JourneyFormer: Encoding Airbnb Guest Journey with Sequence Modeling¶

Fidelity: core mechanism. Paper results, local public-data measurements, and omissions are separate.

Paper information¶

Field	Details
Paper	arXiv 2606.19108
Organization	Airbnb
First public release	2026-06-17 (arXiv v1)
Authors' code	The paper did not provide official or author code when checked on 2026-08-24.
Adapter / method	`journeyformer`
Local reproduction code	`src/auto_research/reproductions/journeyformer/`

Original paper¶

Motivation and method¶

Encode Airbnb's long- and short-term guest journey with event type, time, and listing representations. Caching and sparse computation meet serving latency.

flowchart LR
 A["Long- and short-term journey events"] --> B["Journey Transformer"] --> C["Listing ranking"]

Key figure from the paper¶

Figure 2: Overview of the paper's method and key components. From the original paper; copyright remains with the authors.

Core equations¶

\[ h_u=\operatorname{Transformer}([e_i+e_{type}+e_{time}]_{i=1}^{L}) \]

Paper results¶

The main online result is bookers +0.55% in a three-week A/B test (p<0.01). Offline tables and online metrics use private or paper-specific protocols. The private paper protocol is not directly comparable with local MovieLens measurements.

Local reproduction¶

Local comparison: The baseline is the shared transition + content scorer. The treatment adds only journeyformer on the same users, items, full-catalog candidates, and seed.

MovieLens-100K, 220 users and 360 items, seed 42: NDCG@10 0.0540 → 0.0525 (-2.77%); Hit@10 0.1091 → 0.1091. Validation selects mixture weights only; the test set is not used for tuning.

auto-research reproduce --paper journeyformer --dataset-dir data --seed 42

See the fixed metrics.

Reproduction boundaries¶

The core mechanism runs on MovieLens-100K with the same split, fixed full-catalog candidates, and seed. Private features, the production foundation model, and online traffic are unavailable. The implementation has independent model state and scoring, rather than mapping a paper name to a shared weighted formula. See the adapter's omitted_core_components for omissions. Local changes and paper A/B gains are reported separately.