Production-grade Sequential Recommender System & NLP Pipeline for behavioral candidate job navigation and action classification (
viewvsapply), developed for the ENS Challenge Data 164 (HrFlow.ai).
- 🏆 Official ENS Public Leaderboard: challengedata.ens.fr/participants/challenges/164/ranking/public
- User Participant Account:
ayanmitra(Official Rank 15)
This solution was submitted directly to the École Normale Supérieure (ENS) Challenge Data Platform under account ayanmitra.
- Official Submitted Public Score:
0.3144 - Official Competition Rank: Rank 15 on Official ENS Leaderboard out of all registered participants!
- Action Classification Accuracy:
84.07%(viewvsapply) - Top-10 Retrieval Accuracy:
18.15%(Out of 21,917 distinct candidate job postings)
HrFlow.ai powers talent data orchestration for global enterprise platforms. When candidates search for their ideal job, their journey follows a sequential path across job listings based on evolving preferences.
-
Top-10 Sequential Job Recommendation: Predict an ordered top-10 list of candidate job IDs
$[j_1, j_2, \dots, j_{10}]$ that a candidate will open next, out of 21,917 candidate job listings. -
Action Type Classification: Classify whether the candidate's next action will be an
applyor aview.
Predicting whether a candidate will submit an application or simply view a job based on navigation recency and apply frequency.
Out-of-fold weighted competition score progression across 5-fold session-based splits.
┌────────────────────────────────────────────────────────┐
│ Candidate Session Sequences & Job Text JSON │
└───────────────────────────┬────────────────────────────┘
│
▼
┌────────────────────────────────────────────────────────┐
│ Dual-Branch Feature Engine │
│ ┌───────────────────────┐ ┌──────────────────────┐ │
│ │ NLP Text Encoder │ │ Sequential Graph / │ │
│ │ TF-IDF Vectorizer │ │ Co-occurrence Matrix│ │
│ │ Job Title/Summary │ │ Markov Transition P │ │
│ └───────────────────────┘ └──────────────────────┘ │
└───────────────────────────┬────────────────────────────┘
│
▼
┌────────────────────────────────────────────────────────┐
│ Multi-Stage Candidate Retrieval & Reranking │
│ 1. Candidate Generation: Top-100 via Item-CF + NLP Sim │
│ 2. Action Classification: GBDT (LightGBM) │
│ 3. Reranking: Weighted Scoring Combination │
└───────────────────────────┬────────────────────────────┘
│
▼
┌────────────────────────────────────────────────────────┐
│ Validated Submission (submission.csv) │
└────────────────────────────────────────────────────────┘
-
Sequential Markov Transition Matrix ($P(j_{t+1} | j_t)$):
- Models direct state transitions between consecutive job interactions across 15,882 candidate sessions.
-
Item Co-Occurrence Graph & Popularity Prior:
- Captures global session item co-occurrences and smooths cold-start recommendations with global apply popularity.
-
TF-IDF Semantic Text Encoder:
- Vectorizes 21,917 job title and summary descriptions using 5,000-dimensional sublinear TF-IDF features and computes session-wide cosine similarity matching.
-
GBDT Action Classifier:
- LightGBM model predicting
$P(\text{action} = \text{'apply'})$ with 84.07% out-of-fold accuracy.
- LightGBM model predicting
HrFlow_project/
├── data/ # Raw CSVs & job_listings.json
├── assets/ # Generated visual charts & figures
│ ├── leaderboard_comparison.png # Official leaderboard rank comparison
│ ├── action_confusion_matrix.png # Action classifier confusion matrix
│ └── cv_folds_score.png # 5-fold CV score progression
├── src/ # Modular core Python library
│ ├── data.py # Session sequence & JSON text parser
│ ├── nlp_engine.py # TF-IDF vectorizer & cosine similarity matcher
│ ├── recommender.py # Markov transition matrix & candidate retriever
│ ├── models.py # LightGBM action classifier
│ └── utils.py # MRR@10, Top-10 Acc, & submission validator
├── train_and_evaluate.py # End-to-end training & CV evaluation runner
├── submit_solution.py # Automated ENS Challenge 164 submission script
├── requirements.txt # Project dependencies
└── README.md # Project documentation
git clone https://github.com/am610/HrFlow_job_recommendation.git
cd HrFlow_job_recommendation
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txtpython train_and_evaluate.pypython submit_solution.pyThis project is licensed under the MIT License - see the LICENSE file for details.
- HrFlow.ai for providing the behavioral job recommendation dataset.
- ENS Challenge Data platform for hosting the competition.