Automated classification and cataloguing

[Gary Green]

A few years ago we (my bib team) looked at our cataloguing and classification processes and tried to work out how we could improve the processes and cut down on the human input side of things. I was one of 2 main cataloguing staff working for a large public library service and I knew that it didn't make sense to have every single record we received from our suppliers updated by hand, so that it fitted in with our local needs. We still needed to keep some local information, but we were sure that the majority of changes made by hand could be automated.

To achieve our goals we could have either asked our record suppliers to add our local information into the catalogue record before they sent it as a MARC record to us, or we could somehow manipulate the data they sent to us, so that it populated various MARC fields/subfields and tweaked others based on rules we set up.

We decided to go for the 2nd option and based on a serious of mapping rules I put together, our system supplier wrote a series of scripts so that when a MARC record was downloaded into our system by specfic suppliers it looked at the data in the record and filled out the missing part of the record.

The main areas we mapped were...

(1) Reader Interest Categories (RICs) were based on classification and cataloguing data in each record. RICs are broad subject and genre codes we've been using for shelving purposes since the 1980s. They are similar to E4Libraries codes and tie in with Dewey classification numbers.

(2) Format of stock eg Hardback book, DVD, Spoken word DVD. The format information is used to limit catalogue searches and to help with keeping track of budgets.

With the classification side of things, the rules needed to identify which supplier provided the record, the format of the stock, the Dewey number, Library of Congress subject headings, age ranges and fiction genres. Some mapping rules were fairly straight forward - if it came from supplier 'X', the format was 'Y' and the genre was 'Z' then give it a RIC of 'A'. Other rules did actually use all of the criteria. The added fun was the fact that you can decide where to classify an item in your original rules, but not everyone is going to agree. ie One persons Crime thriller is someone elses Adventure novel, depending upon how you define these genres. As all suppliers of MARC records catalogue/classify records slightly different to each other it also complicated things.

We've cut down on a significant amount of manual input, but haven't been able to remove it all together. I didn't ever expect it to be 100% perfect and accurate. With automated processes it's never totally accurate and you always need someone to keep an eye on things just to make sure the system is behaving itself. When you put data through the blender it's always going to have little oddities that need changing, especially if the mapping rules are complex and the data in the record isn't perfect. I'm pleased with how accurate it is, though.

We've been running things for about 3 years now and I know things can be improved, so I'm having another look at those mapping rules in an attempt to get closer to that 100%.