Sign up for our newsletter and get the latest big data news and analysis.

Exploring Customer Sentiment

Computational_marketingBIG DATA & CUSTOMER INTELLIGENCE SERIES

The first four installments of the Big Data & Customer Intelligence series focused on predicting and influencing customer behavior, looking at determining customer value, and understanding your customer’s social influence. In this final article of the series, we’ll take a hard look at exploring customer sentiment which is another important element of  Computational Marketing.

Does Customer Sentiment Really Matter?

First, let’s talk about how much the negative or positive expression of sentiment about a company or a product matter and in what context does it matter. There have been some big blow-ups in social media in recent years such as viral Youtube videos of the angry UAL passenger with a broken guitar, the anti-NBC campaign over the 2012 London Olympics coverage that was criticized for dismissing a tribute to victims of terrorist bombings, Bank of America being castigated for its plans for debit card fees, and many more.

As it turns out, most outrage expressed over social networks does not significantly add up to corporate damage, i.e. you can’t necessarily align blow-ups with a significant decline in sales. But monitoring sentiment is still important because if offers another valuable view of the customer. The new science of Credibility Analysis is coming on strong to add a new dimension to sentiment analytics (e.g. automated methods for assessing the credibility of Tweets).

A Balanced Goal for Social Sentiment

There are a number of goals that need to be considered when managing social sentiment. Don’t be caught up with your brand’s Facebook Likes or Twitter Follows. Your goal shouldn’t be to stir up positive sentiment by showing cute pictures or hip phrases. Instead, you should work to earn attention, consideration, awareness and purchase intent. The Progressive Insurance’s Flo character may have 5 million Facebook fans, but most were obtained by posting images of puppies. Don’t settle for likability, strive for trust. When the time comes for a customer to make a purchase choice it’s due to trust not funny Facebook posts.

As one example of putting customer sentiment analytics to work in a smart way, consider the PayPal example.  PayPal used sentiment and text analytics to identify specific issues with respect to customers and merchants. The company searched for negative impressions of its password recovery mechanism. As a result of what they found the company made an improvement to some specific password complaints. Following the change, they reapplied the sentiment and text analytics to see if the changes got a positive or negative spin. The process verified proof from the customer voices that the situation had subsided.

Tools for Determining Sentiment

There are a variety of technologies available to help determine customer sentiment. One of the most powerful methods is to develop machine learning algorithms (classification) for performing sentiment analysis, but this method requires data source(s). For example, the Twitter API allows you to obtain unstructured data and there are other free resources that enable you to search through Tweets by keyword to obtain sentiment. This is called exploratory analysis. The screen shot below shows the results of Twitter sentiment analysis for the keyword “James Jeans” which is a fashion brand. This free tool is called Sentiment140 but you need provide a Twitter authentication to use it. The small sample of Tweets used for this analysis shows 100% positive sentiment with Tweets such as “Wow. Great fashion find on James Jeans” and you’re able to give your personal opinion as to the accuracy of the sentiment by clicking on the Accurate or Inaccurate buttons.

Sentiment140

For production sentiment analysis deployments, a more robust data source is required. There are a number of pay-as-you-go vendors providing broad coverage of social media platforms, including historical data. Multiple data sources include: Twitter, Facebook, Tumblr, Instagram, Google+, Reddit, and blogs. Some data vendors include sentiment analysis so you don’t have to develop your own algorithms. One good option for production-ready social media data content, check out DataSift. Your goal in using sentiment analysis is to combine unstructured social media data with transactional data content for a comprehensive view of customer sentiment.

[I hope you've enjoyed the Big Data & Customer Intelligence series. Please leave your thoughts and comments below. We'd love to hear from you.]

Resource Links: