Comprehensive 6.18 Preparation: Load Testing, Deep Personalization, and Recommendation Algorithm Optimizations

Comprehensive Preparation

Since mid‑April, the Search & Recommendation team launched the 6.18 preparation, estimating traffic, allocating data‑center resources, and establishing a detailed plan with emergency procedures and system‑exception testing. To date, more than ten rounds of pressure testing have been conducted, including single‑site and full‑cluster tests, each with clear objectives, steps, and issue verification, followed by a test report.

Load‑Testing Review Meeting

The department places great importance on the 6.18 preparation. To reduce costs, server resources were trimmed this year, and extensive system optimizations were made to support a component‑based codebase. The solutions are already live on the main site, Thailand site, and Indonesia site. Successful completion requires precise traffic forecasting, proactive risk control at every stage, contingency planning for extreme scenarios, and meticulous execution. Numerous late‑night testing sessions saw team members on‑site to quickly locate and resolve issues.

Issues identified during pressure tests are tracked through weekly and site‑level meetings, with clear owners. After resolution, each issue is retested in the next round to confirm it does not recur. The team actively participates in upstream and downstream preparation meetings, synchronizing required support from other departments and responding to their needs.

Full Application of Deep Personalization Technology

Deep learning, a rapidly emerging technology, has achieved great success in image and speech domains. A team of former Google and Facebook senior experts from Silicon Valley applied deep learning to JD.com’s e‑commerce search and recommendation scenario. By mining massive user behavior data, the model learns high‑dimensional representations of user interests and matches them with products, delivering personalized search results. After deployment, key metrics such as UV value, conversion rate, and click‑through rate have all improved, and the deep personalization system is now fully in production for both search and recommendation pipelines, contributing to the 6.18 campaign.

Recommendation Algorithm Optimization

The 6.18 recommendation algorithm received several enhancements: a purchased‑item filter across all homepage positions, a unified large‑scale interface that increases category diversity, cross‑category recommendations, price‑based gating for certain event venues to encourage brand discounting, and support for the “Guide Cube” algorithm to provide richer business units and multi‑dimensional user coverage.

Thoughtful Logistics Support

To keep night‑shift testers energized, the department set up a dedicated “fuel station” and organized a series of morale‑boosting activities—May sports season, “empty‑head energy bullets,” team showcase, and warm night‑snack sessions. Team‑building events such as tug‑of‑war and forest‑run competitions highlighted unity and perseverance.