Professional Documents
Culture Documents
Text Analytics
Out-of-the-box
PAGE 01
DECEMBER 2021 ISSUE 8
Text analysis is tomorrow’s digital gold rush, but first we The marketing perspective :
need to differentiate structured data and unstructured Let’s assume you are a marketing analyst for top-tier consumer
data. Structured data only accounts for 20% of all total electronic company, and you have just released a game-changing
available global data by most recent estimates. The new product into the market. Ordinarily, you would wait a few
remaining 80% of the world’s data is unstructured. months and run a customer satisfaction survey and analyze star
Unstructured data do not follow any specified format and ratings or appoint a third party to collect the responses for you. In
this lack of structure makes handling this data very any case, the first resort would be somehow converting these
difficult. They can in form of text, emails, speech, audio, or qualitative data into a quantitative form. Or collecting data in such
video. way it’s easy to analyze, i.e., the NPS score format. But in both
This is where the protagonist of this article comes into the these cases you’d be ignoring a crucial element of human behavior,
scene, enters text analysis. In simple terms, text analysis or that is irrationalities and specificity when comes to human
text mining is the process of analyzing textual content emotions. We all express ourselves rather uniquely, but there is a
present in unstructured data by converting it into a hidden pattern amongst these responses.
structured format. By transforming the data into a more Using text analysis along with sentiment analysis you can analyze
structured format we can analyze textual patterns and the qualitative data, qualitatively and quantitatively. You will be
trends using machine learning, statistics, and linguistics or able to find out customers' pain points or understand why
sentiment analysis (IBM Cloud Education, 2021). To the customers prefer your brand over others by analyzing, those
more seasoned machine learning technocrats, this is not a specific words they associate your brand with, we call these words
new concept by any stretch of the imagination. Concepts descriptors. Taking things further, you can quantify the correlation
like Natural Language Processing (NLP) have come a between two descriptors, to understand how strongly or loosely
long way since earlier versions of Siri to Google’s frankly those words affect your brand or product. Lastly, and arguably
telepathic predictive text algorithms in Gmail. the most important use case of text analysis is seeing how your
But for the more budding analysts, such as myself, brand or product stacks up against your competitors.
understanding and working on text analytics has been a How you ask? With text analytics you aren’t limited to data
revelation thus far. The process is easy to comprehend and collected from surveys, you can analyze textual data from pretty
replicate. It all starts with identifying what text files you much anywhere, even social media. Your customers tend to expose
would like to analyze and collating them. Then we information about your competitors whilst talking online and they
proceed to tokenization, which is the process of breaking make comparisons between your product and your competitors.
down long-form texts like sentences into a single word or Using text analysis on such data will provide valuable insights into
collection of words called tokens. These tokens are then customers’ reaction to your product; whether it’s working as
used to analyze sentiments or perform text clustering. expected or malfunctioning; and this allows you to compare your
There isn’t one predetermined process. It entirely depends product's performance with that of your competitors.
on the objective of your analysis and the language you This also gives you an idea of market penetration of your newly
are comfortable coding in. launched product.
PAGE 02
DECEMBER 2021 ISSUE 8
Similar to the process used in sarcasm detection, AIs rely on sample data sets to
break down jokes and identify "questions" and "answers" to break down the
"essence of humor". By breakdown jokes down to evaluate their funniness, the
type of joke it is, and what contextual signals make the joke funny, the AI can be
trained to make jokes! Check out the QR code to read in depth how someone
built an AI program that made its own jokes!
While unfortunately my ego was bruised when I found out even a machine can
be funnier than me, here is my favorite AI generated joke:
What do you call a cat does it take to screw in a light bulb? They could worry
the banana. (@JanelleCShane)
PAGE 03