W06 Sentiment Analysis

Question 1

Sentiment Analysis offers…

Answer 1

…direct insights on the firm’s relationship to (potential) customers

Question 2

What is new in algorithm-aided sentiment analysis?

Answer 2

1 automated analysis
2 large amounts of text data
3 collected over the web
4 with objective predictive power

Question 3

Where to find data for sentiment analysis?

traditional. ..?
new. …?

Answer 3

traditional:
surveys, interviews, written complaints
+structured , targeted
- limited responses, sugar-coated, too negative

new:
harvesting social networks
\+ free expression in large quantities
-not structured
-difficult to control
-probably biased

Question 4

Sampling Terminology:

• complete set of elements we study
• collecting data from every population member?
• collecting data from every population member voluntarily?
• subset of members selected from population…must be?

Answer 4

• population
• census
• referendum
• sample subset
• > same characteristics
• > to make inferences about population
• > “well selected” to “well represent”

Question 5

Bias(sed sample)

Answer 5

positive or negative
misleading results

consider for different channels
negativity bias

Question 6

Web Scraping

Answer 6

Extracting Sentiments from Online Sources

Python or Crapy
Efficiency dependent on search patterns

Ethically and legally “grey” area

Question 7

Parts of an Opinion

Answer 7

1 Entity
2 Aspect
3 Sentiment
4 Owner
5 Date

Question 8

Sentiment Analysis via supervised learning

Answer 8

-supervised learning
-score function
p(w I good) - p(w I bad)
/p(w I good) + p(w I bad)

-lexicon
domain specific
1 score for word
2 score for switching words
3 score for contradictions
4 aggregate and weight