Research Assistant – sample size investigation for clinical texts classification using NLP
A team in BHI is looking for a motivated research assistant to work on a project investigating optimal sample size for Natural Language Processing (NLP) classification tasks that require manual annotations. The tasks involve:
· running simulations using deep learning models to investigate model performances in various language settings and training corpora sizes;
· developing a GitHub project page and a well-documented code;
· participating in dissemination activities (e.g. preparing and participating in an interactive online seminar).
A successful candidate can start as soon as possible and work till the end of July, in a collaboration with myself, Angus Roberts, Jaya Chaturvedi, and Daniel Stahl. The hours (14h/week) can be worked as two full days or spread over the week, remotely or in the office (Denmark Hill). Please write to diana.shamsutdinova@kcl.ac.uk to express your interest.
Project Description: Natural Language Processing methods are widely applied to extract information from clinical texts and present it in a structured way. However, unlike in statistical data analyses, there are no methods available for estimating the sample size needed. Our project aims to assess optimal sample size for the development of clinical NLP models and how these requirements change depending on the documents and language properties. By taking a simulation approach and following modern guidance on model validation, we will be able to investigate model performances in various scenarios and provide guidance on sample sizes for clinical NLP tasks.
QualificationsThe role is suitable for a current MSc/PhD student/postdoc/early career researcher in Computer science, Engineering, Health Informatics, Statistics, or a related field.
Skills· Strong analytical skills;
· Knowledge of Python programming language;
· Understanding model validation techniques such as cross-validation and evaluation metrics such as AUC-ROC/sensitivity/specificity/precision,
· Ability to work independently and in a team.
.
You can also use your social account to sign in. First you need to:
Accept Terms & Conditions and Privacy Policy.