The Sentiment of the Internet: What We Can Learn From NLP

As the virtual world continues to grow at an exponential rate, more and more businesses are turning to sentiment analysis for marketing and reputation management. What can we learn from computers that might be able to read online sentiment better than we can and why is it worth it?

Sentiment Analysis is the machine learning process of classifying the emotion or polarity of text as positive or negative, happy or sad, etc through Natural Language Processing or "NLP". A computer can pick out keywords and categorize them based upon surrounding context. As the algorithm being used gets more and more advanced, it gets better, smarter and faster at understanding human communication and the intended sentiment behind it.

The error ratio in sentiment analysis depends on how trained the algorithm is. Current use of Natural Language Processing for sentiment analysis has an accuracy rate of about 70-percent according to data scientists. However, even if it were 100% correct, humans would still disagree about 20-percent of the time.

How is sentiment analysis currently being used?

Public opinion and expression through social media, blogging, and various review sites have taken over market response and can easily become over-saturated. Businesses needed an automation process to filter out all of the noise and get right down to the core of so much instant feedback all at once.

In 2012 it was used in the London Olympics. Brand experience team Ignite and MIT graduates teamed up to create the first light show using the London Eye to show the sentiment of the world's positive or negative reactions on Twitter using hashtags based around the event. The wheel would turn yellow if response was positive, green if it was neutral and purple if it was negative.

Speakers also use sentiment analysis to get instant feedback on audience perception and initial reactions to their speech. TED Talk speaker David Eagleman used it to instantly gauge all of the tweets going out surrounding his talk as he was giving it. He was able to tell how much of the audience reacted positively and how much reacted negatively during any given portion of his talk.

It has been used to fight cyber-bullying by finding negative content on the internet and fishing out any type of potentially harmful online language to report.

But this is only the start of its use.

Why is it worth it?

Some people wonder how a computer can read the sentiment of online language when we as humans can so easily get it wrong. How many times have you misread the emotion or intention behind a text message?

While it cannot do so perfectly, it can do a great job of answering questions like these:
  • Are my consumers satisfied or unsatisfied?
  • Did this ad campaign draw positive or negative emotions?
  • How have bloggers' and news writers' attitudes about political candidates changed through this debate?
  • How do my consumers feel about the new changes to my product?
  • Is this person's actions online mostly positive or negative?
What is the future potential of sentiment analysis?

As a cofounder of social mobile app Shout The Good, we use it to hold people accountable for what they say and do online through providing a sentiment score - the more positive your posts online the higher your score. The more negative your posts online, the lower your score.

Communicating online can be a blurry and messy place with so many different forums to be able to express opinion while hiding behind a computer screen. The rise of anonymous social networking apps like Yik Yak and Whisper have too easily become forums for cyber-bullying and hate messages leading to devastating consequences. What would it look like with more transparency? Would people communicate differently and better? We can take it even further by asking if it can eventually be used to better pick potential job candidates that fit company culture or to find respectable companies to work for. With the growth of dating apps, can it be used for partners to better filter out a proper date?

As the virtual world expands, the use of sentiment analysis and natural language processing has a growing place in it. We are just at the beginning stages of using it to better understand human behavior in a virtual space and to hopefully encourage that behavior to become better.