In the days leading up to Superbowl XLVIII there’s a unique opportunity to capture insightful trends and patterns within social media.
Much of text analytics involves analyzing customer conversations, whether the conversations exist within social media, emails, forums, blogs, survey responses, or call center transcripts.
These conversations, just like the Superbowl tweets, are time sensitive. What is relevant today may not be relevant in a month, a week, or within the next 24 hours (think viral events). Similarly, if you contact a customer one week after they express anger, you miss the window to intervene and incentivize your customer to stay with your organization.
Below are some of the current trends and insights based on Superbowl tweets from the past two weeks.
Graph 1: Twitter volume over time for Denver (orange) vs Seattle (green). Also, what are the top hashtags and who are the most influential authors?
Graph 2: Who is winning the “Twitter Superbowl” based on fan support?
Graph 3: Do fans mention the Seahawks or the Broncos within the context of winning? How about within the context of losing?
Graph 4: Where are the Seahawks and Broncos fan’s located?
What does this have to do with your business? When analyzing text, there are a few key questions you may want to ask yourself:
Why are you analyzing text?
This question is fundamental, but is sometimes overlooked. Organizations know that they have all this textual data and need to be doing something with it, but often fail to define a solid objective that leads to ROI (More on ROI in upcoming blog posts).
- Do you want to identify data-driven trends? (often seen in marketing and customer intelligence)
- Are you looking for root cause or a needle in a haystack? (seen in fraud applications)
- Do you need to extract entities or facts such as IDs, names, demographic information, etc.?
- Are you using textual data to enhance your predictive models?
- Do you want to identify key influencers around a given topic or event?
What topics or categories align to your business requirements?
It's important to approach this from two angles:
- Use a data-driven approach to identify naturally occurring topics based purely on the data. Text mining, clustering, and natural language processing all help to enhance the statistical discovery of topics.
- Provide your domain-knowledge into the model, through business rules, that target the categories and topics you are specifically interested in based on your business requirements.
What data sources are you using (and how did you collect the data)?
Poor data collection methods lead to data quality issues and a large dataset with low relevancy. If you are collecting any data from online sources or 3rdparties, it’s important to understand the data collection process, filtering criteria, and queries, all of which could bias the data and introduce noise if not configured correctly.
- What kind of web crawling techniques/tools are you using?
- If you are using search terms to target and collect data, how did you choose these terms and are they limiting your results or introducing unnecessary noise?
What kind of action should the analysis elicit?
- Do you need a dashboard to monitor trends, influencers and viral conversations?
- Does your model trigger a promotional email, predict customer attrition, or flag a fraudulent event?
- Can alerts help your social media team or call agents proactively reach out to customers with timely offers?
In the days leading up to the Superbowl, I will continue to update the analysis and give you insight into emerging trends and interesting findings. Please check out the software behind the analysis, SAS Text Analytics and SAS Visual Analytics.
Check out the Post-Game Analysis for more insights.