Plain texts written in natural language may have several specific features, such as organizational patterns and an ambiguous and evolving vocabulary. From the literature, entity extraction approaches are not sufficient to consider these specific features jointly. To address this issue, we propose DEEP, a methodology that improves the quality of entity extraction by using organizational patterns through a sequence labelling technique. To this end, DEEP creates a high-quality corpus and relies on an appropriate learning algorithm. DEEP is validated on a real corpus of job offers. Experiments show that (1) considering organizational patterns improves the quality of entity extraction, (2) vocabulary evolution is taken into consideration and ambiguity in vocabulary is reduced, (3) DEEP provides clear guidelines for the creation of a high-quality corpus for entity extraction, (4) the Bidirectional Long Short-Term Memory + Conditional Random Field architecture for sequence labelling is the one that takes the most advantage of the organizational patterns.