What is Sentiment Analysis?
Sentiment Analysis and Opinion Mining (SAOM) is an area of machine learning in computer science which attempts to derive sentiment – a users opinion or feeling – from text. Sentiment analysis has become a hot topic over recent years in academia and in the business world, as its value has become clear. Researchers have created and examined powerful new methods to achieve near human, or in some cases, surpass human ability to recognise sentiment, and far surpass the human ability to analyse the shear quantity of data available. Businesses have recognised it as a tool for customer engagement, project success/failure measurement and much more.
Why would you measure opinion?
Measuring opinion is about creating insight. Insight into customer/user option which can be used to determine the direction of a project, measure how a project has been received, even monitor opinion on an issue in society or political party, and so much more. Computational methods of measuring opinion have many benefits over traditional methods, two of the most important and most valuable allow the analysis process to be automated and constant.
Successfully encouraging users to take an action is hard. Conversion rates are typically in single digits for most actions online. Attempting to convert users to freely offer their time to fill out a form – where the results may be statistically insignificant due to a small sample size is hard and quite often fruitless. Machine learning and sentiment analysis allows the option to gather the type of data users would fill in on a form, passively, without interacting with the user, via the information they freely share online on social networking sites, blogs, forums, and the rest.
The fact this can be done automatically, and in the case of social networks which are in most cases “real-time”, where users share messages while they take actions, allowing near live data. This can be immensely valuable depending on the use-case.
This isn’t to say sentiment analysis should replace focus groups or surveys entirely, these have their place. Sentiment analysis is just another tool in your toolbox.
Sentiment Analysis in Practise
One of my lecturers at university, Dr. Kevan Buckley, was involved in a project called SentiStrength. SentiStrength measures the sentiment of tweets on Twitter, either positive or negative. It was used during the 2012 Olympics in London to display the visual mood of the nation from twitter towards the games in colourful lights on the London Eye, the large ferris wheel permanently stationed on the banks of the River Thames. SentiStrength was also put to use during the the UK Summer Riots of 2011, analysing 2.6 million tweets in an attempt to understand the disorder. Gathering data during these large events without automated, remote and constant analysis would be extremely costly. Hundreds of surveyors would need to be deployed and briefed to gather an appropriate sample size and cross section of the attendees of these events. Not to mention most rioters don’t want to stick around and fill in surveys.
Is Sentiment Analysis Right for Your Project?
That is down to you. I’m an advocate of using computational methods to analyse opinion due to the advantages described above, I hope this short article has achieved its objective and made you consider measuring opinion computationally.
. . .
Sentiment analysis and opinion mining is discussed in more detail in my book, Practical Deep Learning Fundamentals, which will be released on September 25th 2018. Sentiment analysis also forms one of the major projects featured in the book, where the reader is guided through the implementation of a neural network for sentiment analysis in Google’s TensorFlow.
Sign up to read the first four chapters as they become available and be the first to know when the book launches.