Social Media Returns
Anticipation is high for the payback on social marketing investments. The verdict is out for capturing any quantifiable results. A recent CMO poll cosponsored by Duke Business School and Deloite Touche indicates even with an average of 13% of marketing dollars being spent on social marketing only 30% of the respondents indicating that they have shown quantifiable value. Pressure will increase significantly to show value with social marketing estimated to grow to 21.5 % of the marketing budget over the next five years.
Social Media Value – Lagging Indicators
To achieve quantifiable benefit, social media must deliver outcomes with business value. Social media must facilitate the journey through the marketing funnel, for users, customers, and clients. For eCommerce, the outcome is a loaded shopping cart and commitment to buy. For concerns, offering services and/or building brand image, it is attracting eyeballs “from the wild,” and migrating from reach/impressions to followers. Using tools like, Google Analytics, progress is tracked using lagging indicators. The intent is to determine what was an effective post, but resorts instead, to just comparing summary statistics for a given time period. This can show that the process is directionally correct but:
- No detailed data exists on how the how the post-performance was attained
- No clear direction on how to maintain or change or change course for the future
Even sentiment analysis has focused on what has happened. It is used to sense public opinion from social media posts or internal customer issue tracking or product feature feedback. Companies use sentiment analysis to gain insights so they can provide differentiated services, valued product designs, frictionless user experiences and refined business processes. Sentiment analysis is used to determine the predilection or opinions of the author. But again, this is something that has already happen, which is like, “trying to drive a car by looking in the rear-view mirror.”
Sentiment Analysis – Influencing Outcomes
Sentiment analysis can be performed using several approaches. They tend to fall either in the supervised learning or lexicon (bag of words) approaches. Supervised learning, large amount of unlabeled text is fed into algorithms. These algorithms use embedded words learn based on the coming text and develop a model which establishes the sentiment (positive, negative, or neutral) analyzing text sources including social media. Getting large amounts of text and defining the parameters for embedding terms for teaching the model are two of the challenges using this approach.
Bag of Words uses a pre-defined lexicon. There is significant effort to build a lexicon, especially if a crowd sourcing (sending out surveys, online tools) is used. This builds a lexicon with each word being assigned a positive or negative score. The scored words with rules for addressing context and syntax define a sentiment score at the sentence and paragraph levels. This gives an overall score plus insights how the chosen words effected the sentence and or paragraph sentiment score.
Using tools when creating all content influences the reader to continue reading, become a follower, or act. Applying sentimental analysis cannot address all the variability for social media outcomes, it is a part of the variability that you can control. Using a lexicon and consistent approach to content development will influence the audience because, the words make a difference. They make a difference because research has shown that social media with higher positive content
- Increase the quantity of social media traffic and its speed through social media channels
- Used with environmental factors improves the predictability of social media diffusion
- Increased positive content increases social media acceptance and diffusion
The Lexicon Challenge
Having the “right” lexicon is the challenge when taking the supervised approach using a lexicon. How do you manage the effort of creating the “right lexicon,” by audience but not introduce bias? How do you get unbiased word scoring? How often should it be updated? Is a general one-size fits all lexicon the answer or something specific needed for each audience? What choices exist in developing or collecting public domain lexicons and developing audience specific lexicons?