CAISA Lab

Akbar Karimi

Postdoctoral Researcher

Simply put, I’m interested in training computers to understand the human language. The word “understand” covers a wide range of meanings. One example could be for a computer to tell us if a sentence that a person says contains a positive sentiment or a negative one. Or when we ask a computer a question and it gives us the answer. Or a more complex one is when a computer can engage in a conversation with a human being without the person noticing that they’re talking with a machine. The applications are numerous.

During my PhD, I studied several subfields of NLP, namely Aspect-Based Sentiment Analysis (ABSA), Aspect-Based Emotion Analysis (ABEA), and text classification. To address these tasks, I used state-of-the-art language models with various other deep learning methods such as convolutional and recurrent neural networks. In addition, I worked on a novel data augmentation technique that deals with the data sparsity problem in deep learning.

Recently, I have shifted focus from general text processing to medical text processing where there are many significant challenges. For instance, extracting medical entities such as medications and symptoms is one that can help reduce healthcare costs as well as human errors if done automatically. However, models that perform this and similar tasks can still make errors or be vulnerable to adversarial attacks. Therefore, another challenge is to make them more robust so that we can see more reliable outcomes from these models.

Areas of Interest

  - Interpretability and analysis of models for NLP 
  - Large Language Models (LLMs)
  - Machine Learning for NLP
  - Ethics and NLP