CauSeR

Gupta, Priyanka; Sharma, Ankit; Malhotra, Pankaj; Vig, Lovekesh; Shroff, Gautam

doi:10.1145/3459637.3482071

Cited by 22 publications

(13 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…NISER [12] adds a normalized regularization on item embedding and session embedding. As for CauSeR [13], it adopts docalculus [28] to eliminate bad causal effects. Though, these works perform well…”

Section: Popularity Biasmentioning

confidence: 99%

“…• NISER [12]: imposes L2 regularization on item embedding and session embedding respectively. • CauSeR [13]: adopts do-calculus [28] to eliminate bad causal effects of momentum in SGD optimizer, which is biased towards the head items.…”

Section: Baselines and Backbonesmentioning

confidence: 99%

“…Though these works relieve popularity bias for SBR well, they ignore individual's conformity towards popular items and thus decrease recommending performance on popular items. As Figure 1 shows, NISER [12] and CauSeR [13] elevate recommending performance on long-tail item group, but the performance on 50-100 and 100+ item group degrades compared with SR-GNN. Besides, these methods are designed for SR-GNN only, but we aim to explore a general method for more SBR models.…”

Section: Introductionmentioning

confidence: 98%

“…They elevate recommending performance on tail items by various techniques, such as re-weighting [9] and causality [10]. Nevertheless, as far as we know, only three works [11][12][13] focus on popularity bias in SBR. Liu et.al.…”

Section: Introductionmentioning

confidence: 99%

“…[12] normalizes item and session embedding for SR-GNN [6] to promote recommending performance of tail items. Another most recent work [13] utilizes causality technique to eliminate popularity bias for SR-GNN. Though these works relieve popularity bias for SBR well, they ignore individual's conformity towards popular items and thus decrease recommending performance on popular items.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Disentangling interest and conformity for eliminating popularity bias in session-based recommendation

et al. 2023

View full text Add to dashboard Cite

Session-Based Recommendation (SBR) is to predict next item, given an anonymous interaction sequence. Recently, many advanced SBR models show great recommending performance, but few studies note that they suffer from popularity bias seriously: the model tends to recommend popular items and fails to recommend longtail items. The only few debias works relieve popularity bias indeed. However, they ignore individual's conformity towards popular items and thus decrease recommending performance on popular items. Besides, conformity is always entangled with individual's real interest, which hinders extracting one's comprehensive preference. To tackle the problem, we propose a SBR framework with Disentangling InteRest And Conformity (DIRAC) for eliminating popularity bias in SBR. In this framework, two group of item encoders and session modeling modules are devised to extract interest and conformity respectively, and a fusion module is designed to combine these two types of preference. Also, a discrepancy loss is utilized to disentangle representation of interest and conformity. Besides, our devised framework can integrate with several SBR models seamlessly. We conduct extensive experiments on two real-world datasets with three advanced SBR models. The results show that our framework outperforms other state-of-art debias methods consistently. Disentangling Interest and Conformity for... Fig. 1 An illustration of popularity bias in SBR. Items are grouped by their popularity in Diginetica dataset, which is commonly used in SBR experiment. Popularity is defined as the number of interaction of one item in dataset. The vertical axis represents mean reciprocal rank of SR-GNN and corresponding debias methods. The histogram shows the number of items in each item group.

show abstract

Section: Popularity Biasmentioning

confidence: 99%