Unlimited change in scientific terminology challenges integrity in scientific knowledge graph (SKG) representation, while current data and modeling standards, mostly document oriented, hardly allow a resilient semantic upgrade of scholarly content. Moreover, results of a "multimodal knowledge acquisition" are required for an efficient upgrade of search methods: « vital nodes » differ among users of the same keyword, due to distinct needs of scientific communities, rooted in their own interpretations and controversies. Modeling and data are challenged to propose new outcomes, mixing automated information and human choices allowing dynamic community detection: to fulfill this program with GRAPHYP toolkit, we identify a workflow ensuring the objectives of integrity and completeness of search management activities. It encompasses data standards for « routes » of search, modeling of community detection and navigation inside SK bipartite hypergraph, and a first test with extraction of characteristics of communities' preferences from readings of scholarly content. "Search is not Research" and therefore further work should explore the links between modeling and data recording research contents and "search and select" results in SKG data structure. Keywords: scientific knowledge graph, routes of knowledge, bi-partite crown hypergraph, parsing of scientific publication, impact assessment and modeling of search results, graph convolutional networks.