smartThesaurus

Using NLP techniques to improve word relevance results.

In natural language modeling, word embeddings are popular way to capture the semantic similarity between words. However, one drawback they have as a result of their generalized vector representation of a given word, is that they don't capture the local context around it, that may be present at times. We use a context-aware biLSTM model with adaptive word embeddings to improve the word relevance results, of a given target word.

Contributors:

• Aniruddh Nautiyal
• Jae II Lee
• Nathaniel Schub

Directories:

• ./Code: development jupyter notebooks, code (python) and analysis
• ./Report: final project report and presentation slides
• ./datasets: datasets used for evaluating the models
• ./materials: research papers for background as well as materials informing the choice(s) of approaches used
• ./pretrained: links to (big) training datasets or models used