Ant Security Lab Wins Two Golds and One Silver at KDD Cup 2022 with Advanced Keyword Extraction and Self‑Distillation for Product Search

Ant Security Lab algorithm engineer Lin Jinzheng achieved "two golds and one silver" at the KDD Cup 2022, ranking first among 1,699 global participants.

KDD Cup, organized by ACM SIGKDD, is the premier international data‑mining competition that attracts top enterprises, universities, and research institutions each year, focusing on solving real‑world problems.

The "product search" track, described by Amazon, challenges participants to match relevant products to highly difficult search queries, a core e‑commerce technology hindered by massive product catalogs, missing information, and noisy query inputs.

The track comprises three tasks—product ranking for a given query, relevance classification, and substitute product identification. Leveraging years of expertise in image recognition and natural language processing, Ant Security Lab enhanced the model's query‑understanding capability, winning two championships.

To address the common issue of overly short search texts, the lab extracted keywords from related product descriptions and significantly expanded query features; additionally, self‑distillation was employed to mitigate noise, markedly improving model robustness.

The same keyword‑extraction and self‑distillation technologies have been deployed at scale in Ant's interactive risk‑control models, playing a crucial role in extracting fraud‑ and gambling‑related keywords from Alipay's trillion‑transaction data.

In interactive risk control, AI must converse with potentially defrauded users, capture keywords from noisy, ambiguous utterances, determine the fraud type, and retrieve precise deterrent scripts from a corpus to encourage user compliance.

Currently, Ant's interactive risk‑control product engages 71,000 users daily, with average conversation durations exceeding 90 seconds and an 80% increase in fraud‑stop rates.

In May, the International Telecommunication Union (ITU) officially launched the first international standard for interactive intelligent risk control, led by Ant Group.