Web

Analytics

precision

There are different approaches to sentiment analysis, a major component of social listening. If you want out-of-the-box results, precision should top your list of “must-haves.” Here’s why.

Precision drives action

You can’t do anything meaningful with vague results. As nice as it is to know a lot of people like your brand – or don’t, as the case may be – that doesn’t give you a lot to work with. What kind of messaging can you offer, and what kind of innovating can you do, without specifics? These insights only allow you to paint in broad strokes and hope you’re making the right moves.

The whole point of analyzing sentiment is to unearth more accurate information to drive business decisions – so if you still need to make assumptions to fill in the gaps, you’re doing it wrong. Doing it right means getting past the surface layers of sentiment extraction to the gooey caramel center.

Here’s what we mean. Sentiment extraction has three overall layers:

  1. Classification is a basic thumbs-up or down designation. Examples: “I hate technology!”, “I love hamburgers!”
  2. Association, which ascribes a sentiment to a topic or brand, e.g, “I love my iPhone!”, “I hate McDonald’s!”
  3. Fine-grained sentiment extraction, i.e., “gooey caramel center.”

It’s this last that provides the precise complexities to inform brand actions, offering insights like:

  • Who is sharing the emotion?
  • How intense is the sentiment?
  • What precipitated/inspired the feeling?
  • What brands, or brand features, are associated with the sentiment?
  • Is this person’s view of the brand negative or positive?
  • Does this consumer intend to make a purchase?
  • Is the sentiment competitive, e.g., “iPhone is better than Blackberry”?

And this is where alternative sentiment analysis systems – like machine learning – don’t measure up. Most learning systems stop at Classification, and sometimes Association. NetBase incorporates all three layers of sentiment extraction, using deep parsing to answer the questions surrounding social sentiment.

The grammar engineering difference

This isn’t to say machine learning systems are inherently bad – they aren’t. Machine learning is a robust option for document-level insights reliant on keywords. It’s easy to scale, and fast to develop.

BUT –

The coarse-grained approach of machine learning systems doesn’t work well at the micro-blog level – offering only shallow Natural Language Processing (NLP) insights, with no context for understanding.

To separate out the data that matters, you need a Grammar Engineering system, which analyzes based on sentence structure versus keywords. This is ideal for short messages – like those on social media – offering highly precise, fine-grained insights:

keyword-v-parsing

 

You can see in the image above that the keyword/machine learning model accounts well enough for tone – accurately picking up on positive and negative words and phrases. But that’s all they are – singular notations of sentiment with no frame of reference.

The deep parsing/NLP model, however, recognizes the specifics associated with each positive and negative sentiment and puts them into context: It’s the coffee that tastes great, the Dell laptop that doesn’t boot, and the MacBook that’s easy to use.

Machine learning does have the capability to break sentiment down into different, more accurate components – but it takes a lot of work to account for the necessary variables, whereas our NLP system does it straight out of the box.

You don’t have time for assumptions

Why does this matter? Because time is the only constant in social media – as in, it’s constantly moving, with social feeds refreshing every second, offering more data to sift through. This data must be analyzed in real-time to be useful, which makes “straight out of the box” a more efficient option.

And, of course, with consumers taking center stage and wielding all the power on social, accuracy is table stakes if you want your messaging and other brand actions to resonate and keep you top of mind with your audience.

Going to the trouble of analyzing sentiment and only doing it halfway doesn’t make sense – though your competitors will love you. Better to apply your efforts to making sure consumers do.

Want a demo of our Natural Language Processing engine in action? Reach out!

Image from Jaume Escofet