MK&G Neo Lab

Data Exploration Sprint

non-deterministic data exploration

visual search

unflattening metadata

find object by feeling

generous

interfaces / queries

April 2023

☕️ Igor Rjabinin ☕️ Michal Čudrnák ☕️ Philo van Kemenade ☕️

where to start?

personalised collections

multimodal

similarity

clustering interests

Objects of the month

spring

views

looking for photographs

3210 records

filered for iconography containing "blumen"

blumen.csv
images_blumen/
proof of concept (flask)
Application Programming Interface (API)
def cosine_similarity(vecA, vecB):
  return  np.dot(vecA,vecB)/(norm(vecA)*norm(vecB))
(private) repo on GitHub

Choose Your Own Object

 

functional prototype for visitors

 

start with a curated object,
walk through the collection
along 2 axes (visual & metadata)

 

tested in gallery, positive responses

Takeaways

exploration benefits from an queryable API 🤖

 

image embeddings can support intuitive ways
of search, exploration and recommendation 👍

 

metadata and content (e.g. image) features can be combined into meaningful UX 🤗

 

working with image embeddings is easier than I thought 😌

 

Data Sprinting is an interesting format
for insight gathering and rapid prototyping