AI and Deep Learning for Highway Freight Matching – Insights from QCon 2018

Manbang Group, China’s largest truck‑goods matching platform, introduced its logistics background and the need for efficient vehicle‑cargo matching in the QCon 2018 AI and Deep Learning session.

The speaker explained the scale of highway freight, the high cost of transportation, and how most of a product’s price is actually freight cost, emphasizing the importance of reducing empty‑load rates.

Vehicle‑cargo matching differs from passenger ride‑hailing because goods vary in size, weight, and handling requirements (e.g., live pigs vs. pork), creating a complex B2B scenario with thousands of cargo categories.

Key features used for matching include driver/app data, GPS tags, and massive vehicle telemetry (hundreds of terabytes per day). The platform measures feedback rate, transaction rate, and other KPIs to evaluate matching efficiency.

To solve the problem, the team builds a recommendation system constrained by vehicle length, weight, route, and cargo type, then scores each feasible driver‑cargo pair using a deep‑learning model.

The neural network transforms cargo and driver features into high‑dimensional vectors, concatenates them, and passes them through several fully‑connected layers (typically 5–6 layers, with hidden sizes chosen as multiples of 256 for GPU efficiency). Training is performed offline per driver and updated online for real‑time scoring.

For new drivers lacking historical behavior, rule‑based statistical models (e.g., vehicle length + region) provide initial recommendations, though accuracy may be lower.

The system also addresses load optimization, zero‑load reduction, and future extensions such as standardized pallets with RFID tags to better estimate remaining capacity.

During the Q&A, the speaker clarified that only a small subset of driver‑cargo pairs are used for each training step, discussed handling of massive data volumes, and highlighted environmental benefits such as reduced fuel consumption and carbon emissions.