Project for Computational Linguistics

An analysis of idiom sentence similarities for different transformer models

This project investigates on cosine similarities of sentence embeddings. The whole dataset can be found in data/sentences.txt
In model_experiments.py the code for generating all datasets can be found. It includes model-loading from the huggingface library, similarity calculations and different score extractions. Exports are saved in data_exports/
evaluations.py was used to generate plots from that data obtained in model_experiments.py. Plots are saved in evaluation_exports/