In the Amazon paper "Supervised Clustering Loss for Clustering-Friendly Sentence Embeddings: an Application to Intent Clustering", they calculate PRAUC of the model rather differently: "using the true labels of pairs sharing the same cluster as 1 and pairs with different clusters as 0." This results in PRAUC data that is quite different from the results we have.
Week Oct 23:
Find other literature that run ood experiments using CLINC150, Massive, BANKING77, DSTC11, HWU64 dataset so we verify the baseline results we have.
In the Amazon paper "Supervised Clustering Loss for Clustering-Friendly Sentence Embeddings: an Application to Intent Clustering", they calculate PRAUC of the model rather differently: "using the true labels of pairs sharing the same cluster as 1 and pairs with different clusters as 0." This results in PRAUC data that is quite different from the results we have.
Week Oct 23: Find other literature that run ood experiments using CLINC150, Massive, BANKING77, DSTC11, HWU64 dataset so we verify the baseline results we have.