Sentiment analysis is a term that most PR practitioners and communications professionals have heard of, and perhaps even a tool they use as a part of their strategy. However, many industry pros struggle to fully understand the concept and what it can do for them when implemented effectively.
The applications of sentiment analysis are wide-ranging and impactful. For instance, Brandwatch asserts that “shifts in sentiment on social media have been shown to correlate with shifts in the stock market.” British political magazine New Statesman even used the process to determine that President Joe Biden’s recent 2021 inaugural address was “the angriest ever,” based on key linguistic choices.
What is Sentiment Analysis?
Sentiment analysis is the process of identifying the tone or emotion attached to a communication. It can also be referred to as “opinion mining” or Emotion AI. Examples of the types of communication that can be analyzed for tone are nonverbal, like facial expressions and body language, and linguistic.
Analyzing the sentiment of linguistic forms of communication starts with examining a sample of text, which is then assigned a value based on the perceived attitude or tone of the communicator. Usually, the values are coded as positive, neutral, or negative so the data can be easily sorted and later visualized and studied for trends.
Why is Sentiment Analysis Important?
Sentiment analysis is important because it can provide you with a better understanding of your earned media coverage and help you reach your messaging goals. The analysis is part of an integral feedback loop that allows communicators to gauge the success of their communications tactics and identify opportunities for improvement.
Measuring the volume of media coverage by topic can only tell you so much. Without knowing the tone of that coverage, teams can’t determine whether their campaign is a success or a failure. For example, if your company experiences a spike in mentions related to product quality, how can you appropriately respond without first knowing whether that coverage is positive or a potential PR crisis, all of which comes down to sentiment?
Lexalytics explains that sentiment analysis can help companies to gauge “public opinion, conduct nuanced market research, monitor brand and product reputation, and understand customer experiences.” Once you have identified your strengths, weaknesses, and opportunities, you and your team can take advantage of all the practice has to offer.
Using AI for Sentiment Analysis
When analyzing text, computers deploy natural language processing and machine learning techniques to attach sentiment to words, phrases, topics, and themes. When an analysis program runs on an article it breaks the text down into these units. The program then identifies components that have been assigned sentiment in the program’s sentiment library (which stores the system’s human-coded values) – or the library entries they are closest to – and assigns a score to each unit. Finally, the system combines the individual scores to generate a multi-layered analysis score that represents the whole article.
As smooth as this process sounds, there are many areas where problems can arise along the way.
The Accuracy of AI Sentiment Analysis
Because AI uses natural language processing and machine learning to automate the process, it’s a useful tool for freeing up your team’s valuable time. However, fully automating your sentiment analysis can compromise its accuracy.
According to the Institute for Public Relations, no method of sentiment analysis will ever be 100% accurate. However, they argue that relying solely on a tech tool to measure sentiment “can be like flipping a coin, or only 50% accurate, since these platforms often struggle to measure more nuanced posts or are unable to filter and interpret the information through the lens of a company or brand.” Similarly, 5WPR estimates that sentiment algorithms are only about 60 percent accurate.
Linguistic Challenges for AI
Toptal has identified four major pitfalls of AI sentiment analysis: irony and sarcasm, negations, word ambiguity, and multipolarity. Some of these pitfalls can be addressed with approaches like machine learning algorithms or deep learning, but no solution is guaranteed to be fully effective.
Sarcasm is an especially deep pitfall, and its prevalence in consumer-generated content, like social media posts, makes it even more important in many sentiment analysis projects. Even humans struggle to comprehend sarcasm sometimes, so it’s no surprise that computers are often tricked by false-positive statements like, “I love the way [company’s] customer service team put me on hold for two hours.” Research shows that numerical sarcasm like in this statement is especially challenging for AI to comprehend due to its effect on a statement’s polarity.
As a media analyst, I often see articles that dive into complex subjects in detail. The more detailed the article, however, the higher the chances that an AI program will be tripped up by common traps like negatory statements, ambiguity surrounding entities, or articles that discuss both the pros and cons of one idea.
These issues demonstrate some of the imperfections of using AI, which can drastically change the narrative of your media analysis and your subsequent tactical decisions.
Adding a human element to your approach can be the solution to avoiding these major data hazards.
Using Humans to Detect Sentiment
Although using an AI program can help save time, its imperfections can lead to inaccurate results that can impact your communications strategy. Because of these shortcomings, it is essential to include a human perspective to analyze the more linguistically complex elements of your media coverage.
While computers need to be trained to detect subtle context clues, humans have been ‘programmed’ to find them throughout their entire socialized lives, which makes identifying common language tools like irony and negations quite simple. Using human analysts to identify these common contexts and AI to automate the basic tasks that save time can be beneficial for PR professionals as they work to improve the accuracy of their sentiment analysis insights.
The Value of a Hybrid Approach
Both AI and human analyst approaches to sentiment analysis have benefits: AI programs save time with automation, and humans decipher context and increase accuracy. Ultimately, utilizing a combined approach can offer the best of both worlds.
At PublicRelay, our human-AI hybrid approach to media monitoring makes conceptual insights possible. To learn more about using PublicRelay for accurate sentiment analysis, click here.