The Normalization of Occurrence and Co-occurrence Matrices in Bibliometrics using Cosine Similarities and Ochiai Coefficients

We prove that Ochiai similarity of the co-occurrence matrix is equal to cosine similarity in the underlying occurrence matrix. Neither the cosine nor the Pearson correlation should be used for the normalization of co-occurrence matrices because the similarity is then normalized twice, and therefore over-estimated; the Ochiai coefficient can be used instead. Results are shown using a small matrix (5 cases, 4 variables) for didactic reasons, and also Ahlgren et al.'s (2003) co-occurrence matrix of 24 authors in library and information sciences. The over-estimation is shown numerically and will be illustrated using multidimensional scaling and cluster dendograms. If the occurrence matrix is not available (such as in internet research or author co-citation analysis) using Ochiai for the normalization is preferable to using the cosine.