CNRS

Comprehensive Database: We maintain an extensive database of scientists, including their expertise, publications, and research interests.

Input Details: Users can submit their project proposal through a text area, requiring a minimum of 50 words. Additionally, users can select the number of scientists to be displayed in the results through a dropdown menu. The user also specifies the accuracy level of the recommendations: Flexible, Moderate, or Strict. The Strict option provides highly accurate results but takes more time to generate, while the Flexible option delivers quicker results with a slight compromise on accuracy.

Advanced Matching: Our platform leverages an NLP layer to generate results based on the proposal's context rather than merely matching keywords. We use the BART-large model, an open-source model developed by Facebook AI, which is trained on the Multi-Genre Natural Language Inference (MNLI) dataset with approximately 400 million parameters. This model integrates BERT (Bidirectional Encoder Representations from Transformers) and GPT (Generative Pre-trained Transformer) architectures to better understand the context.

Project Profile Generation: The project proposal is processed to create a sample project profile vector, which represents the context and key aspects of the proposal.

Recommendation Process:

• Input Data: Scientists' features (such as expertise and past collaborations) are gathered from the database and other data mining sources.
• Building the Collaboration Network: The system constructs a collaboration network to find the most suited collaborations based on past data.
• Community Detection: Using community detection algorithms, the system forms groups based on the required group size and aggregates the features of these formed communities.
• Aggregation of Specializations: The features are aggregated to form specialized communities relevant to the project profile.
• Match Score Calculation: The match score is calculated by comparing the project profile with the scientists' profiles using pairwise cosine similarity.
• Final Output: The system generates the final output, listing the top scientists or groups of scientists best suited for the research project based on the calculated match scores.

By following these steps, our platform provides an efficient and effective way to identify the most suitable scientists for your research project, fostering impactful collaborations and accelerating scientific discovery.