Disentangling Long and Short-Term Interests for Recommendation

Zheng, Yu; Gao, Chen; Chang, Jianxin; Niu, Yifan; Song, Yang; Jin, Depeng; Li, Yong

doi:10.1145/3485447.3512098

Cited by 66 publications

(19 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We take the mean representation of short-term behaviors as the proxy of short-term interests. As for long-term interests, we argue that directly mean representation of entire long-term behaviors applied by Zheng [32] will lead to sub-optimal performance. Long-term user preference is relatively stable and updates slowly, if we mean representations simply, the temporal variability of long-term interests would be ignored.…”

Section: Interests Disentanglement Modulementioning

confidence: 97%

“…Desired by Zheng [32] who introduces a self-supervised framework for disentanglement, we adapt it for product search. We take the mean representation of short-term behaviors as the proxy of short-term interests.…”

Section: Interests Disentanglement Modulementioning

confidence: 99%

“…The effectiveness of long-term interests proxy updating. Although Zheng [32] proposes a simple strategy for proxy setting to disentangling user interests, we argue that this strategy ignores the nature that long-term interests will evolve slowly, so we propose an updating strategy for long-term interests proxy to address the issue. In this section, we study the effectiveness of the long-term interests proxy updating strategy on the four datasets with respect to different metrics.…”

Section: 82mentioning

confidence: 99%

“…Next, all the representation features are concatenated with original raw context features and fed into fully connected layers to generate the final prediction results. Last but not least, motivated by DLSR [32], IDM injects pseudo labels for both long and short-term interest and employs an self-supervised framework to disentangle long and short-term interests. But different from DLSR, we devise an updating strategy to set long-term interest proxy, which is more suitable and achieves better performance than the settings in DLSR.The contribution of this paper is three-fold:…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Hierarchically Fusing Long and Short-Term User Interests for Click-Through Rate Prediction in Product Search

Shen

Wen

Zhang

et al. 2022

Proceedings of the 31st ACM International Conference on Information &Amp; Knowledge Management

View full text Add to dashboard Cite

Estimating Click-Through Rate (CTR) is a vital yet challenging task in personalized product search. However, existing CTR methods still struggle in the product search settings due to the following three challenges including how to more effectively extract users' short-term interests with respect to multiple aspects, how to extract and fuse users' long-term interest with short-term interests, how to address the entangling characteristic of long and short-term interests. To resolve these challenges, in this paper, we propose a new approach named Hierarchical Interests Fusing Network (HIFN), which consists of four basic modules namely Short-term Interests Extractor (SIE), Long-term Interests Extractor (LIE), Interests Fusion Module (IFM) and Interests Disentanglement Module (IDM). Specifically, SIE is proposed to extract user's short-term interests by integrating three fundamental interests encoders within it namely query-dependent, target-dependent and causal-dependent interest encoder, respectively, followed by delivering the resultant representation to the module LIE, where it can effectively capture user longterm interests by devising an attention mechanism with respect to the short-term interests from SIE module. In IFM, the achieved long and short-term interests are further fused in an adaptive manner, followed by concatenating it with original raw context features for the final prediction result. Last but not least, considering the entangling characteristic of long and short-term interests, IDM further devises a self-supervised framework to disentangle long and short-term interests. Extensive offline and online evaluations on a real-world e-commerce platform demonstrate the superiority of HIFN over state-of-the-art methods.

show abstract

Section: Interests Disentanglement Modulementioning

confidence: 97%

Section: Interests Disentanglement Modulementioning

confidence: 99%

Section: 82mentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Hierarchically Fusing Long and Short-Term User Interests for Click-Through Rate Prediction in Product Search

Shen

Wen

Zhang

et al. 2022

Proceedings of the 31st ACM International Conference on Information &Amp; Knowledge Management

View full text Add to dashboard Cite

show abstract

“…Hence 𝑆 2 -DHCN [44] conducts contrastive learning between representations of different hyper-graphs without random dropout, but its fixed ground truths restrict the improvements. CSLR [53] generates self-supervised signals via long-term and short-term interest but fails to further consider more effective signals. Besides, two works [44,49] about social recommendation [36] and session-based recommendation [5,12,13,26] address this problem by the self-supervised co-training framework.…”

Section: Self-supervised Learning In Sequential Recommendationmentioning

confidence: 99%