We are looking for a Python freelancer with deep expertise in developing text/NLP analysis code (ML/Data Science/Analytics expertise not required but is a plus and could lead to additional work). We have about $2000 to spend to prototype this project, which will go from about April through May 2018. Success may lead to another $1-2000 for a polished module; and/or additional work on a related project. The ideal applicant will be a creative and persistent problem solver with an excellent work history and awesome project collaboration skills. Due to the language-centric nature of the project the applicant should have solid English language knowledge.
[Note, a job similar to this was posted a couple of months ago, and other business priorities prevented us from hiring at that time. The job description is slight updated since then. If you applied for that prior post, feel free to re-apply.]
The Task: Our R&D project needs a module that analyzes whether a sentence uses Passive vs. Active Voice (P/A), and our definition of P/A is somewhat idiosyncratic (it includes special rules for gerunds and present participles, possessives, and infinitives. Contractions should also be taken into account). The first task is to research existing best-practice and example code found on the internet for passive-vs-active language. (The project is for English language, but we will want the consultant to report on methods they ran across related to other languages, for future use.) The second main task is to modify the algorithm based on our idiosyncratic definition of Active-vs-Passive language.
The data we have to work with are about 20,000 free-text responses that can range from a few words to a couple of paragraphs. We will work with a quasi-random subset of this data. Each response needs to be separated into sentences, and probably clauses, and each clause/sentence is then analyzed for P/A classification. Human coders will be involved to determine the 'correct' answer to the P/A classification, to act as the gold standard for evaluation of the new algorithm(s). We are hoping for an eventual 95% success rate on this classification task, with a target of 85% on the prototype (using algorithmic methods, *not* machine learning).
To apply send description of relevant skills and experiences, plus a summary of what approach you would use (to as much detail as you would like). Pointers to past completed projects are encouraged. Accepted applicants will be asked to sign an NDA.
About the recuiterMember since Aug 27, 2017 Azul V.
from Jalisco, Mexico