In this seminar, we introduce an efficient method for monitoring specific phenomena within a text corpus through the use of natural language processing (NLP) techniques. Monitoring phenomena is crucial in disciplines such as sociology, psychology, and economics, which analyze human behaviour within society. Our method diverges from existing strategies that depend on universally applied large language models (LLMs), known for their high computational costs. Instead, we employ less resource-intensive techniques that still achieve rapid data processing and maintain high accuracy. Our strategy is inspired by the cascade optimization approach, initially filtering out irrelevant information roughly before progressively employing more precise, albeit more computationally demanding, models. This allows us to handle large datasets accurately and cost-effectively by leveraging a variety of models.