- A Problem in Information Retrieval and the Countermeasure
- No.4, p.1-16
Analysis of the process of documentation in article retrieval indicates that the greatest problem lies in the process of translation, followed by classification and abstract. It follows, therefore, that the use of a mechanical retrieval system would not offer any fundamental solution to the problem if the machine were to be used solely as a tool for storage and retrieval of information. Hence, for simplifying other procedure, there is an increasing tendency to use facet indexing in mechanical retrieval in place of systematic classification, which takes time.
Performance of the index in three abstract journals using different indexing systems was examined with 13,521 patents. The Uniterm Index to Chemical Patents which uses natural language index without any control on terminology had so many retrieval failure that it had low reliability, and was accompanied with much noise. Index of Photographic Abstracts, which uses systematic classification of the UDC system, and index of Abstracts of Photographic Science and Engineering Literature, which uses descriptors based on systematic classification, were found to give satisfactory retrieval results.
Consideration of a model that explains retrieval functions indicated that natural language index based on the terms collected by merely arranging synonyms, and natural language index or machine language index with functions of ‘roles and link’ were approximately of the same type as the uncontrolled natural language index. Consequently, retrieval functions of the KWIC index, which extracts index words from the title or from a short phrase that takes its place, was comprehended to be still lower.
Indexing by systematic classification gives good retrieval efficiency but requires technical knowledge for classification process and for finding indexing items at the time of retrieval, so that it is not suited for processing a large amount of information.
Indexing by descriptors determined on systematic classifications is a median form of the foregoing two methods. Although it requires time and effort to determine the descriptor terms for compiling a thesaurus, classification procedure becomes easier once the descriptors have been determined, and this might be a useful indexing for mechanical retrieval.
（Fuji Photo Film Research Laboratories, Ashigara）