Deep Learning Ranking Model Enhancements for Recruitment Search at 58.com
This report details how the Search Recommendation team at 58.com upgraded their deep learning ranking model for recruitment by adding multi-valued and semantic vector features, integrating conversion sequences, employing feature‑crossing techniques, optimizing offline data pipelines, and planning future multi‑scene improvements to boost CTR and relevance.
Background : The Search Recommendation department of 58.com provides core search services; after successfully deploying the DIEN deep learning ranking model in the rental business, the team extended it to the recruitment scenario.
Model Feature Mining : New multi‑valued and semantic vector features were introduced, updating the DIEN architecture and achieving a 3‑4 point AUC lift. Long‑tail query issues were mitigated via semantic recall and an automatic daily model‑update strategy. Conversion sequences were added to the interest‑evolving layer, and the multi‑task loss was refined using ESMM principles. Feature crossing was enhanced by integrating the CAN model, further improving offline and online performance.
Offline Process Optimization : The data production pipeline was rebuilt for parallel sequence generation, and three versions of semantic vector ingestion were iterated to drastically cut request volume. Training data were converted to compressed TFRecord format, and TensorFlow dataset pipelines were optimized with prefetch and padded_batch, reducing training time and resource consumption.
Future Work : The team plans to extend optimizations to all recruitment sub‑scenes, explore STAR topology for cross‑scene knowledge sharing, and continue improving relevance through query understanding, query rewriting, and semantic retrieval, aiming for a single model serving all scenarios.
Author & Department : Bai Bo, Senior Algorithm Engineer, 58.com TEG Search Ranking Department, focusing on vertical search ranking optimization.
58 Tech
Official tech channel of 58, a platform for tech innovation, sharing, and communication.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.