Sentiment analysis is a process used by companies and organizations to detect and analyze positive or negative sentiments toward their brand. AI and other specialized social listening tools can analyze text and determine whether people feel positive, negative, or neutral towards a brand. This information can help the company understand if its marketing message is working, if products are well received, or if some external event is impacting consumer trust.
Historically, sentiment analysis simply focused on measuring positive, negative, and neutral sentiments. Today’s tools, however, are more advanced — and, as a result, can provide more useful information to help companies more fully know their audience. Here’s how sentiment analysis has evolved, and how your brand can use sentiment analysis to build consumer trust.
Sentiment analysis has evolved over time. Today, sentiment analysis goes beyond measuring positive, negative, or neutral sentiment to detect more specific feelings and emotions (for instance, angry, happy, or sad), urgency (urgent, not urgent), and even intentions (whether someone is interested or not interested).
The process of sentiment analysis generally falls under four categories. The first, and most basic, is graded sentiment analysis. It measures sentiment on a sliding scale of very positive to very negative, with neutral in the middle. A five-star rating system is an example of graded sentiment analysis.
Machine learning algorithms are necessary for the next category of sentiment analysis: emotion detection. Algorithms can scan social media posts and other content to detect “lexicons” — lists of words and the emotions they convey. This enables the sentiment analysis tool to discern whether a person is happy, irritated, sad, or some other more advanced emotion. The downside is that sometimes these algorithms struggle to detect sarcasm or humor, and can get the sentiment wrong.
Aspect-based sentiment analysis ties a particular emotion to a product or product feature. For instance, a product review that says a pair of jeans is bad quality can determine that the review is sharing a negative opinion about the product.
And, finally, the most complex category of sentiment analysis is multilingual analysis — complex machine learning and algorithms detect, understand, and translate sentiments in multiple languages. This type of analysis takes a lot of complex coding and resources, and therefore is the least common technique.
Despite advancements in sentiment analysis, there are still limitations to what this process can offer. Sentiment analysis tools struggle to understand sarcasm. Sarcasm requires context to detect, and tools are equipped to study the environment in which the sentiment is expressed.
Sentiment analysis tools also struggle with negation detection. “For example, in the sentence ‘The show was not interesting,’ the scope is only the next word after the negation word,” explained Toptal. “But for sentences like ‘I do not call this film a comedy movie,’ the effect of the negation word ‘not’ is until the end of the sentence. The original meaning of the words changes if a positive or negative word falls inside the scope of negation—in that case, opposite polarity will be returned.”
Likewise, word ambiguity can throw off sentiment analysis entirely. Languages are nuanced; and for some words, the context is what differentiates a word’s meaning. Take the word “unpredictable.” An unpredictable story would be exciting; an unpredictable toaster oven would be frustrating. Sentiment analysis tools aren’t always able to discern an ambiguous word’s weight and could misclassify a statement as neutral (or worse, the opposite intended emotion).
[Read more: Why Media Monitoring Sucks]
How should companies analyze customer sentiment if sentiment analysis tools are flawed?
Identifying and tapping into narratives is one way that brands can better understand their customers’ attitudes. Narratives are defined as “big, overarching concepts or ideas, generated when many stories about the same or similar things align.”
“These individual stories are accounts of real people or events that serve to bring the big concept or idea to life and give it meaning, reality and legitimacy,” wrote Stronger Stories. “Narratives are bound by time and culture, evolving with society, and existing within networks of all scales.”
Narratives are formed from the repetition of stories that are shared in TV shows, music, games, and news. Patterns begin to emerge through the media landscape, and these patterns become narratives. News organizations and media outlets often work to shift deep cultural narratives that shape our understanding of social and political issues, such as poverty, immigration, and climate change.
What does a narrative look like in practice? The pandemic offered a useful case study of how narratives form, spread, and impact public opinion. At PeakMetrics, we examined a broad range of media produced since December 1, 2019, and pulled over 135,000 English-language online written news articles from a wide range of both local and national scale media outlets.
We searched for trends in how people responded to the spread of Covid-19 over time — what narratives were forming from the media coverage of this event?
Our findings concluded: When we take action (or fail to take action), the media recognizes it. There were a number of pivotal moments where the saturation spiked and continued increasing, at milestones that can be identified. How will COVID-19 play out in the media in the near future? It depends on how we respond to the challenge.
Sentiment analysis can be better informed by identifying and mapping narratives. Understanding how the media shares a story can help a brand can better monitor what shapes consumer opinions, influence attitudes, and find opportunities to engage and build consumer trust.
To learn more about PeakMetrics’ tools, request a demo with one of our experts.
Ask us anything! We are happy to talk through any questions or thoughts you may have.