Meso-level retrieval: IR-bibliometrics interplay and hybrid citation-words methods in scientific fields delineation

In this position paper, we comment on various approaches to the delineation of scientific fields or domains, a typical prerequisite for a wide class of bibliometric studies. There is growing evidence that this meso-level, between micro targets of typical IR and large disciplines handled by macro-level bibliometric studies, takes full advantage of hybrid approaches. Firstly, delineation tasks gain to combine the a priori thinking of traditional IR, which typically involves clearly targeted expectations, and the a posteriori thinking of bibliometric mapping, where the decisions are built on external structuring of the domain in a wider context. The combination of the two ways of thought is far from new, with IR increasingly building on bibliometric networks for query expansion, and bibliometrics building on IR for evaluating and refining its outcomes. Secondly, delineation benefits from the multi-network perspective, which gives different representations of the scientific topics, usually all the more converging than the objects are dense and well separated. Focusing on two basic networks--words and citations--various sequences or combinations of operations are discussed. Bibliometrics and IR, especially when properly combined in multi-network approaches, provide an efficient toolbox for studies of domains delimitation. It should be recalled however that the context of such studies is often loaded with policy stakes that ask for cautious supervision and consultation processes.