Enhancements for IR Metrics Evaluation, Data Visualization, and Reranking Functionality

[Matth-info]

Pull Request Description

This pull request introduces several significant enhancements and new features aimed at improving the evaluation and visualization of Information Retrieval (IR) metrics. The key updates include:

1. Experiments

Impact of Test Dataset Size and Diversity on IR Metrics

• Conducted experiments to assess how varying the size and diversity of the test dataset affects IR metrics.
• These experiments are crucial for understanding the robustness and reliability of IR systems under different dataset conditions.

2. Data Visualization

Visualization Function for Experiment Results

• Added a new visualization function to plot the results of the experiments.
• This function provides clear and interpretable visualizations to better understand the impact of test dataset variations on IR metrics.
• It supports various types of plots to represent the experimental data effectively.

3. Reranking

Reranking Function Based on Log Probability of LLM Logits

• Introduced a reranking function designed for evaluation purposes (not for production use).
• This function utilizes the log probabilities of LLM (Large Language Model) logits as a pseudo judge to evaluate the relevancy of documents.
• Inspired by the Perplexity metric, this approach offers a novel way to assess and improve the relevancy of retrieved documents in IR systems.