On the Inverse Relationship of Recall and Precision

It is now ten years since some slight experimental evidence was presented which appeared to support the hypothesis that there was an inverse relationship between recall and precision. The idea of this was certainly not new; Fairthorne had more than implied it in his discussions on OBNA and ABNO systems, i.e. Only‐But‐Not‐All (high precision) and All‐But‐Not‐Only (high recall). However, it was one of the propositions arising from Cranfield I which met with strong opposition and was quite rightly attacked. In reply to the critical review by Swanson, I had to agree that the simple hypothesis required modification. By the following year test results coming from the experiments by Salton and from Cranfield II made further modification necessary, and the hypothesis was finally put forward to read as follows: ‘Within a single system, assuming that a sequence of sub‐searches for a particular question is made in the logical order of expected decreasing precision, and the requirements are those stated in the question, there is an inverse relationship between recall and precision, if the results of a number of different searches are averaged. This, it will be noted, has four important qualifications to the basic statement.