Extracting triples from dialogues for conversational social agents

Piek Vossen, Selene Báez Santamaría, Lenka Bajčetić, Thomas Belluci · 2024

Read Extracting triples from dialogues for conversational social agents by Piek Vossen, Selene Báez Santamaría, Lenka Bajčetić, Thomas Belluci (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-18364.txt.

Description

Piek Vossen and co-authors build tools to pull fact triples out of social dialogue for chat agents in 2024. A triple states who did what to whom in a compact form. Past triple tools train on encyclopedia text. Chat is unlike that genre. It has dropped words, shared references, and yes no exchanges across turns.

The team releases data sets for training and testing on chat. Five extraction models were built and tested. On single lines, the best model got about half of full triples right. Parts of triples scored higher. Scores fell a lot for facts that span turns. The gap shows how hard true chat is for knowledge extraction.

Catalog record

CREATORPiek Vossen, Selene Báez Santamaría, Lenka Bajčetić, Thomas Belluci
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE45 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-18364

Browse the full library · About the Tower of Babel