No single temporal tokenization strategy is best for all event data; performance depends on matching the tokenizer to the statistical shape of the data.
We use a per-device train batch size of 4 with 4 gradient ac- cumulation steps, resulting in an effective batch size of
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Temporal Tokenization Strategies for Event Sequence Modeling with Large Language Models
No single temporal tokenization strategy is best for all event data; performance depends on matching the tokenizer to the statistical shape of the data.