Capstone: A Mini Search Engine

Capstone: A Mini Search Engine

Index and score

Second-order scoring

What you built, and what it still misses

A search engine serves millions of queries against millions of documents. What does the inverted index change about scoring a query, compared with scanning every document for every query?

BM25 uses the idf weight $\ln\frac13.54.5 + 1$. In the capstone corpus $N = 17$ documents, the term crossing appears in $df = 4$ of them. What is its idf, to three decimal places?

Three queries each return a ranked top 3. The first query has 3 relevant documents in its top 3, the second has 2, and the third has 3. What is the mean P@3 over the three queries, to three decimal places?

In the capstone, the query cheap tickets and the document about discount fares share no words, yet second-order scoring ranks it inside the top 3. What connects them?

A theatre review containing storm scene keeps displacing a genuine storm-delay notice in every stage of the capstone, lexical, semantic and hybrid. What kind of evidence would actually separate them?