Susan Dumais

Technical Fellow at Microsoft Research & Co-Inventor of Latent Semantic Indexing

Lifespan 1954–

Technical Depth intermediate

Key Breakthrough Latent Semantic Indexing (LSI / SVD Dimensionality Reduction, 1990)

Focus Areas

information retrieval natural language processing human-computer interaction

Topic Keywords

#search-engines #information-retrieval #nlp #lsi #microsoft-research

Source: Historical Biographical Archive / Wikimedia Commons

Susan Dumais

Technical Fellow at Microsoft Research & Co-Inventor of Latent Semantic Indexing

Biographical Overview

Technical Fellow and Director at Microsoft Research. In 1990, Dumais co-invented Latent Semantic Indexing (LSI), applying singular value decomposition (SVD) to document-term matrices to discover latent semantic concepts, revolutionizing automated text retrieval and search engines.

"Words are ambiguous and people are unpredictable. Search is about discovering the underlying meaning behind the words."
— Susan Dumais

Lifespan 1954–

Technical Depth intermediate

Key Breakthrough Latent Semantic Indexing (LSI / SVD Dimensionality Reduction, 1990)

Focus Areas

information retrieval natural language processing human-computer interaction

Topic Keywords

#search-engines #information-retrieval #nlp #lsi #microsoft-research

Source: Historical Biographical Archive / Wikimedia Commons

💡

Historical Context & Impact

In short

Decades before modern transformers and dense word vectors (Word2Vec, BERT) took over artificial intelligence, Dumais mathematically proved that words could be projected into a low-dimensional semantic space where synonyms cluster together naturally.

Key Technical Breakthroughs & Inventions

Latent Semantic Indexing (LSI / LSA, 1990) Co-invented SVD-based dimensional reduction of term-document matrices, allowing search algorithms to match conceptual synonyms (e.g., "car" and "automobile").

Personalized & Contextual Desktop Search Invented the Stuff I've Seen (SIS) unified search index at Microsoft, integrating user activity history, query context, and temporal relevance into personal desktop retrieval.

Question-Answering & Web Information Architecture Developed scalable algorithms for desktop indexing, web query log modeling, and real-time interaction feedback in Microsoft Bing.

Human-Computer Interaction in Information Seeking Pioneered user-centered search evaluation metrics, bridging cognitive psychology with large-scale mathematical retrieval algorithms.