Automatic Evaluation of French Research Projects in the Acquisition Process of Research Tax Credit (CIR).

In this work, we evaluated research projects of French companies using Natural Language Processing. To this end, we designed a system able to estimate the probability of obtaining a research tax credit (CIR) for a project based on its technical description. This system is designed around two modules whose outputs are concatenated and fed to a fully-connected neural network that predicts the probability of success for the project. The first module uses the FastText algorithm and a Convolutional Neural Network to extract a Text Embedding vector. The second module uses an unsupervised knowledge graph extraction method and a Graph Neural Network to extract a Graph Embedding vector. The texts used as data in this study describe the research projects of companies and are written in French. Due to their high confidentiality, no similar examples exist in the literature. This data is provided by a partner consulting firm whose work consists in helping companies raise funds for their research projects. Since the methods in the literature were not effective in extracting the knowledge graphs used in the second module for our data, we present a new Knowledge Graph extraction using an unsupervised Named Entities Recognition, NER, as our contribution. [ABSTRACT FROM AUTHOR]