I have 40 items rated on 25 semantic differential scales. I have taken the average of each item's rating on each of the scales. I would like to now compute a distance measure between items, based on those averaged scores. Would Euclidean distance be appropriate here? Is there a better distance measure to use?
If you consider the item as a point on 25D scale then Euclidean would be the choice, but if you deal with the scores of different scales as a vector representing the semantic of the item (I prefer this way), then using cosine similarity is the most matching calculation from my opinion.
Hi Amr, thanks so much for the reply! Would you mind elaborating on the difference between the two possibilities you mentioned?
It depends on your items. In my case the items were pieces of text represented as points in a 1000-dimensional semantic space. I used Euclidiean distance. The key problem was how to assign reasonable values to the coordinates so that the distance would work. In 1000-dimensional space it turned out to be rather hard, because most of the coordinates were zeros.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts