Review Relevance @ Thumbtack

Thumbtack is a Sequoia backed online marketplace, connecting millions of customers to professionals who can get the tasks done. Customers need to know that a pro is accountable, reliable, timely, easy to communicate with and safe to bring into their home. Customers also want to know whether a professional has successfully completed similar projects in the past. In this talk we will dive into how we leverage the millions of reviews for professionals on Thumbtack, to help our customers find the most relevant professionals for their job. We will talk about how we represent review text, and train and evaluate models using traditional machine learning and deep learning techniques using Scikit, Tensorflow etc. for various review relevance problems like review re-ranking, review snippet extraction, review keywords extraction etc., and dive specifically into the problem of review snippet extraction which involves a set of challenges like review re-ranking, snippet extraction and term highlighting. We will also cover aspects around how we productized our approach using Apache Airflow. We will also dive into the challenges we currently face around review relevance and how we are evolving our process within a startup like Thumbtack so as to be able to scale our efforts. We hope to provide useful insights for other machine learning practitioners working on similar text representation and modeling problems in the industry especially at other startups. The talk would be intended for a technical audience with at least a beginner level background in data science & machine learning.