58.com AI Algorithm Competition: Award Ceremony, Top Teams, and Solution Sharing

58.com’s core recruitment service includes a recommendation system where increasing job click‑through and application rates is critical. To foster innovation, the 58 Technology Committee organized an AI algorithm competition, providing real‑world recruitment data for participants to predict user browsing and application probabilities using machine‑learning methods.

The competition ran for 53 days, attracting more than 210 teams from over 60 universities and 80 companies, with 280+ participants in total, and culminated in a live award ceremony on September 16, 2021.

During the ceremony, 58’s Vice President and Technical Committee Chair delivered opening remarks, followed by presentations from the top‑three teams and a senior engineer from 58 AI Lab who demonstrated a deep‑learning solution achieving a score of 0.83.

Team solutions included:

Third place: tag_pooling method that counts how often a job appears in other users' click sequences to capture popularity, alongside item_embedding and embedding_pooling using Word2Vec.

Second place: extensive feature engineering with 89 sequence features, multi‑value tag processing via Word2Vec, pooling, and CountVectorizer, and a five‑fold cross‑validation ensemble using LightGBM.

First place: multi‑channel feature generation (continuous, discrete, decision‑tree, sample‑augmentation, BERT‑based, and graph‑embedding channels), tree models with focal loss, PLE enhanced by ESMM, attention‑based gating, and Bayesian hyper‑parameter optimization, achieving a best AUC of 0.792089.

The organizers also provided the 58 AI Algorithm Platform (WPAI) with GPU resources for model training and evaluation, and released videos and PPTs of the award ceremony and solution talks.

Participants are invited to continue submitting results to the open leaderboard, explore the upcoming NLP‑focused competition, and contact the 58 technology assistant for further resources.