Ready-to-use FAQ-matching solution even when there’s little to no data
Based on my team’s experience building an FAQ-matching solution for 3 different public health helplines, we present a suite of FAQ-matching models useful even in cold-start or unsupervised settings. In some cases, we can pre-trained or custom word embeddings while allowing for user-defined domain-specific contextualisation. In other cases, we just need to provide just several representative questions for each FAQ, and train a match-scoring layer on top of powerful pre-trained models like BERT [1,2,3]. We present an open-source library called FAQT that implements these models and Flask application templates to start an FAQ-matching service.
Suzin You is a Data Scientist at IDinsight, based in New Delhi, India. Other than the usual data science work, she spends a lot of her time thinking about how the team can work better and how we can build data science solutions responsibly.
visit the speaker at: Homepage