Fuzzy Content for Fuzzy People

Suppose you asked someone to classify some objects such as an ashtray, a painting and a sink, as either "furniture" or "home furnishing". That would seem to be a straightforward task.

If you also asked them whether the same objects belong in a single group comprised of both "furniture and home furnishings," you would expect that any object that they classified as either one or the other would belong in the combined or parent group. A logical disjunction.

Such assignment tasks are very much like those that we require of enterprise content management system (ECM) users to assign metadata about a content (i.e. digital files) they are adding. Such metadata helps subsequent retrieval through searching and browsing, and potentially supporting dependent business processes (e.g. a triggered workflow).

There's a problem though. Often people will not make the classification you expect. They may place an object in one of the original categories, but not the larger or parent one if it is the only choice they have! There is a tendency for people to delay making a decision if there might be an outcome they don't know. Apparently this phenomenon has been documented over two decades by psychologists and is referred to as the 'disjunction effect'.

I learned about this in a New Scientist article posted yesterday (5 September 2011): Quantum minds: Why we think like quarks.
The article describes one of the first observations of the disjunction effect:
"In the early 1990s, for example, psychologists Amos Tversky and Eldar Shafir of Princeton University tested the idea in a simple gambling experiment. Players were told they had an even chance of winning $200 or losing $100, and were then asked to choose whether or not to play the same gamble a second time. When told they had won the first gamble (situation A), 69 per cent of the participants chose to play again. If told they had lost (situation B), only 59 per cent wanted to play again. That's not surprising. But when they were not told the outcome of the first gamble (situation A or B), only 36 per cent wanted to play again."
Traditionally in ECM we have held that it is difficult to get users to add metadata to describe the content they are adding; in essence that users are lazy. We have not considered that the choices presented to users, and any concurrent information presented, will actually change whether they provide the necessary data, when they provide the data, or indeed the actual values they choose.

ECM taxonomies are built on the assumption that users can make logical decisions to correctly describe content. Typically we present mutually exclusive choices, often organized in hierarchical (parent-child) fashion. But as the New Scientist article notes, people employ a kind of quantum logic that allows for something to be a bit of two exclusive alternatives, and for the context of the classification (the measurement in quantum terms) to affect the outcome. As a result their content classifications are fuzzier then we expect or perhaps need.

Content is often described as unstructured information. Metadata schemes are commonly applied to impart a structured framework to manage that unstructured content, but the fuzziness of human logic may make this doomed to failure.

No comments:

Post a Comment