Research in science is usually built uponcomplex background knowledge and assumptions, making it difficult to organizeand overview. We propose using question network to dynamically maintain scientific knowledge, with each nodes beingeither a question or an answer, linked with relations such as specification, contrast and so on. Publications can then be fittedinto nodes of the network. By constructing example networks around cognitive concepts, we observed a big question (e.g.What is curiosity?) being answered with theoretical speculation initially, then specified into the operationalized definition(How to measure curiosity as a personality?) and computational algorithms. Similar patterns are repeated in differentbranches of the network. We also compare research topics starting with similar questions yet develop differently.