An intelligent, real-time Query Autocompletion Framework using GPT-2

We are thrilled to announce that the research paper (IntelliQSense: An intelligent, real-time Query Autocompletion Framework using GPT-2), authored by Crux Intelligence Data Scientists Taaniya Arora, Shashank Srivastava & Neha Prabhugaonkar, has been published in Lattice, the international peer-reviewed machine learning journal, hosted and managed by the Association of Data Scientists (ADaSci).

It is a major milestone in extending the NLP-based data insight capabilities of our adaptive question answering system, ASK. 

The Crux Intelligence Advantage

By enabling business users to ask questions in natural language about enterprise data, Crux Intelligence makes Business Intelligence accessible to every business user. Business leaders can make informed decisions with Crux Intelligence by quickly understanding the “why” behind a variety of business issues. 

The research paper proposes a unique approach to Query Autocompletion (QAC), specifically designed for the ASK engine within our Augmented Analytics Platform, where queries are often complex business questions, framed in natural language. Query Auto Completion is a feature that helps complete a user’s partially typed query in many text-based applications.  

The research work combines semantic search and natural language generation to build an intelligent framework for question autocompletion. It uses GPT2 language model, fine-tuned on custom data to generate autocomplete suggestions in natural language. The approach was evaluated using datasets from multiple business domains and proven effective and efficient. Moreover, it demonstrated its domain-agnostic capabilities. In addition, optimization techniques were incorporated to make the framework real-time  

With QAC, we aim to simplify the user experience of gaining valuable insights from enterprise data by: 

• Boosting productivity and saving time through reducing the need to manually type out full queries 
• Elevating the user experience with more accurate insights through the provision of intuitive and context-aware autocomplete suggestions 
• Helping users explore and learn to frame questions supported on ASK 

This is a novel approach to autocompletion in the research area of query autocompletion as it focuses on autocompletion of questions in natural language with domain agnostic capabilities for Question Answering system in our Augmented Analytics Platform, in contrast to conventional approaches to autocompletion which are only suitable for short search queries, typically used by search engines, social media and e-commerce platforms.