This describes a computer system that answers user questions. It takes a question, spoken or typed, analyzes it using techniques like morphological analysis for entity names, and then finds relevant documents by comparing the question to similar ones using word embedding. It processes potential answers from these documents through specialized question-answering units and delivers the most reliable answer to the user.
Why it matters: Filed just as large language models (LLMs) were gaining widespread capability. Many of the described steps, such as question analysis, document selection, and generating answer candidates from passages, can now be performed or significantly enhanced by modern LLMs, potentially streamlining the system's development and improving its accuracy.
AI gives you a few directions you could take this. Pick one, and we check whether your version is different enough to patent, then write the filing.
Reinvent this with AI