In this workshop, the attendees will learn to build a virtual job interviewer with the capability of asking follow-up questions based on candidate answers.

Building a questioning agent can be useful for various applications where information gathering is the main goal. Conversational surveys, personality identification, understand individuals and help teaming in an organization, initial screening for job interviews are some example applications. Recent studies also show several benefits of chatbots for information elicitation, such as eliciting higher quality information than using traditional form-based methods. But with large scale adoption, predefined questions may eventually become repetitive and uninteresting. Dynamic and controlled question generation would be a feasible option to gather information and sustain an engaging dialogue.

We consider the use-case of an asynchronous job interview system and build a virtual interviewing agent equipped with the ability to ask follow-up questions. Using pre-trained language models such as OpenAI’s GPT-2, we will train a follow-up question generation model with the help of a small in-domain corpus. Leveraging AWS Sumerian, a toolkit and platform to build virtual reality apps, we will deploy the question generation model to a 3D avatar as the interviewing agent.

Participants will learn to build a question generation model by fine-tuning a language model.

Participants will also be exposed to the basics of AWS Sumerian and the state-of-the-art Natural Language Processing library HuggingFace Transformers.

An approximate outcome of the workshop can be as seen in this demo.

Knowledge:

• Intermediate knowledge of Python
• Basic understanding of Machine Learning

Equipment:

• Own laptop (availability of GPU is recommended)
• Google account (for Google Colab)
• AWS account (for Sumerian)