References and citations in automatic indexing and retrieval systems - experiments with the boomerang effect
Birger Larsen · 2004
The objective of the present dissertation work is to investigate the use of references and citations as an integrated part of automatic indexing and retrieval techniques operating on scientific full text documents. There are two main motivations behind the dissertation: Firstly, the few scattered studies from both the system-driven and the useroriented tradition that have investigated references and citations for information retrieval (IR) purposes, have generally shown promising results. Secondly, scientific documents are for the first time becoming available in large quantities in electronic form. This offers new possibilities for combining conventional automatic indexing and retrieval techniques with the exploitation of references and citations in IR. Analytical and empirical investigations are carried out with the aim of investigating which factors that affect the behaviour and performance of automatic indexing and retrieval techniques given that references and citations are an integrated part of the document representation of scientific full text documents in the IR system. It is investigated why references and citations might be useful in IR in an analytical review of the literature on citer motivations and citation behaviour. A normative and a social constructivist position on the motives of authors in citing other work is identified. The most unifying theoretical explanation is found to Small’s (1978) notion of references as ‘concept symbols’ that stand for an idea or concept that is being used in the course of an argument. This explanation is found attractive from the point of exploiting references and citations in IR, because regardless of whether or not some references are omitted, forgotten, biased etc. the references actually given function as symbols for a concept. This can explain why references and citations are useful in IR: Because the references represent concepts when used for document representation, they are well-suited for IR purposes as long as the user’s information need can be expressed in the same concepts, i.e., as seed documents. A review of earlier studies which have employed references and citations showed that the most common use of them for IR purposes was as seed documents in a forward chaining, where documents that refer to the seed document are retrieved. The ability of these seed documents to express the user’s information need was found to be the main factor affecting the behaviour and performance of references and citations in IR.