HOT TOPICS: Customer Experience Marketing Automation Social Business SharePoint 2013 Document Management Big Data Mobile DAM

Context is the Key with Sentiment Analysis #SAS12

Businesses using sentiment analysis need to tune out the hype around understanding sentiment to help predict customer behavior. The link between them is tenuous and complicated.

Lots of people are excited about social media, but it is too hard to make business decisions based on it, Kate Niederhoffer, a researcher with a Ph.D in social psychology said at the Sentiment Analysis Symposium this week.

Sentiments expressed in social media, the kinds that are easier to monitor and analyze, may be a good source for divining emotions or finding trends, but they don't necessarily predict behaviors, Niederhoffer said.

"Sentiment analysis seems like an okay place and seems like it's moving forward, but we shouldn't fool ourselves," she said.

image-sentimentanalysisconf-2012.jpg
Like genes, sentiment rarely causes a given behavior, thus, alone rarely leads to a business decision.

Academic Research is Spotty on Sentiment

Niederhoffer brought the rigor of academic research to bear on her presentation at the symposium, and her conclusion is that predicting customer behavior by looking at sentiment is highly context based. Tracking social media influencers and evangelists, and using sentiment analysis can help as long as there is somebody in the company who is immersed in the data and can readily see the important signals.

"We've made a decision to use sentiment to help explain behaviors. But, these behaviors, like the decision to buy something, are very complex, like why people vote certain ways," Niederhoffer said. 

Sentiment needs to be combined with company culture, processes and investment in social for it to be effective in making business decisions, she said.

While there is no real direct link between sentiment and predictive behavior, they are related. Studies do show the relationship between our moods and how we express them via Twitter, for example. One study released in late 2011 showed how tweets from certain times of the day correlated quite nicely with our sleep patterns and work day. 

Identify Things We Don't Know

It might be helpful then to start out by saying we don't really know if:

  • Sentiment captures expression and experience of affect
  • Sentiment predicts actual buying behavior
  • Executives believe in sentiment (and social media more generally)
  • Other data complements or social proves a leading indicator

"What is the right cadence with which we should look at sentiment compared to other metrics?" Niederhoffer said.

The key is to have the right mix within a company to do things like ensuring the sentiment analysis methodology is transparent. It's also important to explore sample characteristics through entity extraction (buying history, purchase intent) and use language processing.

Furthermore, it's helpful to substantiate the sentiment metric (standardize/weight according to events or influence) and build up the data set, Niederhoffer said. Immerse yourself or someone on your team in the data.

 
 
 
Useful article?
  Email It      

Tags: , , , , , , , , ,
 
 

Resources

 

Featured Events  View All Events | Add Your Event | feed Events RSS