Research talk: Is phrase retrieval all we need?
- Danqi Chen | Princeton University
- Microsoft Research Summit 2021 | The Future of Search & Recommendation
DensePhrases is an extractive phrase-search tool based on natural language input that achieves dense retrieval of billion-scale phrases with extreme runtime efficiency. In this talk, Assistant Professor Danqi Chen of Princeton University will highlight some of the technical challenges that she and her research team encountered and the solutions of learning dense representations of phrases at scale. She’ll demonstrate the strong performance on open-domain QA and slot-filling tasks, and she’ll show how phrase retrieval, the most fine-grained retrieval unit, can also be used for passage or document retrieval tasks. Finally, she’ll cover how phrase filtering and vector quantization can make the phrase index much smaller, making dense phrase retrieval a practical and versatile solution in multi-granularity retrieval.
Learn more about the 2021 Microsoft Research Summit: https://Aka.ms/researchsummit (opens in new tab)
The Future of Search & Recommendation
-
-
Keynote: Universal search and recommendation
- Paul Bennett
-
-
-
-
-
-
-
-
-
Research talk: System frontiers for dense retrieval
- Jason Li,
- Knut Risvik
-
-
-
-
-
Research talk: Domain-specific pretraining for vertical search
- Tristan Naumann
-
Panel: The future of search and recommendation: Beyond web search
- Eric Horvitz,
- Nitin Agrawal,
- Soumen Chakrabati
-
-
-
-
Panel: Causality in search and recommendation systems
- Emre Kiciman,
- Amit Sharma,
- Dean Eckles
-