Concise 0.3.0 Release: with 'Dead End' Collocational Network Feature

As suggested in my previous article Collocation and Interactive Collocational Network, collocational networks are networks consisting of words that co-occur in a statistically significant way in a text.  In Concise 0.2x, we introduced an interactive way to explore the co-occurencetial relationship.  Now, with Concise 0.3.x, a 'dead-end' collocational network is featured.

The 'dead-end' collocational network provides a whole picture of your fundamental 'core' word.  It keeps expanding the network until nothing left.  However, the 'core' word you're using is not exactly the 'core' of its network.

Take the upper collocational network for instance.  I was looking for the collocational network of 'farmer' (農民) among some of Council of Agriculture's (農委會) official documents.  The top five collocates (sorted by co-occurrence) are 'agriculture' (農業), 'counsel' (輔導), 'conduct' (辦理), 'promote' (推動), and 'develop' (發展).  These nodes suppose to be the central part (the 'core') of the network if documents are randomly selected.  But these official documents have very strong tendencies toward agricultural policy.  That is the reason why this dead-end network is mostly comprised of policy words.

Camilla Magnusson, in her Text Visualization for Competitive Intelligence, believes collocational network method is useful handling sequence text.  To test her theory, I put my documents into two collection by time.  The first collection ranges from 1996 to 2002; and the second collection ranges from 2003 to 2009.

Figure 1: Collocational Network of Collection 1

Figure 2: Collocational Network of Collection 2

Something interesting did show.  Figure 1 remains simple compared to the top figure, but figure 2 is much complicated.  Lots of things come together: birds flu, foot-and-mouth disease, mudslide . . . .   Of course, they are not directly related to the word 'farmer'.  But figure 2 did show the potential foci of the 2nd collection.

Magnusson may be right.  Collocational network, the text visualizational presentation did show the differences.  But does it work in the field of agriculture?  Maybe!  Maybe not!  I have not tested it.


If you are interested, the latest Concise can be found at SourceForge: https://sourceforge.net/projects/concise-text/files/