[FEATURE] Absolute-Timestamp Splitting

[lthoang]

Description

A data-splitting strategy to use specific absolute timestamps to cut item sequences for training and evaluation, respectively. This strategy aligns with real-world scenarios but is not widely used in research. Researchers are encouraged to experiment with this splitting strategy.

Expected behavior with the suggested feature

Specially, given a chronological user interaction sequence of length N:

• Training part: item interactions with timestamp range (-∞, t_1);
• Validation part: item interactions with timestamp range [t_1, t_2);
• Testing part: item interactions with timestamp range [t_2, +∞).

Reference: https://dl.acm.org/doi/pdf/10.1145/3569930

Other Comments

The Amazon Reviews 2023 dataset has already been split here: https://amazon-reviews-2023.github.io/data_processing/5core.html#absolute-timestamp-splitting

[qtuantruong]

This is good and we should have it for UIRT format. We could do something like:

• Add a new eval method TimestampSplit for UIRT data only.
• Accept two absolute cutoffs, val_timestamp and test_timestamp, with semantics:
  • train: timestamp < val_timestamp
  • validation: val_timestamp <= timestamp < test_timestamp
  • test: timestamp >= test_timestamp
• Reuse the existing BaseMethod build/evaluate flow. The only new logic should be how raw UIRT rows are split.
• Validate inputs clearly: require both cutoffs, require val_timestamp < test_timestamp, and fail if train or test ends up empty.