Cross-Language Semantic Search Using OpenAI Embedding
Yoshihiro Adachi, Akito Sakai, Minoru Uehara · 2025
Embeddings generated by the OpenAI embedding model for sentences that have similar semantic information (topics) are highly similar, regardless of whether they are written in English, Chinese, or Japanese. Utilizing this phenomenon, we implemented a cross-language semantic search system using OpenAI embeddings and evaluated its search accuracy. As a result, we verified that cross-language search in this semantic search system has a high search accuracy, almost equal to that of monolingual search. We confirmed that cross-language search for queries containing the logical operators AND, OR, and NOT works well. We also compared the search accuracy of a monolingual semantic search system using Japanese embeddings generated by a simple contrastive learning framework with that of a system using OpenAI embeddings.