Term Identification via Neighborhood Matrix

This example came from user study March, 2022 using Vis dataset. It will demonstrate how our participant identify interesting terms that lead to further searching.

Participant sought to use a known relevant paper to find other works related to her topic. The closest neighbors were diverse and she wanted to select the relevant ones so she used the Neighborhood Matrix View to identify salient words and observed that the word fact was extremely common, and she knew it to be important. She used this term to sort the neighbors, creating an item-based description of an expanded region which was a good relevance check. Observing the salient terms for those documents, she noticed the term “insight” and observed that it was a term used in a similar way to fact, which led to further searching.

Another participant also used the similar approach with robotics full dataset. She started with searching for neighbors of her abstract and using neighbor matrix to quickly identify related work in the neighborhood. She identified the term “projection” using Specter vector, which did not appear in her abstract but was highly related to her work. She then sorts the neighborhood matrix to find recommendations that used term “projection” most frequently and identified two related work.