This study develops a general procedure for forming a well-defined collection of documents on a research topic, which is suitable for bibliometric analysis. The procedure is applied to one research topic in Library and Information Science, Bradford's Law of Scattering. An analysis is made of the underlying concepts, viz. 'document', 'selection', 'collection comprehensiveness', 'topic', 'research topic', and 'on' and 'relevant to' a research topic. An important distinction is drawn between graphical and semantic attributes of documents, and between their modes of analysis. The central problem of the study is that, while 'topic' is a problematic semantic attribute of documents, a well-defined collection of documents requires a selection criterion formulated on unproblematic graphical attributes. The solution proposed is to let specialist A&I Services legitimate a research topic and to provide a sample of its documents; then to extract a diagnostic graphical pattern from the sample, and from this construct a criterion which can be mechanically applied to all documents. Modifications introduced into the general procedure include the iterative development of the criterion from the growing collection, and allowing a content analysis of documents to suggest diagnostic patterns in the text. The graphical selection criterion developed for the specific collection was composed of six alternative pairs of word-stems separated by at most two words in the body of the text of documents. It has an estimated precision of 96% against the background literature and retrieves 90% of all known documents which might be judged to be on the topic. The final collection consists of 1187 analytical-level and scholarly documents written in 19 languages; it is well-defined, accords with convention, and is judged to be near completion for documents more strongly on and influential in the topic. Inadequacies in the procedure and in the collection formed were examined and improvements have been suggested. For example, the mixing of semantic and graphical methods in the pattern extraction process is clarified, recall can be enhanced by the addition of several small sub-collections, and measures of topic aboutness and topic influence were installed in the collection.

This study develops a general procedure for forming a well-defined collection of documents on a research topic, which is suitable for bibliometric analysis. The procedure is applied to one research topic in Library and Information Science, Bradford's Law of Scattering. An analysis is made of the underlying concepts, viz. 'document', 'selection', 'collection comprehensiveness', 'topic', 'research topic', and 'on' and 'relevant to' a research topic. An important distinction is drawn between graphical and semantic attributes of documents, and between their modes of analysis. The central problem of the study is that, while 'topic' is a problematic semantic attribute of documents, a well-defined collection of documents requires a selection criterion formulated on unproblematic graphical attributes. The solution proposed is to let specialist A&I Services legitimate a research topic and to provide a sample of its documents; then to extract a diagnostic graphical pattern from the sample, and from this construct a criterion which can be mechanically applied to all documents. Modifications introduced into the general procedure include the iterative development of the criterion from the growing collection, and allowing a content analysis of documents to suggest diagnostic patterns in the text. The graphical selection criterion developed for the specific collection was composed of six alternative pairs of word-stems separated by at most two words in the body of the text of documents. It has an estimated precision of 96% against the background literature and retrieves 90% of all known documents which might be judged to be on the topic. The final collection consists of 1187 analytical-level and scholarly documents written in 19 languages; it is well-defined, accords with convention, and is judged to be near completion for documents more strongly on and influential in the topic. Inadequacies in the procedure and in the collection formed were examined and improvements have been suggested. For example, the mixing of semantic and graphical methods in the pattern extraction process is clarified, recall can be enhanced by the addition of several small sub-collections, and measures of topic aboutness and topic influence were installed in the collection.