9
Sentiment analysis
What is it?
Sentiment analysis, also known as opinion mining, seeks to extract subjective opinion or sentiment from text (Chapter 8), video (Chapter 11) or audio data (Chapter 12).
The basic aim of sentiment analysis is to determine the attitude of an individual or group regarding a particular topic or overall context. The sentiment or attitude may be a judgement, evaluation or emotional reaction.
For example, we have known for decades that the words we use to communicate and express ourselves and our opinions only account for 7 per cent of comprehension. The vast majority of our communication is picked up non-verbally through body language and tonality. Most of us have, for example, experienced asking our children to clear the dishes, or asked an employee to stay late at work and although the words coming out of their mouth may indicate they agree – we are left in no doubt about what they really want to do!
Sentiment analysis seeks to get to the real truth behind communication so that businesses can make better decisions by working out if stakeholders feel positively, negatively or neutrally about our products, business and brand.
When do I use it?
You would use sentiment analysis when you wanted to know stakeholder opinion.
Say you have a lot of text data from your customers. That may originate from emails, surveys, social media posts, etc. There are several hundreds of thousands of words in the English language and while some are neutral, others have a distinctly positive or negative vibe. This polarity of sentiment can therefore be applied to your customer text to establish what your customers as a stakeholder group really think of you.
There are number of software tools that can help you to measure text sentiment around your product or service. Twitrratr, for example, allows you to separate the positive tweets about your company, brand, product or service from the negative and neutral tweets so you can see how well you are doing in the Twitterverse.
What business questions is it helping me to answer?
Sentiment analysis can help you to gauge opinion, which can in turn guide strategy and help decision making. It can help answer:
- How positive do our customers feel about our brand?
- How does the perception of our product compare to that of products from our competitors?
- What is the perception of our employment brand?
How do I use it?
Obviously what someone thinks and feels is very subjective so the data you have in order to analyse this subjective element would need to indicate sentiment in some measureable way.
This could be text, audio or video. You can, after all, tell whether someone is happy, angry, happy, excited, etc., by the words they use, the pitch and tonality of their speech, and their facial expressions.
As a result you can apply sentiment analysis to text, speech (audio) and visual interactions (video).
Advanced, ‘beyond polarity’ sentiment analysis can also go further by making a classification as to the emotional state involved. For example, text, audio tonality or facial expressions can determine whether the person is ‘frustrated’, ‘angry’ or ‘happy’. This type of analytics is becoming increasingly popular with the rise of social media, blogs and social networks where people are sharing their thoughts and feelings about all sorts of things – including companies and products – much more readily.
It is also being used to measure emotional charge on telephone waiting queues or insurance claims lines. Tonality of voice may, for example, indicate that someone is getting upset or it may indicate that they are telling a lie!
In the current business landscape it is increasingly important that we know what our customers, competitors and employees think about the business, products and brand, and sentiment analytics can help us do that – often relatively inexpensively.
If you want to know more about the various sentiment analysis techniques and how to use them you can explore the links at the end of this chapter. Alternatively there are many commercially available sentiment analysis tools on the market that can help you.
Practical examples
Researchers at the Microsoft Research Labs in Washington discovered that it was possible to predict which women were at risk of postnatal depression just by analysing their Twitter posts with text-based sentiment analysis.
The research focused on verbal cues that the mother would use weeks before giving birth. Those who struggle with motherhood tended to use words that hinted at an underlying anxiety and unhappiness. There was more negativity in the language used with an increase in words such as ‘disappointed’, ‘miserable’, ‘hate’ as well as an increase in the use of ‘I’ – indicating a disconnection from the ‘we’ of impending parenthood.
Co-director of Microsoft Labs Eric Horvitz acknowledged that this type of information can be incredibly useful in reaching out and helping women at this vulnerable time and also to help break down the stigma around postnatal depression. It would be a relatively simple step, for example, for a welfare group to create an app that could run on a smartphone and alert pregnant women to the onset of potential postnatal depression and direct them to resources to help them cope.
Audio sentiment analytics is also being used to measure stress levels in call centre’s so that customer service representatives can measure how upset the caller is and intervene earlier before things escalate. For example, people often talk into the receiver, even when they are on hold or listening to the soothing music; they can also make various sounds such as heavy sighing which can indicate the caller is getting increasingly frustrated.
Tips and traps
Sentiment analytics is pretty funky stuff because it can tell us things we didn’t know and had no way of understanding in the past. This makes it appealing and sexy so make sure you are not just sucked in because it sounds like a useful thing to do.
Like all analytics it is only useful if there is a commercially viable reason for doing it. Make sure there is.
Further reading and references
Sentiment analytics is usually performed using commercial (and sometimes free) software and many vendors like SAS, IBM and others provide very good reading material. See for example: