As the name may disclose, Sentiment Analysis refers to a process by which data is mined in order to extract contextual meaning. The alluring aspect, albeit hidden from the name, is that this process is carried out by a machine. Through Sentiment Analysis, an automated engine is proficient at determining and categorising the emotional tone of a given text; positive, negative or neutral.
The capacity to scrutinise voluminous amounts of public opinion is undoubtedly a powerful tool. With millions of users willing to share their thoughts on a broad range of topics, microblogging websites such as Twitter become rich sources of data for Sentiment Analysis and opinion mining. These insights are key for brands, companies and researchers wishing to track trends and unlock the value in consumer behaviour. In turn, allowing for shrewd business and political decisions.
Just Do It
An excellent example of Sentiment Analysis was carried out by global media company Forbes, to prove their predictions of a net positive for Nike’s brand and marketing reach during their polarising sponsorship of Colin Kaepernick. The former NFL player was thrust into public debate after initiating a wave of protests amongst other players who joined him by refusing to stand during the United States national anthem played before the games; a protest he described, against racial inequality and police brutality.
Forbes analysed the sentiment of a series of social media announcements that introduced the “Just Do It” campaign featuring Kaepernick. Popular hashtags that emanated from the Nike Tweets were surveyed; #JustDoIt, (positive) #BoycottNike (negative) and #MAGA – Make America Great Again (negative). Initially, the results demonstrated a spike in negative intent, yet these were later overpowered by positive counteraction. In effect, Nike’s controversial endorsement deal catered to its core demographic, a surge in customer purchase intent has increased as a result.
Thanks to the astute nature Sentiment Analysis, companies are able to fight off negative emotions by detecting core weakness in their business models. Contextual Semantic Search (CSS), a deep learning powered advanced text classifier, allows companies to conduct categorisations of concepts. By filtering messages that share similar contexts, such as ‘expensive’ and ‘flat-rate’, CSS is able to group together vast amounts of messages under one tag, such has ‘Price’. This can complement the analysis of sentiment data, making readings of issues related to a company’s products or services more specific.
For instance, e-commerce giant Amazon, may decide to unpick its customers’ views on categories such as ‘delivery’ and ‘product’ and therefore use CSS to filter all delivery-related feedback in one category and all product-related information in another. This would allow for a more productive reading, as negative or positive sentiment could be examined through identifiable topics.
An experimental programme
As revealed by The Statistics Portal, an online statistics and consumer survey site, in the second quarter of 2018, Facebook has reached 2.23 billion monthly active users, Twitter 335 million and Instagram 1 billion. With each user sharing, posting and liking content daily, huge masses of data are generated and publicly available for re-use. Aside from businesses seeking to manage reputation and fine-tune their marketing strategy to fit customer expectations, Sentiment Analysis has also been used to track voter activity and preferences in political campaigns.
The Obama 2012 campaign used an intimate process of data analytics to interpret supporter’s concerns; as well as techniques to win over potential voters and predict their likeness to become an Obama supporter. Every voter in the US was ranked a set of scores based on the probability that the individual would vote and support Obama. Voter profiling was crafted from mass interviews conducted by the campaign’s call centres, complimented by data from voter registration records, consumer data warehouses and past campaign contacts. A crafty method of capturing voter individuality and predicting future decisions.
Sentiment was mined from reactions to topics the campaign termed “entities”. Such entities were tracked across social media and newspapers in order to gauge opinion on themes the campaign considered crucial, like the public’s assessment of Mitt Romney’s position in relation to Obama. Specific entities of interest were also sponsored across social channels in the form of ads, with the intention of shaping conversations on Twitter and maximising public opinion on issues related to the campaign.
Unquestionably, the disruptive innovation and large-scale use of data by the Obama campaign surpassed traditional forms of TV advertising in U.S. politics. It revealed just how predictable voters could be and how Sentiment Analysis could shape a political campaign. There are those who question the morality of data collection for political gains or business profit; although the information pooled from social media posts are public, users may not be aware or expect re-use of the data that they generate. Yet Sentiment Analysis and opinion mining are extraordinary means that allow us to understand ourselves better, and can, if there is no breach of public privacy, transform services or policies in ways that benefit consumers and citizens.
Hold, as ’twere, the mirror up to nature
At PangeaMT, we consider Sentiment Analysis a lucrative and powerful tool. The need to stay competitive and maximise productivity is essential to any business; truly understanding customer opinion can have a huge effect on the growth of a service or product. Moving forward, Sentiment Analysis will not only indicate a positive, negative or neutral reading, but also reveal deeper and broader observations.
A fine-tuned algorithm able to learn more about us than we know about ourselves bears an eerie resemblance to something more akin to a science fiction novel. However, such developments in Natural Language Processing are closer than we think. As the number of online users grows; as does the volumes of data they generate, computational processes in Machine Learning will undoubtedly sharpen. An effective model for identifying, categorising and revealing mass public opinion, a thought-provoking idea indeed.