Page 115 - AI for Good-Innovate for Impact Final Report 2024

AI for Good-Innovate for Impact

Use Case – 25: Enhancing SME Online Business through An

Automatic Recommendation Powered by LLM and Knowledge
Graph 25-Ant Group

Country: China

Organization: Ant Group CO., Ltd.

Contact person: Guanchen LIN, guanchen.lgc@ antgroup .com

25�1� Use case summary table

Domain Business, economy

The Prob- • Accelerate SMEs' ability to quickly find their customers under limited
lem to be resources.
addressed • Ease the competitive relationship between SMEs (e.g., mini-program
merchants) and large internet platforms.
• Assist SMEs in implementing recommendation systems at a low cost to
drive their revenue growth.

Key aspects of • Structured knowledge representation: Ant Personalization Engine (APE)
the solution utilizes advanced NLP technologies including LLMs to create a universal
knowledge graph, generates diverse entities from texts and logs, and
establishes complex entity relationships using Entity-bert methods.
• Multi-source heterogeneous knowledge understanding: APE devel-
ops a multi-dimensional cold-start recommendation strategy using
graph neural networks and knowledge transfer learning to address
data sparsity and sample knowledge deficiency in long-tail scenes and
private-domain scenarios, inspired by zero-shot and few-shot recommen-
dation tasks in machine learning.
• Cross-modal knowledge mining: APE establishes internal standards for
content creation, explores generative techniques including stable diffu-
sion (SD) model, LoRA, ControlNet, Roop, and iteratively enhances AIGC
quality.
• Machine Learning Operations (MLOps) Automation: APE provides SMEs
with the capabilities for automated data integration, automated model
training and deployment, as well as automated search and recommenda-
tion.
Technology NLP, LLMs, Knowledge Graph, Cold-start Recommendations, AIGC, MLOps
keywords

Data availabil- Private in Alipay
ity

Metadata (type Structured and unstructured data
of data)

99