Negative Finder

NLP course project. In a team of five people, we implemented a solution for negation and uncertainty detection in medical documents. The dataset included medical exams in Catalan and Spanish. We implemented three algorithms with different approaches: rule-based with regex, machine learning with CRF, and deep learning with LSTMs. Additionally, we preprocessed the initial data. We successfully trained models and achieved good results, with the best average score of 75%.

- Collaborated in a 5-person team to develop models for negation and uncertainty detection in Catalan and Spanish medical documents
- Implemented and compared three approaches:
- Rule-based (regex),
- Machine Learning (Conditional Random Fields), and
- Deep Learning (LSTM network)
- Preprocessed raw clinical text data and optimized feature extraction for token-level classification
- Achieved a 75% F1 score with the CRF model — the best-performing approach among all tested methods

Used: spacy, nltk, scikit-learn, pytorch, Numpy