Inzata Announced as a 2020 Gartner Cool Vendor in Data Management!

Business IntelligenceData Analysis

Modeling Intent & Anticipating Outcomes with Sentiment Analysis

By July 13, 2020 No Comments

Sentiment analysis is one of the more established areas in the modern fields of statistics and machine learning. It’s widely used by many businesses with data operations to model consumer intent and to anticipate outcomes, particularly in the world of marketing. Let’s look at how this analysis works, why companies employ it, and a few particular challenges you should keep an eye out for. We’ll also explore some use cases along the way.

How Sentiment Analysis Works

Generally, sentiment analysis is run on bodies of text. A data scientist will collect sentiments from a specific set of sources, such as news articles or social media feeds. Marketers rolling out a campaign for a new sneaker, for example, might pull all of the Twitter feeds of known influencers and their followers who’ve mentioned something about the shoe.

Sentiments will be categorized by using one of two methods. The analysts will either:

  • Use an existing corpus of classified words with strongly associated sentiments, such as “good,” “bad,” “cool” and “fun”
  • Develop a corpus by training a model based on selected entries that are classified by humans

Once the analysis is run, each entry will be scored as “positive,” “negative” or “neutral” in sentiment. This data can then be used to develop insights about how the rollout of the marketing campaign for the sneaker is performing.

Why Do Organizations Use Sentiment Analysis?

Sentiment analysis is typically meant to measure performance after the fact or to monitor response in close to real-time. A company might use sentiment analysis to break down customer reviews on Amazon, and they would then use the insights to address the most common issues that caused negative sentiments. National political campaigns, on the other hand, might be interested in seeing how messaging performs in real-time. A candidate’s team might monitor Twitter sentiment to see how statements during a debate prompted certain responses, for example.

These approaches can be useful in an array of jobs. You might want to:

  • Scan all comments on a forum to filter out spammy or useless statements
  • Review customer service logs to identify and reconnect with consumers who simply quit on their interactions
  • Identify which influencers prompt the best engagement when they speak to their followers
  • Monitor your brand’s reputation over time
  • Determine who is excited about a pending product launch

The best organizations in this sector don’t just monitor issues and deal with them. Many actively seek to anticipate and address the concerns they see. Expedia, for example, used sentiment analysis to identify the growing annoyance that TV viewers had with an ad featuring a violin. Rather than just withdraw the ad, the company created a new one where the violin was destroyed.

What to Watch Out For

Several challenges tend to emerge when using sentiment analysis. These include problems like:

  • Listening to the whole world instead of your established customers
  • Depending on machines at the expense of having humans deal with issues
  • Labeling data poorly
  • Excessive elaboration alongside minimal action
  • Conducting analysis before a statistically meaningful set of sentiments has appeared
  • Identifying problematic word usages, such as slang and sarcasm

It’s important to understand that a host of problems can emerge while modeling intent and trying to anticipate outcomes. Biases can be induced by:

  • Making subjects aware that they’re being monitored, potentially leading to gamesmanship, anger or taunting directed at your organization
  • Publishing standards that third parties can play to, such as search engine optimization standards
  • Narrowly defining the data set, leading to selection biases
  • Training models based on one set of cultural norms, such as taking a Eurocentric view while doing global analysis

Conclusion

There is an old maxim in the world of data science: “All models are wrong, but some are useful.” It’s wise to internalize that idea and move forward. 

A good data operation seeks to achieve continuous improvement. Especially in sentiment analysis, it’s essential to evolve as the world evolves. By staying aware of the potential pitfalls of the process, sentiment analysis can help you respond quickly and competently in an ever-changing cultural, political, and economic environment.

Back to blog homepage

Scottie Todd

Scottie Todd

Digital Marketing Assistant

"Level 4 marketing wizard on a quest for data insights one blog post at a time."