visual exploration of how structural similarity and semantic similarity interact at scale.
Source: https://factory.universalhex.org/explorer (select UHT-PACMAP for this specific visualisation)
Data is mostly from wikidata, with some recent 'community' additions.
Happy to go into detail on any aspect, if anyone is interested!
#education
source
| 4,000 mlx | |
| 6 | |
| 0 |