An introduction to Entity search systems

Abstract

Entity seeking queries ("marvel latest movies" or "Which Indians have won the Nobel prize?") form a significant subset of Web search queries. Such queries often trigger a special search system called "Entity search system" that returns entity responses from knowledge graph (KG).

Syntactic form of the query is a vital clue used by entity search systems to understand it. However, Web search queries are very diverse in their syntactic form, ranging from syntax-poor collection of keywords to syntax-rich natural language queries. Further, even if query interpretation is correct, question answering may still fail due to noisy and incomplete nature of the information sources.

In the first part of this talk, speaker will give a broad overview of various Entity search systems and discuss challenges that arise from query format diversity as well as imperfect information sources. In the later part, speaker will discuss her research which aims to perform better question answering by utilizing two complementary information sources - the structured knowledge graph and unstructured Web corpus. Further, she advocates joint query interpretation and answer ranking instead of performing these two tasks independently. speaker will conclude the talk with a discussion of future directions.

Speaker Bio

Uma Sawant is an Applied Research Scientist at LinkedIn, Bangalore. She is also a final year PhD candidate in the Computer Science and Engineering department of IIT Bombay. Her interests include machine learning, Web search, books and music.

Agenda

11:00 AM - 12:30 PM Talk & QA

12:30 - 01:00 PM Networking