Proceedings of the Web Conference 2021 2021
DOI: 10.1145/3442381.3449815
|View full text |Cite
|
Sign up to set email alerts
|

ECLARE: Extreme Classification with Label Graph Correlations

Abstract: Deep extreme classification (XC) seeks to train deep architectures that can tag a data point with its most relevant subset of labels from an extremely large label set. The core utility of XC comes from predicting labels that are rarely seen during training. Such rare labels hold the key to personalized recommendations that can delight and surprise a user. However, the large number of rare labels and small amount of training data per rare label offer significant statistical and computational challenges. State-o… Show more

Help me understand this report
View preprint versions

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
2
1

Citation Types

0
17
0

Year Published

2022
2022
2024
2024

Publication Types

Select...
5
3

Relationship

2
6

Authors

Journals

citations
Cited by 23 publications
(17 citation statements)
references
References 59 publications
0
17
0
Order By: Relevance
“…Therefore, new scoring functions are designed to promote infrequent label prediction by giving the model a higher "reward" when it predicts a tail label correctly. Propensity-scored P@𝑘 (PSP@𝑘) and propensity-scored NDCG@𝑘 (PSNDCG@𝑘, abbreviated to PSN@𝑘 in this paper) are thus proposed in [18] and widely used in LMTC evaluation [16,32,40,55,62]. PSP@𝑘 and PSN@𝑘 are defined as follows.…”
Section: Performance On Tail Labelsmentioning
confidence: 99%
See 2 more Smart Citations
“…Therefore, new scoring functions are designed to promote infrequent label prediction by giving the model a higher "reward" when it predicts a tail label correctly. Propensity-scored P@𝑘 (PSP@𝑘) and propensity-scored NDCG@𝑘 (PSNDCG@𝑘, abbreviated to PSN@𝑘 in this paper) are thus proposed in [18] and widely used in LMTC evaluation [16,32,40,55,62]. PSP@𝑘 and PSN@𝑘 are defined as follows.…”
Section: Performance On Tail Labelsmentioning
confidence: 99%
“…Following previously established parameter values [18,55,62], we set 𝐴 = 0.55, 𝐵 = 1.5, and 𝐶 = (log |D| − 1)(𝐵 + 1) 𝐴 . Therefore, the less frequent 6 When reporting PSP@𝑘 and PSN@𝑘, previous studies [16,32,40,55,62] normalize the original PSP@𝑘 and PSN@𝑘 scores by their maximum possible values (just like how DCG@𝑘 is normalized to NDCG@𝑘). Following these studies, we perform the same normalization in our calculation.…”
Section: Performance On Tail Labelsmentioning
confidence: 99%
See 1 more Smart Citation
“…• Propensity-scored Precision (PSP@k): Originally introduced in [20] for extreme classification scenarios [44,19,35], the PSP@k metric intuitively accounts for missing labels (items in the case of recommendation) by dividing the true relevance of an item (binary) with a propensity correction term. More formally, given recommendation lists Ŷu ⊆ I K ∀u ∈ U, the set of positive interactions I + u ⊆ I ∀u ∈ U, and a propensity model φ : I → R:…”
Section: B6 Additional Experimentsmentioning
confidence: 99%
“…Besides, collecting labeled pairs is time-consuming, expensive and sometimes infeasible, for example, launching an e-commerce system in the emerging locale, where no user behavioral signals are avaiable. In spite of these constraints, most existing methods (Dahiya et al, 2021b;You et al, 2019;Mittal et al, 2021;Dahiya et al, 2021a) followed this XMC setup. It can be seen in Figure 2 that Astec (Dahiya et al, 2021b), one of the state-of-theart extreme classifiers, is incapable of handling the scenario without supervision, which leads to zero performance in both Precision@5 and Recall@100.…”
Section: Introductionmentioning
confidence: 99%