Holistic Evaluation of Travel Blogger Campaign ROI and Attribution Models

A system-level, reproducible framework for evaluating ROI and attribution across multi-channel travel marketing campaigns.

Overview

Travel marketing operates across distributed channels (blogs, social, newsletters, affiliates). Traditional attribution (especially last-click) compresses multi-stage journeys into single events, leading to misallocation of value.

This repository provides:

• A Signal-level research paper (paper/paper.md)
• A production-grade dataset generator (data/generator.py)
• Attribution models: last-click, linear multi-touch, time-decay, hybrid (models/)
• A reproducible experiment pipeline (experiments/)
• Comparable outputs (results/*.json, results/comparison.csv)
• Publication-ready visuals (results/figures/)
• Citation + DOI-ready metadata

Core Idea

Attribution is a system, not a report. The goal is to represent how influence is distributed across time, channels, and interactions.

The hybrid model combines:

• Time decay (recency)
• Channel weights (context)
• Interaction weights (intent)

Structure

.
├── paper/
├── data/
├── models/
├── experiments/
├── results/
│   └── figures/
├── methodology/
├── CITATION.cff
├── .zenodo.json
└── README.md

Quick Start

pip install pandas matplotlib numpy
python data/generator.py --out data/campaign.csv
python experiments/run_all.py --input data/campaign.csv
python experiments/make_charts.py

Open the notebook:

jupyter notebook experiments/analysis.ipynb

Outputs

results/
  last_click.json
  multi_touch.json
  time_decay.json
  hybrid.json
  comparison.csv
  figures/
    last_click.png
    multi_touch.png
    time_decay.png
    hybrid.png
    comparison.png

Example Comparison

Channel	last_click	multi_touch	hybrid	delta
blog	2000	5500	7000	5000
instagram	1000	5200	4000	3000
newsletter	1500	5300	6000	4500
affiliate	18000	6000	9000	-9000

Reproducibility

• Dataset is generated programmatically
• Models are deterministic given a seed
• Full pipeline can be rerun end-to-end

Citation

See CITATION.cff

License

MIT