Network database
A network model database management system has a more flexible structure than the hierarchical model or relational model, but pays for it in processing time and specialization of types. Some object-oriented database systems use a general network model, but most have some hierarchical limitations.
The neural network is an important modern example of a network database - a large number of similar simple processing units, analogous to neurons in the human brain, 'learn' the differences and similarities between a number of inputs. At any one time the 'weights' assigned to different connections between layers of neuron-like processing units constitute a set of assertions about what is most closely related to what.
This enables prototype models to be built or inferred from the weights, and possibly used to define formal typed link or typed object schemas. Internet search engines like Google use a technique similar to this, to determine whether a link could normally be expected to a given page on the World Wide Web, from a page containing the search terms.
As this example suggests, network models are particularly useful to text analysis. One approach, called singular value decomposition, enables many-dimensional maps of large corpus text databases to be drawn, and to generate an orthogonal base of vectors of difference between text. This approach yielded interesting results, such as the fact that a given medical specialty typically contained 100-140 degrees of freedom in its text corpus - a number remarkably similar to the number of human beings in a village of optimal size according to anthropology. This led to speculation that tracking differences in implications of medical papers and tracking differences in behavior of one's neighbors relied on the same mental module in the brain. It remains controversial, but if it were so, that would suggest that the human brain's cognition resembles a network database more than the other models.
See also: massive parallelism, collective intelligence
|
