This repository contains the source code, datasets, and documentation for our Master's graduation project, which focuses on scientific expert search using deep learning. Our project was presented at TAMARICS 2022, an international conference held at the University of Tamanghasset, Algeria Check out my LinkedIn!.
Scientific expert search is a crucial task in the academic world, used for identifying researchers for supervision, evaluation, or collaboration. Our work aims to enhance expert search systems by leveraging deep learning and natural language processing (NLP) techniques, specifically transformer-based models like BERT, SciBERT, and RoBERTa.
- New indexing approach: Sentence-based indexing instead of full-document indexing.
- Query expansion: Augmenting search queries using definitions to improve accuracy.
- Score distribution adjustment: Mitigating the dominance of highly prolific authors.
- Dataset creation: A new test corpus extracted from ACM for better benchmarking for the Algerian Expert Search task.
- Optimization: Implemented FAISS indexing with CUDA acceleration for efficient similarity search..
/Project-Name
│── /Evaluation # Scripts for evaluating our methods and comparing them with previous works
│── /Results # Comprehensive collection of results described in the article (binary files and CSVs), including plots
│── /Scraping # Web scraping scripts used for dataset construction
│── /Test_&_Tutos # Testing scripts and tutorial notebooks
│── /Web_Site # Codebase for the web portal implementation
│── /Weighted_Sampling # Utility package used within the code
│── /Masters-Graduation-Project # Script for building the FAISS Index for efficient similarity search ([FAISS - AI Meta](https://ai.meta.com/tools/faiss/))
│── /main_&_execution # Backend scripts handling app connection, indexing, query augmentation, and similarity search
│── /docs # Documentation and articles
│ │── full-article.pdf # Detailed article about the project (French)
│ │── TAMARICS_2022.pdf # Summary article presented at TAMARICS 2022 (English)
│ │── slides.pdf # Presentation slides (English)
│── README.md # Overview of the project
- TAMARICS 2022 Article: A summary of our project presented at an international conference.
- Full Project Report: A comprehensive document detailing our methods and findings.
- Presentation Slides: Key insights from our work in a slide format.
Here are some key visual results from our project:
Figure 1: Performance comparison of different indexing methods.
Figure 2: Effect of query expansion on retrieval accuracy.
Figure 3: Analysis of gain in time while using CUDA (Nvidia Gtx 1660 Ti) compared to CPU (Ryzen 7 4800H).
-
Clone the repository:
git clone https://github.com/chakibMH/Masters-Graduation-Project.git cd Masters-Graduation-Project -
Run the main AI part:
python main_ranking.py
This project was developed as part of our Master’s thesis at University of Science and Technology Houari Boumediene and was supervised by [ BELLAZZOUGUI Djamal and Mr. CHAA Messaoud ]. Special thanks to TAMARICS 2022 for providing a platform to present our research.
📌 For more details, check the articles in the /docs folder.