You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hello @albernar ! I'm not the authors but I can try to provide an answer.
The reason this is done, I believe, is for better learning and data augmentation. If you look at Table 7, you see that the model was able to help predict the query <Florence, vacationer, D.C. Henrie> by seeing the inverse relationship of <Florence, contain^-1, Italy>. Without the inverses, the model could not follow this path as the original query was probably <Italy, contain, Florance>. Thus, starting at Florence as the u node, the model would not have followed this path because it couldn't have.
Activity
BraydenJamesE commentedon May 17, 2026
Hello @albernar ! I'm not the authors but I can try to provide an answer.
The reason this is done, I believe, is for better learning and data augmentation. If you look at Table 7, you see that the model was able to help predict the query <Florence, vacationer, D.C. Henrie> by seeing the inverse relationship of <Florence, contain^-1, Italy>. Without the inverses, the model could not follow this path as the original query was probably <Italy, contain, Florance>. Thus, starting at Florence as the u node, the model would not have followed this path because it couldn't have.