Verfahren und Systeme zur Identifizierung von Beziehungen zwischen Entitäten

Details

Veröffentlichungs-Nr.

EP4811208

Anmeldetag

21. März 2025

Veröffentlichung

23. September 2026

Rechtsraum

EP

IPC

G06F40/295, G06F40/35

Abstract

A computer-implemented method of identifying one or more pairs of related entities in a source text, the method comprising: obtaining a set of one or more occurrences of a first entity in the source text and identifying a set of candidate entity pairings. Identifying the set of candidate entity pairings comprises: for each occurrence of the first entity in the set, identifying in the source text one or more occurrences of a further entity, which is different to the first entity, which are nearest to the respective occurrence of the first entity in the source text; and identifying a candidate entity pairing as comprising the respective occurrence of the first entity and one of the identified one or more occurrences of a further entity. The method further comprises, for each candidate entity pairing: determining a distance in the source text between the respective occurrence of the first entity and the respective occurrence of the further entity so that each candidate entity pairing has an associated distance value; carrying out a first filtering stage, wherein the first filtering stage comprises filtering the set of candidate entity pairings based on the associated distance values for the respective candidate entity pairings to obtain a shortlist of candidate entity pairings; and carrying out a second filtering stage, wherein the second filtering stage comprises filtering the shortlist of candidate entity pairings based on one or more semantic characteristics and/or one or more structural characteristics associated with at least a portion of the source text to obtain a final list of one or more pairs of related entities.

Anmelder

Firma

British Telecommunications Public Limited Company

Land

🇬🇧 Großbritannien

Fachgebiete