This project is founded by ANR which stands for Agence Nationale de la Recherche for humanitarian goals. Its main objective is to investigate whether the situations in which migrants are, are suitable or not, by analyzing a set of images. The project is multifaceted, with one aspect focused on extracting objects and background contexts from images using advanced deep-learning techniques such as panoptic segmentation. Once objects are extracted, various approaches can be employed for further processing. One notable method involves constructing a co-occurrence graph where nodes represent objects, and edges denote the frequency of object occurrences across pairs of images. This graph analysis allows us to identify significant objects within the dataset by analyzing their frequencies and degrees, potentially augmented by employing algorithms for community detection, including Graph Neural Networks. Another facet of the project involves processing textual data, leveraging similar methodologies. Specifically, a text co-occurrence graph can be constructed to elucidate relationships among textual elements. Moreover, integrating the object co-occurrence graph with the text co-occurrence graph offers a comprehensive understanding of the relationships between text, objects, and images concurrently.