FWPRec Notes
Experiments
Not a run

Important things to note

Mixed results on real data. Report the losses. Do not treat one run as proof.

Read this first

Status

Mixed. Tested on real data. Reported as-is. Not a clean win.

Win

Beats other models on Amazon — explicit ratings. Strongest evidence.

Loss

Loses to standard models on MIND (news) and KuaiRand (video). Do not claim overall superiority.

Efficiency

No history re-read. True. Still uses more memory and compute than a tuned GRU. Not a clean efficiency win either.

Promising

Splitting stable taste (anchor) from fast recent mood (memory) looks real — especially a fixed anchor on MIND.

Caveat

That fixed-anchor MIND result is one run. More seeds before it counts as solid.

Rules

No cherry-picking. No mixing incompatible setups. No reusing the test set to keep tuning. No treating one-off runs as proof.

In order

Next steps

Do these in this order. Then update the paper.

  1. Confirm the fixed-anchor result
  2. Align training with real-world use
  3. Recheck smaller, compact versions
  4. Test on a dataset with real explicit intent
  5. Redo the efficiency test on Amazon
  6. Update the paper

Baselines

Models we compare against. Hybrids check whether combining ideas helps.

Simple

StaticProfile

Average of past likes. No updating.

Tracker

PositiveEMA / SignedEMA

Recent-average trackers. Newer events count more.

Recurrent

GRU4Rec

Classic recurrent sequence model.

Recurrent

LRURec

Another recurrent-style sequence model.

Transformer

SASRecProfile

Attention-based sequence model.

Transformer

HSTU

Newer, stronger Transformer-style model.

State space

Mamba4Rec

Newer state-space sequence model.

Hybrid

SASRecSignedEMA / SASRecFWP

Transformer plus EMA, or plus this paper’s fast memory. Tests whether combining helps.

Datasets

What each corpus is for. Side checks and synthetic data do not support claims.

Main

MIND

Real news clicks. Main test.

Check

EB-NeRD

Another real news set. Double-check only. Not a final test.

Strongest

Amazon Reviews

Video Games ratings. Explicit likes and dislikes. Strongest result.

Main

KuaiRand

Real short-video watches.

Side

KuaiRec

Related video set. Backup only. Not for main claims.

Sanity

MovieLens

Old movie ratings. Code check only. Not for real claims.

Never

Synthetic

Made-up data. Testing only. Never use it to support a claim.

This page is the guardrail. The argument is on The idea. Layout is on the method page. Scores live on the experiment pages.