Comparing Apples and Oranges: Human and Computer Clustered Affinity Diagrams Under the Microscope
summary
Evaluates seven text‑mining models and picks fastText (Spearman r = .44) to pre-cluster short affinity notes, finding automatic clusters match human diagrams with overlap M = .694 (SD = .034) and average Jaccard ≈ .3 but far above random (< .008). A usability study with four teams on 112 notes shows pre-clustering increases discussion overhead, reduces efficiency and judged quality versus manual clustering, and experts see algorithmic groups as shallow word‑sorting; recommends intervening support (highlight matches, convey confidence) rather than full pre-clustering.
AI-written summary · about
related talks
25mThe Art of Clustering: The Good, The Bad and The Beautiful
30mWords in Space
47mAssessing the quality of a clustering
01h 20mClustering Data
19mHow to look at your data
34mWhen life gives you Orange, make data speak volumes!
45mWords in Space: A Visual Exploration of Distance, Documents, and Distributions...
06mPerson-Centered Predictions of Psychological Constructs with Social Media Contextualized by ...