SkillBridge: Canadian Labour-Market Skill Recommender

Data mining team project, University of Victoria (2026). Team of 6.

SkillBridge recommends the skills that matter for a given occupation in Canada, and predicts labour shortages, salary and regional demand. It is built on five national data sources merged into one modelling base.

Results at a glance

Component	Task	Result
Skill recommender (BPR)	Rank the core skills for each occupation	nDCG@5 0.617, MRR 0.821, ROC-AUC 0.99
Shortage classifier	Predict which occupations face labour shortages	weighted F1 0.76
Salary and demand regressor	Predict salary and regional demand	MAE $12,088, R² 0.64 (target was 0.4)
NLP skill extractor	Map free-text job postings onto the skill taxonomy	micro-F1 0.37

• The recommender was benchmarked as a five-model ladder against real baselines. BPR won.
• Fairness was audited across TEER skill levels.
• Weak spots are reported openly. The NLP extractor is the weakest component and is documented as such.

Data

Five sources merged into one base:

• 900 occupation competency profiles (181 descriptors each)
• Four months of Job Bank postings
• LinkedIn Canada postings and skills
• Employment projections
• The NOC occupation taxonomy

The ETL step removed 51,103 duplicate records.

The key decision

The first framing, link prediction on occupation-skill pairs, was broken: 162,899 of 162,900 cells were already observed, so the right answer was always "yes". We caught this before modelling and reframed the task as ordinal matrix completion, which gives a meaningful binary target with a 14.1% positive rate.

Repository structure

SkillBridge/
├── datasets/
│   ├── raw/
│   └── clean/
├── figures/        charts used in the report
├── results/        model outputs and metrics
├── scripts/        pipeline and experiment scripts
├── skillbridge/    core package
├── webapp/         demo web app
├── README.md
└── requirements.txt

Setup

The dataset is too large for the repo and is distributed as a release file.

1. Download data.zip from the latest release.
2. Extract it into the project root.
3. Move the raw/ and processed/ folders into a datasets/ folder, and rename processed/ to clean/.
4. Install dependencies:

python3 -m venv .venv
source .venv/bin/activate          # Windows: .venv\Scripts\activate
pip install -r requirements.txt

Credits

Forked from aarij13406/SkillBridge, the team's original repository.