Sentiment analysis

1. Sentiment analysis

In the last lesson, we alluded to social media comments being a source of text for analysis. The type of text analysis most commonly performed on social media data is called sentiment analysis.

2. Sentiment analysis basics

Analysts can often find social media data overwhelming - it spans many platforms, like Twitter, Instagram, and Facebook. But it also can be noisy without the right analysis techniques. One way of monitoring social media health is via sentiment analysis. Sentiment analysis is a technique to assess if data is positive, negative, or neutral. For example, "great service" sounds like positive sentiment, while "horrible service" sounds overwhelmingly negative.

3. Sentiment complexity

As we can imagine, sentiment analysis can be subjective, which makes it complex! Brands want to be able to monitor sentiment across different social media platforms in real-time. One of the reasons that sentiment analysis can be difficult to scale is that context is very important. If someone uses the term "well done" while congratulating a competitor at a sporting event, it is very different from asking for a well-done steak at a restaurant. Another complicating factor is that sentiment can be expanded beyond "positive/negative/neutral" to emotions like angry, happy, or sad. Sentiment can even include timing aspects like "urgent" versus "non-urgent." When we add in the social media elements of emojis or sarcasm, sentiment analysis can become unwieldy for analysts to apply at scale.

4. Sentiment model types

Fear not, Marketing Analysts! There are multiple ways to approach sentiment analysis, depending on the scenario. The two main options are either: manually create rules and apply them or use NLP modeling to automatically determine sentiment. NLP can take words that are commonly associated with sentiment and classify for us. Fortunately, there is no one right modeling approach; analysts should evaluate the level of complexity and size of the data first. Automatic sentiment analysis works best with large datasets containing multiple languages. Manual rules are easier to interpret but are harder to scale beyond a small data set. Some analysts may find that a hybrid approach works best, where they use manual at first (especially if there are rules specific to our business) to train an NLP model later.

5. NLP model options

At this point, we may be thinking, "NLP seems to have a lot of benefits, but how do we get started?" Luckily, many marketing tools offer some type of natural language processing. One option is to use an out-of-the-box model that assigns sentiment for data within that social media platform. Another option is to build an in-house NLP model. For analysts that are new to NLP, an out-of-the-box model is a great starting place. If an analyst is knowledgeable about NLP modeling, training a custom model will be more flexible over time. The last factor analysts should consider is reporting needs; these can influence how much time to invest in maintaining a model.

6. Let's practice!

Let's practice some sentiment analysis!

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.