fPLSA: Learning Semantic Structures in Document Collections Using Foundation Models