Measuring Sentiment With WebFOCUS Social Media Adapters

By Efrem Litwin

“People are talking about us but what are they saying?”

In today’s society, social media tools such as Facebook and Twitter give people a forum to express their thoughts publically on a wide variety of subjects. People express their thoughts through posts and comments on Facebook or tweets on Twitter. They might be discussing life events or a wide variety of topics and experiences. Comments can express positive and negative sentiments – or no sentiment at all.

For example, a comment like, “I bought this new car last week and I am so happy with it,” would be an example of a positive sentiment. A comment like, “This is the worst airline I’ve ever flown on,” would be a negative example. Something like, “I’m sitting and watching television,” would be considered neutral, which is to say expressing no sentiment at all.

Because of the power implied in the expression of sentiment, an increasing number of companies is using social media as a tool for making key business decisions. For example, a particular manufacturing company might want to determine the reasons for a sharp decline in their sales for a few months. The company’s financial statements indicate only that sales are declining because their customers are not buying as much as they used to. But why is that? Customers, in general, would buy products based on demand. If there is no demand, the manufacturing company would see a decline in sales.

So this company could look at all its internal reports and still not find the answer to what actions are required to increase sales. Instead, the answer may lie in what the general population thinks about the company’s products.

Let’s say the company produces all different types of widgets. We are now going to turn to social media to try and determine the answer for the decline in sales. As people use social media to convey their thoughts about a particular subject, it is probable that we can find the answer in the sentiments they express. For now, we will be using Facebook and Twitter to determine the answers.

On Facebook, the company would create its own page for users to post comments. The comments are either in reference to a post from the company itself or a comment about the company’s products.

For example, a company might post the following on its Facebook page: “Next generation of widgets to be released next month.”

A Facebook user might comment on that particular post as follows: “I’m very unhappy with the quality of your widgets. They always break after a month of usage.”

This negative comment about the quality of the product could provide a clue to why there is a decline in sales. The company now has to determine if this particular person’s sentiment is isolated or shared by other users of the widgets.

In Twitter, people are able to create a comment of up to 140 characters in the form of a tweet where a company’s Twitter screen name or products could be referenced. The products referenced could be in the form of a tweet grouping known as a hashtag grouping. For example, typing #widget in a tweet allows the Twitter search engine to return tweets containing that particular hashtag.

So, a Twitter user might tweet the following: “I’m very unhappy with the quality of widgets. They always break after a month of usage #widget.”

With so many people using Facebook and Twitter, a company now has to be able to sift through all of the comments to determine what is relevant in helping it make good business decisions. So getting back to our original case, how can this company who manufactures widgets use social media, sifting through Facebook posts/comments and tweets, to quickly pinpoint the main issues as to why their sales are declining?

With WebFOCUS 8, Information Builders released our first set of social media adapters:

  • Facebook adapter
  • Twitter adapter
  • Words Analysis adapter
  • Wand Sentiment Analysis adapter
  • Alchemy Sentiment Analysis adapter

The Facebook adapter allows WebFOCUS to be used to report on Facebook posts and comments for specific Facebook pages. Using Create Synonym for the Facebook adapter, metadata and sample WebFOCUS reports are created in a specified application. So, for the company that is manufacturing widgets, a WebFOCUS report can be created to pull all posts and comments for a specified timeframe from the company’s Facebook page.

Screen 1 shows an example of a WebFOCUS report on Facebook posts and comments for the Information Builders Facebook page.

The Twitter adapter makes it possible to use WebFOCUS to report on Twitter Tweets based on specific search criteria. Using Create Synonym for the Twitter adapter, metadata and sample WebFOCUS reports are created in a specified application. So, for the company that is manufacturing widgets, a WebFOCUS report can be created to pull tweets searching keywords like #widgets or the company name.

Screen 2 shows an example of a WebFOCUS report reporting on tweets searching on “Information Builders.”

The Words Analysis adapter is used to perform word counts on textual data. Using Create Synonym for the Words Analysis adapter, metadata and sample WebFOCUS reports are created in a specified application. A default stop words file also is created containing words to be excluded from the analysis. Textual data is passed to the Words Analysis adapter in the form of a Join. The field containing the textual data is joined to the DOCUMENT fieldname defined in the wan_document metadata.

wan_sample_join is a sample report that joins textual data contained in a sample file to the Words Analysis adapter. The report displayed is a list of the words analyzed and sorted from the highest to lowest occurrences.

wan_sample_join_tagcloud is a sample report that joins textual data contained in a sample file to the Words Analysis adapter. A tag cloud graph is displayed showing the words analyzed with the higher occurrences in a larger font than the lower occurrences.

Screen 3 shows an example of a tag cloud graph analyzing the posts and comments from the

“Information Builders” Facebook page.

So, getting back to our widgets analysis, Facebook posts/comments and tweets can be joined to the Words Analysis adapter to display the occurrences of the analyzed words in a tag cloud. Words showing the highest occurrences appear in a larger font. A drill-down report can then be used to display the posts/comments and tweets that contain the word that is clicked on. This gives an indication of what most people are saying about the company’s widgets.

Screen 4 shows an example of a drilldown report listing the posts and comments for the

“Information Builders” Facebook page containing the word “developers.”

The Wand Sentiment Analysis adapter and the Alchemy Sentiment Analysis adapter are used to perform a sentiment scoring on the textual data passed to it. The difference between the two adapters is that Wand Sentiment Analysis is purchased software and Alchemy Sentiment Analysis is a subscribed service. The sentiment score ranges from -1 to 1 where -1 is the lowest negative sentiment, 1 is the highest positive sentiment, and 0 is a neutral sentiment.

Using Create Synonym for the Wand and Alchemy Sentiment Analysis adapters, metadata and sample WebFOCUS reports are created in a specified application. Similar to the Words Analysis adapter, textual data is passed to either the Wand Sentiment Analysis adapter or Alchemy Sentiment Analysis in the form of a Join. The field containing the textual data is either joined to the TEXT fieldname defined in the wandscore metadata or the DOC fieldname defined in the alchemy metadata.

wand_sample_join and alchemy_sample_join are sample reports that join textual data contained in a sample file to the respective Sentiment Analysis adapter. The report displays the textual data being analyzed and sentiment score of the text.

Screen 5 shows an example of a report analyzing the posts and comments from the “Information Builders” Facebook page with a sentiment score next to each post or comment.

The widget company can use the Sentiment Analysis adapter to quickly determine which posts, comments and tweets are negative in nature without having to read each one. A tag cloud drill-down report can used display the sentiment score of the posts, comments and tweets for the words that are occurring most often. The drill-down can also be a graph visualizing the percentage of negative, positive, and neutral posts, comments and tweets.

To summarize, combining the functionality of the Facebook, Twitter, Words Analysis, and Sentiment Analysis adapter could prove useful in identifying problem areas and aiding in making the appropriate business decisions to increase sales. The widget company can use the social media adapters to determine if the posts, comments and tweets that reflect a negative sentiment are related to a specific issue.

The Facebook adapter pulls the posts, comments and tweets from the respective applications, the Words Analysis tag cloud graph visualizes the most used words, and the Sentiment Analysis adapter indicates which posts, comments and tweets are positive, negative, or neutral.