Verfahren und System zur Sprachmodell-Gestützten Antwort-Erzeugung unter Zuhilfenahme Externer Datenquellen

Details

Veröffentlichungs-Nr.

EP4738145

Anmeldetag

31. Oktober 2024

Veröffentlichung

6. Mai 2026

Rechtsraum

EP

Abstract

An extended retrieval-augmented generation system stores documents in a vector database (ME-VDB), wherein for at least one text chunk of each document, an embedding model (EM) computes a vector embedding (VE), and wherein both are stored in the vector database. Metadata of each document are included in the vector database as well. An application (EA) then receives a query from a user (U). The embedding model (EM) receives the query as input and forms a query embedding (QE). The vector database receives the query embedding as input and identifies documents matching the query embedding. The application retrieves the matching documents and their metadata. Next, the application sends at least one prompt containing the query and the retrieved documents to a large language model (LLM). Later, the application receives a response from the large language model. A metadata score generator computes at least one score for the retrieved metadata. In addition or as an alternative, a metadata summary generator (MSMG) sends at least one prompt containing the retrieved metadata to the large language model or to another large language model. The metadata summary generator then receives a summary of the retrieved metadata. Finally, the application outputs the response together with the at least one score and/or the summary to the user. The system supports the user in quickly assessing quality/trustworthiness of the LLM response by providing a score/summary based on the metadata (e.g. authors, publication date, rating) of the retrieved documents. The system can offer the user different orthogonal scores and/or summaries of the metadata of the retrieved documents, e.g. both the recency of retrieved documents and the diversity of the author affiliations, as opposed to RAG systems which choose or filter their data sources in advance.

Anmelder

Firma

Siemens Aktiengesellschaft

Land

🇩🇪 Deutschland