We are developing a POC for a text classification model. We need a developer and/or a data scientist who can finetune a BERT or similar model to predict labels/metadatas from a given span of text. The input would be a sentence or a sentence fragment. The output would be 15-30 labels, many of which will be unique. See below example. We anticipate 3,000-5,000 unique labels total, with a dataset of 30,000-40,000 instances, but we can discuss these upon hire. Given the nature of the data, the dataset will be skewed. In terms of deliverables, we'll need the model to be placed on our server with endpoints for testing/auditing. We will also need to see f1 score (or other objective measure) to show the model's performance. See attached for the sense of the type of input and output we would expect. When applying for this job, please describe any similar classification models you've developed, the results you were able to acheive, your proposed timeline and your proposed compensation.
Example input: This Services Agreement, dated as of February 13, 2024 (this “Agreement”), is made between UpWork Inc., a US company ("Upwork"), and the freelancer named below (the "Freelancer").
Example output: [Clause-Loc: (Preamble)], [DocTitle: Services Agreement], [EntityDocDate: February 13, 2024], [DefDocTitle: Agreement], [ChunkPartyDesc: UpWork Inc., a US company ("Upwork")], [EntityPartyA: UpWork Inc.], [PartyAJur: US], [PartyATyp: company], [DefPartyA: UpWork], [ChunkPartyDesc: the freelancer named below (the "Freelancer")], [EntityPartyB: freelancer named below], [PartyBJur: na], [PartyBTyp: na], [DefPartyB: Freelancer]
This job is already closed and no longer accepting applicants, sorry.
