#VoxxedDaysCERN26

Search technology has evolved from matching words to understanding meaning. In this talk, we’ll unravel how this transformation happened, from the early days of sparse vector models like TF-IDF and BM25, which relied purely on keyword overlap, to today’s dense vector embeddings powered by neural networks and transformers. We’ll explore how semantic search captures context, relationships, and intent, allowing machines to “understand” language rather than just count words.

Using intuitive examples and visual demonstrations, we’ll demystify how embeddings represent meaning in high-dimensional space, how similarity is computed, and why this shift is revolutionizing information retrieval, recommendation, and question answering.

Attendees will leave with a clear conceptual map of how sparse and dense approaches differ, and how they can work together to build smarter, more intuitive search systems.

Devoxx

A beginner-level introduction to semantic search, contrasting it with lexical (keyword-based) search. Covers the limitations of lexical search such as lack of context awareness, vocabulary mismatch, and poor handling of long queries. Explains how semantic search works by transforming documents and queries into vector embeddings using machine learning models, then comparing them in a vector space to find the most similar results. Introduces key concepts including dense vectors, k-nearest neighbors (KNN), approximate nearest neighbors (ANN), and similarity metrics like cosine similarity and Euclidean distance. Concludes that semantic and lexical search are complementary, and hybrid approaches are also viable.

A Beginner’s Guide to Semantic Search by Pietro Mele