Marco BonechiFollow


Semantic matching utilizing large language models (LLMs) to convert text or images into embeddings and scoring them can outperform keyword matching in various ways by matching on meaning rather than word equality. However, semantic matching lacks explainability. This disclosure describes dual-encoder LLM techniques to confer explainability to LLM-based semantic matches within information retrieval systems. Semantic meanings are attached to abstract mathematical embeddings to generate gravitational fields that enable dynamic, high-quality information retrieval as measured by precision/recall, query-understanding, concept-matching, speed, scalability, etc. while providing justifications and user-visible corroborations of search results. Information retrieval is also improved in diversity, personalization, and efficiency, with high query throughput at low latency.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.