Aleksandr Perevalov, Kirill Davydov, Thomas Fritz, Artur Mikhailov, Aleksandr Vysokov - Data Science Kooperationsprojekt

Cerence Complex Question Answering system

Projekt: Cerence Complex Question Answering system

 “Question Answering & Chatbots”; Winter Semester 2019/2020; supervised by Prof. Dr. Andreas Both

The Cerence Complex Question Answering System is intended to work in the automotive industry. Since there are many systems that are capable of answering “simple questions” - the main goal of this system is to answer non-trivial questions using such features as coreference resolution and geospatial questions. Imagine the driver asks the Question Answering system: “What I see around me” and the system gives an answer based on the driver’s geolocation.

An example of non-trivial dialog:

  • Q1: Who was the director of Django Unchained?
  • A1: The director is Quentin Tarantino.
  • Q2: Tell me more about it.
  • A2: Django Unchained is a 2012 American western film …

In this example the goal is to link "Django Unchained" from Q1 and it from Q2 to give more information about the coreference entity.The Cerence Complex Question Answering system is designed in accordance with microservice architecture and based on DBpedia data. Technologies that were used: Python, Tensorflow, HuggingFace, Docker, Gunicorn, Stardog and the Qanary framework.

Project partner: Cerence Inc., Kuldeep Singh

Project team:

  • Aleksandr Perevalov
  • Kirill Davydov
  • Thomas Fritz
  • Artur Mikhailov
  • Aleksandr Vysokov

Link to implementation: