BEYOND COSINE SIMILARITY

This repository contains the official implementation code for reproducing the experiments in the paper:

BEYOND COSINE SIMILARITY
ArXiv Link: https://arxiv.org/pdf/2602.05266

To reproduce the results reported in the paper, please follow the steps below.

1. Environment Setup

Create and activate a virtual environment:

bash
python -m venv recos_env
source recos_env/bin/activate

2. Install Dependencies

Install all required Python packages:

pip install -r requirements.txt

3. Compute Embeddings

Create a subdirectory named embeddings and run the provided Jupyter notebook to generate the necessary embeddings.

mkdir embeddings
jupyter notebook get_embeddings.ipynb

The computed embeddings will be saved in the embeddings directory.

4. Compute Similarities

Ensure the supplementary Python script recos.py is in your working directory. Then, create a perf subdirectory and run the similarity computation notebook.

mkdir perf
jupyter notebook get_similarities.ipynb

This process will save all similarity scores in the perf directory. The final aggregated performance results are stored in perf/all_performance.csv.

---

To reproduce the statistical test results, run the statistical_test.R script in R version 4.5.2 after installing the required packages: rstatix 0.7.3, lmerTest 3.2.0, and effsize 0.8.1.