lambdamart

Code and data for "Universal Approximation Functions for Fast Learning to Rank: Replacing Expensive Regression Forests with Simple Feed-Forward Networks"

Dataset and experiments from the CIKM 2020 Resource Track

Learning-to-Rank search engine combining BM25 retrieval with LambdaMART (LightGBM) for relevance-optimized document ranking. Achieves NDCG@10 of 0.9449 — a 6% improvement over baseline. Includes feature engineering pipeline and interactive Streamlit search UI.

UCL COMP0084 Information Retrieval and Data Mining (2023/24)

Learning-to-Rank system for e-commerce product search using LambdaMART, MiniLM, TF-IDF, and marketplace-specific relevance features.

Ranklib for .NET is an open source learning to rank library

Bachelor Thesis 2024 - Automatic Identification of Duplicate Questions in Indonesian Consumer Health Forums Using Learning-to-Rank

Production-oriented two-stage retail search system with dense retrieval, LambdaMART reranking, Airflow retraining, and FastAPI serving—improving NDCG@10 by 11.18% on a frozen Amazon ESCI test set.

Learning-to-rank search for short-term rental listings, built on public Inside Airbnb data and deployed on Azure ML.

Learning to rank for Expedia hotel search with LightGBM LambdaMART ensembles

Reciprocal ranking in a two-sided market: LambdaMART over early interaction signals, evaluated with NDCG@K and wave-based splits to prevent leakage.

A three-stage recommender under a 50 ms deadline: two-tower retrieval, HNSW search, LambdaMART reranking. Measured p99 of 20.5 ms at 50 req/s, sheds quality instead of timing out under load, catches a corrupted index in one canary cycle, corrects position bias in clicks. CPU-only, offline.

A personalised news ranking system that learns and adapts to each user's preferences through interaction feedback such as clicks, dwell-time, likes, shares, and bookmarks.

Two-stage recommendation & ranking system on Retailrocket implicit feedback using ALS candidate retrieval and LightGBM LambdaMART.

Learning-to-rank for hotel search with XGBoost LambdaMART — group-wise ranking on Expedia booking data, with PCA and undersampling for class imbalance.

Experiments in semantic relevance modeling, learning-to-rank, neural reranking, and multi-stage ranking for search.

Permission-aware hybrid search engine with retrieval-time ACL filtering, BM25, dense vector search, and learned ranking

A command line tool for training and evaluating ranking models using LightGBM and FastTree

Two-stage book recommender over 103K titles: multi-source retrieval (ALS, item-kNN, BM25, sentence embeddings) → LightGBM LambdaMART ranking. Temporal split, cold-start routing, FastAPI + Docker, GCP/Vertex scaffolding.