Byte-token Enhanced Language Models for Temporal Point Processes Analysis

Quyu Kong, Yixuan Zhang, Yang Liu, Panrong Tong, Enqi Liu, Feng Zhou · 2025

Read Byte-token Enhanced Language Models for Temporal Point Processes Analysis by Quyu Kong, Yixuan Zhang, Yang Liu, Panrong Tong, Enqi Liu, Feng Zhou (2025) 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-2502-07139.txt.

Description

Quyu Kong and co-authors fuse language models with event timing models in this 2025 paper. Temporal point processes describe streams of timed events, like posts, reviews, and purchases, but classic versions ignore the text attached to each event. Language models read text well yet have no sense of timing. Their framework, Language-TPP, joins the two.

The key trick turns each time gap into special byte tokens that a standard language model can read directly. No custom machinery is needed inside the model. Across web data sets, the system set new marks for predicting event times and types. Adding timing also improved the text it generated, with better summary scores and sentiment match. Code is public for other teams.

Catalog record

CREATORQuyu Kong, Yixuan Zhang, Yang Liu, Panrong Tong, Enqi Liu, Feng Zhou
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE52 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-07139

Browse the full library · About the Tower of Babel