Intro to Sentiment Analysis: What it is, how to conduct it, and what are its limitations?
What are they saying? Sentiment Analysis Professor Matthew Kushin, PhDShepherd University | Department of Mass Communication | 2012
Preview Last class: Looked at what people say about a brand online Today, we’ll explore: Can we more systematically evaluate text content (such as tweets)?
Defined Sentiment analysis – process of categorizing text, based on the “sentiment” or “feelings” embedded in the message. Aka “opinion mining” For assessing opinions Form of content analysis – systematic process of coding content of media for interpretation
Simple Example Tweet: “Can’t wait to see Zan the Ram at the game this weekend!!!” Sentiment: Positive Problem How do we know that this text is positive?
Basics: How it works A database of words and symbols (e.g., !) is created. Each word is assigned a value Positive = 1 Negative = -1 Neutral = 0 Example: “love” = 1; “hate” =2; “blue” =0
How it works, contd Computers or person evaluates each piece of data (e.g., a Tweet), searching for words in database. Total number of positive/negative/neutral counted in data, and a sentiment score or % is given
Example Tweets about Shepherdstown over a one-week period: 60% positive 20% negative 20% neutral
Potential, potential Attitude towards your brand Perception of products, ideas, brands, people, etc. Reputation management Able to respond to posts Evaluate over time to see if sentiment is changing as part of campaign goals
Example: Taco Bell Beef! Last class an unrepresentative sample of Tweets we happen to look at Sentiment offers: Much more systematic evaluation of tweets Evaluate thousands of social media posts Very quick & little cost
Tools Many tools (paid and free) exist for assessing sentiment Study of linguistics is applied backed by years of research & understanding of human language. Common limitations: Language contains context Is subjective Inability to assess sarcasm & other nuances of human communication
Key Considerations We must interpret what the sentiment means Sentiment alone is only a minor indicator of the true feelings of the crowd. Need to ask “Why is it positive/negative?” “What products do people mention?”
Beyond Basic Sentiment Computer-assisted content analysis of social media posts virtually limitless potential More detail than “pos” “neg” “neutral” What about: “satisfied” “dissatisfied” “concerned” “informed” “afraid” “glad” etc. Count all mentions of ANY term Ex: product mentions in tweets: “Fries” “McRib” “McFlurry”
Participation: Obama v. RomneySentiment Getting the PDF files: Obama Tweets PDF: http://bit.ly/402_ObamaTweets Romney Tweets PDF: http://bit.ly/402_RomneyTweets