Data Types for Conversation Analytics

July 29, 2017

Conversation analytics is work done to undertand, monitor, and track information within speech or text from conversations. This is powerful within sales, service, and media. We use conversation analytics to determine topic, sentiment, intent, and compliance to drive conversion, quality, and efficiency.

Here we demonstrate a few types data types that are useful for capturing and modeling the speech conversation data type so it is analytics ready. These examples can be found on github at


Switchboard is one data type where it captures speaker phrases as they go back and fourth. More details can be found at []

The python description says: A specialized list object used to encode switchboard utterances. The elements of the list are the words in the utterance; and two attributes, speaker and id, are provided to retrieve the spearker identifier and utterance id. Note that utterance ids are only unique within a given discourse.

Here is a sample of python code to load the sample data:

import nltk
from nltk.corpus import switchboard

Here is the sample data:

[<B.2: u'Uh , yeah , currently we have a poodle .'>,
 <A.3: u'A poodle , miniature or , uh , full size ?'>,
 <B.4: u"Yeah , uh , it 's , uh miniature .">,
 <A.5: u'Uh-huh .'>,
 <B.6: u'Yeah .'>,
 <A.7: u'I read somewhere that , the poodles is one of the , the most intelligent dogs , uh , around .'>,
 <B.8: u"Well , um , I would n't , uh , I definitely would n't dispute that , it , it 's actually my wife 's dog , uh , I , I became part owner six months ago when we got married , but , uh , it , uh , definitely responds to , uh , to authority and , I 've had dogs in the past and , uh , it seems , it seems to , uh , respond real well , it , it she 's , she 's picked up a lot of things , uh , just , just by , uh , teaching by force , I guess is what I 'd like to say .">,
 <A.9: u"Oh , uh-huh . So , you , you 've only known the dog , wh- , how long did you say .">,
 <B.10: u'Well , about a year I guess .'>,
 <A.11: u'Oh , well , uh , is it , uh , uh , how old is the dog ?'>,
 <B.12: u'It just turned two , I believe .'>,
 <A.13: u"Oh , it 's still just a pup .">,
 <B.14: u'Pretty much , yeah , yeah .'>....

Not only is this conversation quite entertaining, this is also a useful data format for various analytical methods.

comments powered by Disqus