Presented By O’Reilly and Intel AI
Put AI to Work
April 29-30, 2018: Training
April 30-May 2, 2018: Tutorials & Conference
New York, NY

From answering questions to questioning answers: Challenges of large-scale QnA systems

Mridu Narang (Microsoft)
2:35pm–3:15pm Wednesday, May 2, 2018
Implementing AI
Location: Grand Ballroom East
Average rating: ***..
(3.00, 3 ratings)

Who is this presentation for?

  • Machine learning practitioners, applied scientists, and software engineers

Prerequisite knowledge

  • A basic understanding of machine learning and question-answering system concepts

What you'll learn

  • Understand the challenges faced by large-scale open-domain question-answering systems and approaches to solve them


In a world of information overload and manipulation, knowledge acquisition techniques are expected to provide instant, precise, and succinct answers. Question-answering (QnA) systems must serve answers with high accuracy and be backed by strong verification techniques. These systems extract and rank best answer candidates from the web to answer questions automatically and differ from crowd-sourced forum type systems in that respect. The kind of queries in such systems are not restricted to particular topics (i.e., open domain) and can cover a diverse range of topics, from sports to politics to science, making them very hard to generalize for. Users’ engagement and interaction with such systems is much less tolerant compared to showing top 10 blue links or recommendations. As a result, accuracy is key for such systems, requiring innovative, creative, and strong verification techniques.

Mridu Narang offers an overview of the challenges of and approaches taken by these large-scale open-domain question-answering systems that generate answers from the web in response to user queries (whether search engine queries or conversational questions over personalized agents). Mridu begins by outlining the typical architecture designs of large-scale open-domain question answering systems before diving into the challenges these systems face to generate answers for user queries. Mridu then walks you through approaches to solve them. You’ll also learn the pitfalls to watch for in such systems.

Photo of Mridu Narang

Mridu Narang


Mridu Narang is a senior engineer at Microsoft, where she builds foundational algorithms and scalable machine learning systems focused on solutions for natural language question-answering systems. In her time at Microsoft, Mridu has contributed on entity linking, temporal fact extraction, and photosynth projects.

Comments on this page are now closed.


Albert Kim |
05/18/2018 7:21am EDT

Can you please make your presentation available?