ORB Visualization

(soon)

3/9/2004 8:08 PM

The Orb structure can be developed using more complex parsing rules, including ambiguation/disambiguation rules, and rules in which the current very simple word level 5-gram is replaced by a more general measurement of catalytic indexicals (or memes).

On March 9^th, Nathan harvested the URL’s linked from

BCNGroup Glass Bead Games

At a depth of 2 hyper-links.

The resulting harvest consists of x documents, having 2138 atoms. The set of atoms are a subset of the set of unique words. An atom is a word that co-occurs with a word that also co-occurs with a word different from the first two. Co-occurrence patterns are encoded as ordered triples of the form < a, r, b > where the atoms are the a and the b.

(This may seem to be an odd rule, but it is one that was convenient for us. We are claiming that many types of rules exist, and will create a system of subject matter indicators that have value. The selection of rules is a key to increasing the fidelity of Orb constructions. Our work has been developed on minimal funds and we understand exactly where we would make changes that would increase the measure of fidelity between subject matter indicators and concepts that are experienced by human reading of this text. )

Figure 1: Orb harvested using a stop word list

A stop word list was used to harvest 1521 atoms. The use of the stop word list eliminates common words from the input text before a word level 5-gram measurement is made.

The elimination of commonly occurring words means that on average there is less connectivity between atoms. This characteristic of this Orb and the one developed in Tutorial One is seen in the number of gather iterations needed to bring the central core of the limiting distribution together. Figure 1 took 6,000,000 iterations as opposed to 500,000 iterations to produce Figure 3 in Tutorial One. The time to compute this result was still within a few minutes.

Downloading the data folder for Tutorial Two and substituting this folder for the data folder in the first tutorial software download allows one to double click on the A node in Figure 1, to produce the visual images seen in the following figures.