Design and evaluate document retrieval using lexical matching, vector similarity and hybrid ranking. A fixed fictional collection supports worked examples of relevance judgements, recall and precision. The book assumes programming experience and basic linear algebra, and separates retrieval evaluation from answer quality.
Book ID
book-016
Reading level
Advanced
Currency
GBP
Category
technology
searchretrievaldataevaluation
This is a fictional book in a demonstration store. It is not available to buy.
Reader reviews
6 fictional reviews · 3.7 / 5
★★★★★5 out of 5
Separates finding from answering
The distinction between retrieval quality and answer quality is handled carefully. Using one fixed collection makes the lexical and vector comparisons easy to follow.
★★★★☆4 out of 5
Detailed evaluation chapters
The relevance judgements and recall examples are excellent. The linear algebra arrives quickly, so I needed to revise a few concepts before continuing.
★☆☆☆☆1 out of 5
Far beyond a first search project
I wanted a quick guide to adding search to a small site. This is a detailed technical treatment with mathematical prerequisites I do not have.
★★★☆☆3 out of 5
Good theory, demanding implementation
The ranking explanations are clear when read slowly. Translating the examples into a working project took me longer than the surrounding prose suggested.
★★★★★5 out of 5
Helpful comparison of approaches
I appreciated cases where lexical matching wins instead of treating vector search as an automatic upgrade. The hybrid examples make the trade-offs concrete.
★★★★☆4 out of 5
Would like more error analysis
The fixed collection and worked metrics are useful. A longer inspection of individual irrelevant results would complement the careful attention to aggregate scores.