Five topics are selected in respect to the volume of outputs, large or small, and three search formulations, wide, normal and narrow, are tested for each query, in order to examine retrieval efficiency and cause of search failures. It was formed that the retrieval efficiency depends on the affinity of the search query to the system, and that the distribution of the causes of search failures is reversed depending on seach patterns. The average recall and precision is 90% and 79% respectively by normal search formulations. The silence (retrieval miss) was formed at about 50% each on both keyword assigners' and search formulators' sides, but the noise can be ascribed to indexers by 50%, to retrievers by 20% and to INIS data base itself by 30%. Whole tests are analyzed individually with examples.
View full abstract