How Businesses Use Sentiment Analysis in Social Media to Monitor Your Feelings

What is Sentiment Analysis?

Have you ever typed out a positive review of your favorite appliance and posted it online? What about an emotionally charged tweet series, lambasting the quality of a fast food chain? A neutral rating of an Amazon product? All of these posts have one thing in common: your emotions. It is relatively easy for humans to decipher emotions and context in text, although there are definitely times when people misinterpret the emotions behind text. If humans aren’t accurate all the time, how could machines possibly be any better?

Let’s start with what text actually is. Text is simply unstructured data — information that doesn’t conform to a preset data model or pre-defined schema. Unstructured data makes up a good 80–90% of data. The internet is mainly composed of unstructured data such as text, video, audio, and logs. On the other hand, structured data is information that has clearly pre-defined data models and data types that are usually stored in relational databases, readily available for analysis. Compared to structured data, unstructured data is hard to store, hard to process, and hard to analyze, so why would deciphering the convoluted emotional meaning of text be any easier? It turns out that with a natural language processing (NLP) task called sentiment analysis, machines can figure out the emotional content of text just as well — or maybe even better — than humans can.

With text data, sentiment analysis analyzes and identifies the sentiment and intent behind a chunk of text. There are different approaches to sentiment analysis, from supervised machine learning models to lexicon-based methods to hybrid approaches. With supervised machine learning approaches, words and sentences are often extracted into features for Machine Learning models to work on. Humans label training datasets with the correct outputs, then the models learn from the data and associated classifications and scores. With lexicon-based approaches (also known as knowledge-based approaches), words in text are labeled as positive or negative by use of a dictionary. Hybrid approaches are a mixture of different approaches.

Figure 1: Machine Learning vs. Lexicon Based Sentiment Models (Source: Machine Learning and Lexicon Based Approaches To Sentiment Analysis)

Businesses: How Sentiment Ties Into Choices

Businesses often turn to the public when deciding on what direction to take in terms of service, advertisements, products, and branding. For example, the American breakfast brand Aunt Jemima rebranded to Pearl Milling Company in 2021 because of controversy surrounding the history of the name. Negative reviews and press explaining the racist archetypes behind the company’s name and logo caused the company’s reputation and products to suffer, so they changed their brand name. When consumers have so much purchasing power, opinions matter.

Sentiment analysis is also commonly known as opinion mining. The act of opinion mining can help organizations gather unstructured data from social media sources and gather insights from it. Businesses can use opinion mining to gather information about their brand, reputation, and potential changes they could make in the future.

Figure 2: Rebranding of Aunt Jemima (Source: Aunt Jemima Reveals New Name and Logo)

Let’s say you write an Instagram post or comment about the popular cookie brand, Oreos. They have recently changed their long-standing and delicious recipe, and have put those new cookies out into grocery stores. You are disappointed with the recent formula change and decide to complain about it online. You, and everybody else who cares about the formula change, post their opinions, whether they be negative, positive, or neutral. Oreo’s manufacturer, Nabisco, can then mine the reviews from different web sources and conduct sentiment analysis to see whether or not the change was received positively or negatively, and then reevaluate their future endeavors.

Different Types of Sentiment Analysis

Sentiment analysis can be categorized into different categories depending on how an organization wants to interpret consumer feedback. There are two main categories: coarse-grained and fine-grained.

Coarse-grained sentiment analysis allows us to look at sentiment at a wider perspective, where we can find and decipher the sentiment on a full document or full sentence level. There are two components to coarse-grained analysis: subjectivity classification and sentiment detection. Subjectivity classification determines whether a sentence is factual or opinionated. Sentiment detection determines whether a sentence has a sentiment and, if so, whether it is positive or negative.

On the other hand, the scale is smaller with fine-grained sentiment analysis. We try to identify the topic of a sentiment on the sub-sentence level. The sentence is first broken up into different relevant pieces. Then, every part is looked at in relation to the other sub-sentences, which allows analysts to identify the subject of a sentence and the target of the feedback. It helps businesses understand why a writer would express their opinions in a certain way. Fine-grained analysis helps us understand polarity, which defines categories in ways you might often see in surveys, with levels such as “Very Positive” to “Very Negative”.

Figure 3: Visualization of Polarity from Negative to Positive (Source: Customer Sentiment Analysis)

Let’s take a look at an example.

• Text: Oreos are much better than Nilla Wafers or Chips Ahoy.
• Analysis: Oreos are the subject in this sentence; they are being compared to the two other products: Nilla Wafers and Chips Ahoy. The sentiment is negative towards the other products due to the use of the words “much better”.

Here’s another one:

• Text: I was trying to use my microphone’s mute button, but the terrible thing was broken!
• Analysis: The subject is the microphone, the feature being discussed is the mute button, and the sentiment is decidedly negative due to the consumer describing it as “terrible” and “broken” , which leads businesses to realize their feature needs to be improved.

Due to both fine-grained and coarse-grained sentiment analysis, businesses can gauge the reception towards products and features on a large scale.

Ethics and Other Concerns

There can be ethical issues with businesses using sentiment analysis for financial and commercial gain. While companies may very briefly inform you in fine print that your information and online activity may be mined and analyzed for profit, many people do not suspect that their data is being collected. Voluntary participation and informed consent are crucial ethical considerations in research, but sentiment analysis can sometimes blur the line between legality and morality. There is also a chance that when sentiment analysis is used by large corporations for commercial gain, it has the potential to negatively impact individuals or groups.

Natural language processing and artificial intelligence as a whole can be especially prone to human biases. On the surface, having machines learn which words are associated with positive and negative sentiments can seem innocent enough. When we look deeper, it is entirely possible for sentiment analysis models to learn discriminatory bias from its inputs, and may somehow distribute human discrimination and biases into its results. Luckily, tools and metrics are being developed to analyze the bias of sentiment analysis models so that people can identify and mitigate model bias. In addition, machines cannot always be 100% accurate when interpreting human emotions. Machines may fail to recognize tone and sarcasm in text, unlike humans who can generally pick up sarcasm through text and speech quite easily. It is possible to get sentiments and opinions wrong, and decisions made off of inaccurate information can have real-world implications for a brand.

Sentiment analysis can provide great insight into people’s opinions and feelings. It allows data scientists and analysts to identify whether a customer has a negative or positive view of a business’s products or services. With that knowledge in hand, businesses can create strategies that leverage consumer sentiment and make changes for the future.