In an era of scarce resources, positive or negative social constructions of groups attract growing scholar attention as an important condition for obtaining policy benefits. Yet, measuring these perceptions over time, across contexts and for various types of groups, is still a methodological challenge. The media plays a pivotal role in shaping societal perceptions and constructing narratives that influence the collective understanding of social groups. This paper introduces a new project that seeks to detect specific social group mentions based on an innovative approach that leverages a multilingual supervised BERT model trained on French and German newspaper archives spanning a period from 1995 to 2023. In the first step, the model is trained to identify media mentions of social groups in general, following recent advancements in the field. In the second step, we fine-tune another model to identify mentions of specific, prominent social groups within the sentences detected in the first phase. Our measures are validated through precision and recall tests conducted on manually coded data, complemented by existing and newly collected survey data on group deservingness. Preliminary findings demonstrate the potential of this approach for tracking social constructions of various groups across time and contexts.