Teaching / CSCI 4141
Summer 2026 · Dalhousie UniversityInformation Retrieval
How does a system find the right document, image, product, or answer? This course builds search from first principles, then follows the same ideas into vector databases, recommender systems, retrieval-augmented generation, and agent memory.
What you will learn
Turn text, images, queries, and users into representations that a retrieval system can compare.
Build ranking functions and select metrics that reveal whether search results are genuinely useful.
Use inverted indexes, approximate nearest-neighbor search, and vector databases.
Connect retrieval to recommendation, language models, RAG pipelines, agents, and memory.
The course journey
Weekly syllabus
Introduction to IR; Boolean search
Vector space and bag-of-words models; ranked search; development environment tutorial
Indexing, web ranking, and evaluation metrics
Machine-learning foundations; probabilistic retrieval; PyTorch and image-processing tutorial
Image-to-image retrieval; approximate nearest-neighbor search
Word embeddings, query expansion, query-performance prediction, and vector databases
Search personalization and midterm synthesis
Collaborative filtering and matrix factorization
Linear and neural recommender systems
Generative recommendation, prompt engineering, and pretrained language models
Retrieval-augmented generation; agentic AI and memory systems
Model Context Protocol; integration and review
What you will build
Construct an inverted index and retrieve ranked documents.
Search visual embeddings using a vector database.
Estimate preferences from patterns across users and items.
Ground language-model responses in retrieved documents.
Assessment at a glance
The course uses four individual implementation assignments, a midterm and final exam, and active participation. It is designed as an applied machine-learning course: students should leave able to explain an IR method, implement it, and evaluate whether it works.