Xiao Zhou

Xiao Zhou

Auto-generated profile, based on publicly available information

Xiao Zhou

Meituan.com, Beijing, China

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
21 October 2023
Verified
Conference Paper
Open access
XZXiao Zhou
Xiao Zhou
RWRan Wang
Ran Wang
HLHaorui Li
Haorui Li
QLQiang Liu
Qiang Liu
XWXingxing Wang
Xingxing Wang
DWDong Wang
Dong Wang
As a local-based service (LBS), search ad retrieval in online meal delivery platforms should be broader to bridge the gap between vague consumption intentions of users and shortage of ad candidates limited by users' queries and positions. Recently, graph neural networks (GNNs) have been successfully applied to search ad retrieval task. However, directly applying GNNs suffer from noisy interactions and intents indistinguishability, which seriously degrades systems' effectiveness in the broad retrieval. In this paper, we propose a Language-EnhAnced Denoising and Intent Distinguishing graph neural network, LEAD-ID, which is developed and deployed at Meituan for sponsored search broad retrieval. To denoise interaction data, LEAD-ID designs hard- and soft- denoising strategies for GNNs based on a pretrained language model. A variational EM method is also employed to reduce high computational complexity of combining LMs and GNNs jointly. To distinguish various intents, LEAD-ID generates intent-aware node representations based on meticulously crafted LMs (language model) and GNNs; and then, it is guided by a contrastive learning object in an explicit and effective manner. According to offline experiments and online A/B tests, our framework significantly outperforms baselines in terms of recall and revenue.