Build ML classifier to generate subject tags on 200M academic papers - Freelance Job in Machine Learning - Less than 30 hrs/week - undefined - Upwork
ab1630's bookmarks 2021-10-06
"Here's the project: We have a stream of data about newly-published research papers (about 50k/day). For each paper we have standard bibliographic metadata including title, authors, and journal. In some cases we also have subject tags from the MESH vocabulary . We want to run each paper through a classifier that will tag it with one or more predifined Fields Of Study. So for example, a paper called "The role of mosquitoes in malaria transmission" might be tagged with Biology, Medicine, Malaria, and Mosquitoes. Your job is to create the classifier that will do the tagging. You would select features, train the model, assess the accuracy, and deliver the model and Python code we need to run it in our production environment. We have a good gold standard for training: the Microsoft Academic Graph (MAG) , which includes 200M papers tagged with 700,000 tags. We want to reproduce the results of their classifier as closely as possible, because they are discontinuing their service, and we are going to replace it ...."