A Proof of Concept for Semantic Search and Intelligent Recommendations in the Domain Name Industry
Researchers from the University of Donja Gorica (UDG), in collaboration with industry (Alicorn), have developed and evaluated an AI-based approach for recommending premium internet domain names. The research, published in IEEE Access, demonstrates how large language models (LLMs) and semantic search can support more effective recommendations from large collections of existing domain names, with potential applications in commercial domain marketplaces.
CHALLENGE
Recommending premium domain names presents a specific challenge: suitable names must be selected from an existing inventory rather than generated freely. Large inventories, limited semantic information contained in domain names, and the need to balance relevance, diversity, and computational cost make conventional search and generative AI approaches challenging for this task.
SOLUTION
The developed Proof of Concept combines LLM-based semantic expansion, vector embeddings, FAISS similarity search, and different reranking strategies to identify and recommend relevant premium domain names.
The approach was evaluated using a sample of 30,000 domain names from a collection of nearly 700,000, with assessments involving 22 domain industry experts and 180 students. Different retrieval and reranking methods were compared, including cosine similarity, maximal marginal relevance (MMR), and six LLM-based approaches.
The research also has direct industrial relevance for Alicorn, a Montenegro-based software company developing AI-powered solutions for the domain name industry, including its Name Sentinel platform.
BENEFITS
- Improved semantic search: Identification of relevant premium domains based on user intent rather than keyword matching alone.
- Relevance and diversity optimization: Comparative evaluation of recommendation strategies identified trade-offs between relevance and variety, providing evidence for selecting appropriate ranking methods.
- Validated retrieval performance: The tested semantic retrieval configuration achieved an nDCG@10 of 0.791 and Precision@10 of 0.940 in the first-stage candidate evaluation.
- Practical inventory search: Retrieval-first selection of existing premium domains, avoiding the low inventory match rates and associated costs of unrestricted name generation.
- Evidence-based model selection: Comparative evaluation of retrieval and LLM-based reranking methods, including computational cost considerations.
- Industry innovation potential: Opportunities to enhance commercial domain recommendation platforms using experimentally evaluated AI methods.
HPC Montenegro Support and Next Steps
The research was supported in part by the EuroCC2 and EuroCC4SEE projects, funded through the EuroHPC Joint Undertaking, and contributes to HPC Montenegro’s efforts to strengthen AI expertise and collaboration between academia and industry.
Future development could explore the use of HPC resources for processing larger domain inventories, parallel evaluation of AI models, and testing open-source embedding and language models to assess opportunities for improved scalability and cost efficiency.
Scientific publication: Premium Domain Name Recommendation via Semantic Retrieval and Reranking of Fixed Aftermarket Inventories – IEEE Access