The Language of talkRA

Looking at the language in a subject area can often provide fascinating, or at least interesting, insights into how thoughts and ideas change over time. A couple of weeks ago I asked Eric for a copy of all the posts ever put on talkRA. I had been playing around with Python’s open source Natural Language Toolkit (www.nltk.org) and wanted to test this on something familiar. Eric kindly obliged with the data and this is the result of my cursory analysis.

I used the more than 500,000 words in the talkRA blogs and responses as a proxy for the changing discussions in the world of revenue assurance. Some explanation first on reading the images. All words on talkRA were sorted chronologically so the first word written by Eric, in the first blog called “The RA Truck Stop” is in position 1, the second is in position 2 and so on up to a week or so ago. When you look at the image (click to expand) each time a word, or phrase, appears on a talkRA, the position number is noted and then plotted at that position. The more points on the graph, the more the word or phrase has been used, and the bigger the number for any individual point (or higher position on the image), the more recently it has been used. Hope that’s clear and so on to three quick pieces of analysis:

Firstly, I’ve taken some of the key phrases we hear. “Revenue Assurance” appears, not surprisingly, pretty consistently throughout the history of talkRA blogs. “Risk management” is emerging more consistently and “business assurance” is a late entry but starting to become more relevant.

Secondly, I thought I’d look at the business functions or processes that get talked about. Again you can see there is an interesting story to be told. “Billing” is a constant theme and the “switch” and “finance” are of interest but discussion on signalling and provisioning is limited.

Lastly, I looked at some (I acknowledge the list is incomplete) of the organisations where RA gets talked about. Again some interesting trends – I expect all readers will be familiar with the names mentioned and perhaps its safer if I let you draw your own conclusions.

I still have the data and the above are only my ideas. If any readers have any further words or phrases (preferably 2 words only) then provide a comment and in a week’s (give or take) time, if there is interest, then I’ll provide an updated graph with readers’ requests.

Mike Willett
Mike Willett

Mike is a Partner at Ernst & Young, Australia. He is responsible for enterprise intelligence, helping clients to improve their management and use of data. He can be contacted at: mike.willett@au.ey.com.

 

Mike was previously the Director for Fraud & Revenue Assurance at Telstra. He started his career at BellSouth (now Vodafone) in New Zealand and then moved to Praesidium Services in the UK. Mike graduated from the University of Auckland in New Zealand with degrees in psychology and marketing.

  • Great post! Good read!

  • Güera Romo

    I would be interested in looking at this tool. I did a similar exercise manually using an open coding technique that supports Grounded Theory. ATLAS.ti can help with auto coding much as I understand you have done here.

    Do you have the ability to add definitions to your key words? While some words may be obvious such as “switch”, it becomes a hairy discussion when your and my understanding of the difference between “RA process” and “RA methodology” is not the same. We would interpret the graphs differently.

  • Mike Willett

    Hi Guera,

    It’s fair to say that for this is early days with this software package for me so I’m not well placed to answer your question. This is open source software though and, fortunately, unlike some open source projects the website (www.nltk.org) has pretty good documentation on how to use it. In addition some books, including cookbooks, have been released.

    It’s true that some context of the words used would be beneficial, and my quick analysis only looked at the presence, or absence, of the words or phrases shown. One thing I know it can do is a key word search (like in all applications), but also provide the sentence around it. To interpret this though is a more manual task.

    I myself have more learning to do, in relation to its capabilities for more advanced text analytics.

    Mike