8a. Text Mining

Question 1

What is the relevance of text data today?

Answer 1

Highly relevant…
-> mainly due to data growth in unstructured data

Question 2

Is text data unstructed?

Answer 2

Somewhat..

Oftentimes text data is not completely unstructured due to punctuation and other semantic features -> weakly structured

Question 3

Text data is _________ structured

Answer 3

Text data is WEAKLY structured

Question 4

What are some typographic elements that give structure to a document?

Answer 4

• Punctuation
• Capitalization
• Numeric Special characters
• Asterisks

Question 5

What are the 3 types of text data?

Answer 5

1. Customer
2. Firm
3. Insitutions/ Society

Question 6

What is “customer text data”?

Answer 6

• Customer reviews of online products on Amazon
• Customer emails, chats, or even (transcribed) calls with customer service staff
• Customer posts and comments on social networking sites such as Facebook or Instagram.

Question 7

What is “firm text data”?

Answer 7

• Owned meda (eg. company social media)\
• Adverts
• Communication with investors
• Packaging + labels

Question 8

What is “Insitutions/ Society text data”?

Answer 8

• News content
• Movies
• Songs
• Books

Question 9

What is the definition of “text mining”?

Answer 9

“A computer-assisted methodology that allows researchers to rid themselves of measurement straitjackets, such as scales and scripted questions, and to quanity information contained in textual data as it naturally occurs”.

Question 10

What is the goal of “text mining”?

Answer 10

convert text into numbers to be analyzed with statistical and machine learning techniques

Question 11

How does “text mining” relate to linguistics/ Natural language processing/ ML/ Statistics?

Answer 11

Question 12

What are two main analytical techniques for analysing unstructured big tech data?

Answer 12

1. Text retrieval
2. Text mining

Question 13

What is the difference between “text retrieval” and “text mining”?

Answer 13

Text retrieval:
- obtaining the data
- also referred to as scraping (and sometimes mining, which is why its a major source of confusion).
- finds most relevant data needed from a large collection of text data

Text mining:
- processing the data
- assists user in analysing patterns in text data and discover actionable knowledge for decision-making

Question 14

Between “text retrieval” and “text mining”, which takes place first?

Answer 14

Text retrieval first! (obtaining relevant data)
= text mining (discovering actionable knowledge)

Question 15

What are 4 main applications of “text mining”?

Answer 15

• marketing
• IS
• management
• finance

Question 16

Why is text said to “matter”?

Answer 16

• Generates framing (Gamson 1992) in the way the text conveys a product, company, market or an idea. This has been shown to change perceptions and behaviour across many different contexts.
• Agenda-setting, psychological priming, availability, associations, etc.

Question 17

What are the two main functions of “text mining”? (goals)

Answer 17

• Prediction
• Understanding

Question 18

How is “prediction” a goal of “text mining”?

Answer 18

Goal: predict a particular outcomes with the highest level of accuracy
-> focus on how features can best be combined to achieve the best prediction

eg. “which movie will be more profitable?” using IMDB reviews (prediction)

Question 19

How is “understanding” a goal of “text mining”?

Answer 19

Goal: gain a deeper understanding of why and how something has happened.
-> Focus on uncovering which text features drive which outcomes and why

eg. “Why do some movies become more popular than others?”

Question 20

What are the two types of “Automated text analyses”?

Answer 20

• Top-down
• Bottom-up

Question 21

What is “Bottom-up” text analysis?

Answer 21

Bottom-up means you analyze the text without a preset dictionary and let themes, patterns, or topics emerge from the data itself.

Question 22

What is “Top-down” text analysis?

Answer 22

Top-down = you use a dictionary or predefined coding scheme to look for specific things in the data, so it’s more deductive.