HITS (Hubs and Authorities): Topic-Scoped Authority in Search
Counting links picks the wrong page. HITS gives each page a hub and an authority score that define each other, computed per query, so authority is measured per topic.
Rocchio Relevance Feedback: Vector Space Query Expansion in Search
The Rocchio algorithm optimizes search queries in vector space by adjusting the query vector to shift toward relevant documents and away from non-relevant ones, resolving vocabulary mismatch.
Latest
Locality-sensitive hashing: near-duplicate detection at scale with MinHash and SimHash
SimHash and MinHash compress high-dimensional document features into compact signatures that preserve similarity, letting search engines detect near-duplicates at web scale in sub-linear time.
PrefixSpan and GSP: finding frequent paths in search sessions
GSP and PrefixSpan both find ordered behavior that appears often enough in a database of sequences. GSP generates candidate patterns and counts them; PrefixSpan projects the database to the suffixes after a prefix and grows patterns inside those smaller databases.
Finding communities in a graph: Label Propagation, Louvain, and Leiden
Three algorithms split a graph into communities using only local moves, with no preset number of groups. Search systems use them: one in a web-ranking patent, another in language-model retrieval.