Technical Deep Dive of JD’s Intelligent Customer Service 2.0: AI‑Driven Intent Recognition, Emotion Analysis, and Smart Scheduling

JD’s Technology Gold Award recognizes outstanding R&D teams; the Intelligent Customer Service 2.0 project is highlighted as a case study of AI‑driven innovation.

To overcome the limitations of early chatbots, the team introduced the ABSQ (Action/Business/Scene/Question) multi‑dimensional classification, enabling finer intent detection and reducing category overlap.

At the algorithm level, a Hierarchical Attention Network‑based Sentence Attention Hybrid Networks model was built, combining sentence‑level attention with conversation‑level context to improve intent accuracy by 5‑10% across scenarios.

Emotion analysis was added using a convolutional neural network that identifies seven user emotions (anger, anxiety, worry, loss, confusion, happiness, calm) with high precision and the ability to gauge intensity, leveraging transfer learning and back‑translation data augmentation.

For voice channels, ASR and NLP technologies were integrated via the MRCP protocol, enabling voice navigation and real‑time quality inspection; the system reduces call handling time by half and doubles processing speed.

Smart dispatch combines user intent, emotion, and agent skill profiling using a collaborative semi‑supervised regression model and low‑rank matrix completion to predict agent‑intent match scores, raising overall service capacity by 15%.

Future directions include self‑learning, reinforcement learning, multimodal fusion, and unsupervised training to further enhance accuracy and scalability.