Local + Global Explanation Identifies Spurious Recommendation

This example uses Vis dataset and shows how local + glocal explanation helps identify spurious recommendations.

all match words are dimmed, indicating this recommendation is not so related.

Looking at the neighborhood distribution of 2 selections, which are very different. This confirms that the 2 selections are not very related.

Another example in Vis dataset. Serendip paper and its spurious recommendation Nested Model

Compared to other neighbors (highlighted - ‘text’ ‘document’, corpus), the common word is “level”. This explained why this was chosen, also Assessed this is not relevant and Identified the difference

Compared the neighbor distribution, all its’ neighbors are away from serendip’s