SENTIMENT ANALYSIS

Unit 2, Module 2.6

1

HOW DO COMPUTERS KNOW HOW PEOPLE FEEL?

2

2

NATURAL LANGUAGE PROCESSING

• Natural Language Processing (NLP) is how computers process and understand human language
• This allows computers to communicate with us via written and spoken language
• NLP techniques can be used to analyze text or speech the user inputs, and can also use machine learning to train models on text or spoken language
• These models must be trained on large amounts of text

3

MODULE OBJECTIVES

Students should be able to:

• Define what Natural Language Processing is
• Describe the difference between subjectivity and polarity in sentiment analysis
• Describe limitations of sentiment analysis tools

4

ACTIVITY: GOOGLE TALK TO BOOKS

• Natural Language Processing is fundamental to tools we use on a daily basis
• NLP is integral to search engines, allowing users to type natural language queries and get relevant results
• Google Talk to Books - search engine that gives results as complete sentences from books that best answer your question
• Link: https://books.google.com/talktobooks/�
• Try asking questions and seeing what answers you get here versus when searching on Google
• Try some abstract/subjective questions like “How can I find happiness?”

​

REFLECTION: What are some challenges you think search engines run into when trying to understand and answer user’s questions?

5

6

6

Sentiment Analysis With Machine Learning

© AI4ALL, 2019. May be reproduced with permission.

​

Credit to this article for activity idea

​

7

7

Sentiment: a view or opinion of something (Oxford)

Sentiment Analysis: a process that a computer performs to measure the positivity or negativity of a text

There are two aspects of sentiment analysis:

• Determine subjectivity of text (factual vs opinion)
• Determine polarity of opinion (positive vs negative)

​

​

Sentiment

© AI4ALL, 2019. May be reproduced with permission.

8

8

​

• Subjectivity - how much of an opinion vs fact something is
• In our activities, the subjectivity rating will be somewhere between 0 Some articles are just informational articles. Others are opinion pieces.
• (entirely informational) and 1 (entirely subjective).

​

​

Subjectivity

© AI4ALL, 2019. May be reproduced with permission.

9

9

Understanding Check

​

The subjectivity of each description of the puppy.

0 = Fact

0.5 = somewhat subjective

1 = Entirely Subjective

• Cute
• 4-legged
• Small
• That dog is 1 foot tall.
• My neighbor’s dog is so annoying!

© AI4ALL, 2019. May be reproduced with permission.

10

10

• Polarity - positivity or negativity of opinion
• May be strong or weak
• Numerical range from -1 (very negative opinion) to 0 (neutral) to 1 (very positive)

​

​

​

Polarity

© AI4ALL, 2019. May be reproduced with permission.

11

11

​

Rate the polarity of each description of the puppy.

-1 = very negative opinion

0 = neutral

1 = very positive opinion

• Adorable
• Fluffy
• This puppy destroys everything.
• It can be hard to take care of a puppy, but often worth it
• Puppies are the best!
• Puppies require feeding and exercise.

© AI4ALL, 2019. May be reproduced with permission.

Understanding Check

12

12

DISCUSSION

​

• In the next activity we will do sentiment analysis ourselves on songs
• What do you think gives a song a positive, negative, or neutral sentiment?

​

© AI4ALL, 2019. May be reproduced with permission.

13

13

Song Sentiment

​

• Examples:
• A song with positive sentiment might be happy, or talk excitedly/fondly about love or friendship
• A song with a negative sentiment might be sad or angry
• A song with a neutral sentiment might have an equal mix of lyrics with positive and negative sentiments, or not have lyrics with strong emotions

© AI4ALL, 2019. May be reproduced with permission.

FREE SENTIMENT ANALYSIS DEMO

14

Key:

Sentiment score [-1...1]

​

Polarity - [positive, negative, neutral], [-1...1]

​

Magnitude - [0 - infinity]

Volume of sentiment conveyed

​

Subjectivity - subjective, objective, unknown

​

​

Twitter-like content == shorter, unformatted text, slang, missing punctuation

​

​

TOOL: FREE SENTIMENT ANALYSIS DEMO

15

15

Try It #1 - Song Sentiment Analysis

​

Activity Overview

• Google the lyrics of a song that you think have different sentiments (positive, neutral, or negative)
• Analyze the sentiment of the lyrics, and pick out specific words or phrases that highlight the sentiment
• Recall that:
• Subjectivity is how much of an opinion vs fact something is (0 - 1)
• Polarity is the positive or negative strength of opinion (-1 to 1)
• On a piece of paper,
• Prediction - write your song name and prediction
• Actual - and your answers for them, and then
• Discussion
• Share results

© AI4ALL, 2019. May be reproduced with permission.

STUDENT ACTIVITY SHEET

STUDENT NAME

Song Title

Artist

16

Use Insert Video to add a quick thumbnail to the song

​

​

Make a copy of this slide and complete!

STUDENT NAME

Song Title

Artist

17

18

18

DISCUSSION

​

• Give a brief introduction to your songs
• How did you rate your songs in terms of subjective and polarity?
• What words or phrases in particular lead you to believe that the songs are positive or negative?
• Did others rate your songs similarly? If not, why do you think they disagreed?

​

© AI4ALL, 2019. May be reproduced with permission.

TRY IT #2 - COMPARISON OF SENTIMENT ANALYSIS TOOLS

FREE SENTIMENT ANALYSIS DEMO

SENTIMENT ANALYZER

TONE ANALYZER

Try out these three online demos and compare the outputs.

​

19

Not sure you agreed with the sentiment ratings of the other tool.

​

Let’s try out some other tools …

​

Goal: Explore

Ratings of song lyrics

• Look for similarities and differences in the ratings
• Explain why there are differences
• Do you agree with the rating?

​

Representations of Sentiment

• How does this tool represent sentiment?

​

​

​

TONE ANALYZER

20

SENTIMENT ANALYZER

21

STUDENT NAME

Song Title

Artist

22

Use Insert Video to add a quick thumbnail to the song

​

​

23

23

DISCUSSION

​

• Do you disagree with any of the sentiment scores? Why? What do you think made the tool rate the song the way it did?
• Do you think the average polarity is a good measure of the overall polarity of an artist’s songs? Why or why not?
• Besides song lyrics, what types of applications can you think of for sentiment analysis?

​

© AI4ALL, 2019. May be reproduced with permission.

24

24

Takeaways

​

• Tools are trained on different datasets which effects what they will recognize or not recognize as positive or negative
• The dataset used to train these models must label the data. Biases or preferences from labelers might impact the ratings of words. [Note: there should be processes and procedures in place to prevent this.]
• Ambiguity and context of word use will affect its ratings.

​

© AI4ALL, 2019. May be reproduced with permission.

APPLICATION ACTIVITY

25

25

CONNECTING TO STUDENTS LIVES

Possible Applications of Sentiment Analysis

• Cyber-Bullying
• Misphrased Text Messages
• Measuring the impact of your words and others
• Signs of mental health concerns

26

​

• HOW BAD IS IT?

​

• IS IT INTENTIONAL OR UNINTENTIONAL?

​

• IS THIS AN IMPORTANT ISSUES WE SHOULD TRY TO PREVENT?

27

WHAT IS CYBER BULLYING?

DETECTING CYBER BULLYING

28