Automatic Conversion from MARC to FRBR

摘要

We applied our prototype system to a set of 4379 BIBSYS records that are related to Henrik Ibsen. These records often describe complex works, containing multiple expressions and/or parts of other works. We chose them because singlecelular works are often easy to convert and studies "[...] suggest that the majority of benefits associated with applying FRBR [...] could be obtained by concentrating on a relatively small number of complex works." [8]. When putting the system into a strict mode that relies on the certain identification of the original title, we identified 48 works by Henrik Ibsen. Of these works 16 were falsely identified because of varying or erroneous spelling, 10 were collections and translations that the system failed to recognize as such. These works could be clustered into 1111 expressions, which are contained in 1072 manifestations, 35 of which contained more than one expression. But due to the strict setting of the rules 3307 records where ignored, because it could not be determined, whether they contain an original title or not. Without this knowledge it is impossible to reliably identify the work cluster. With non-strict settings our system identified 580 works by Ibsen and 3706 expressions in 3567 manifestations, clearly no satisfying result. The next steps are to apply a more fault tolerant dissimilarity measure to the clustering process, and to use authority files in the attribute layer in oder to cope with spelling variations and errors. In addition, we will try to leverage the reliable information found in high quality records to process incomplete records. It is planned to apply the system to a 100.000 records large subset of the BIBSYS catalog and to evaluate the results.