10 Sentiment Analysis Issues to Be Aware Of

Sentiment analysis’ popularity has grown exponentially in recent years and it is now being
used to track customer reactions, monitor competitor positions, anticipate election outcomes, forecast investment trends, and predict box office revenues. Speakers at the Sentiment Analysis Symposium on Tuesday in New York included corporate users as well as suppliers in the field, all providing different perspectives on this complex topic.

Chris Frank, VP at American Express, emphasized the need to focus on the learning obtained
from sentiment analysis, since, to date, more resources have been devoted to listening. Wayne St. Amand, marketing VP at Crimson Hexagon added that it’s critical to understand social intelligence, including the drivers of sentiment that help explain the underlying reasons behind comments.
Despite the amount of time devoted to listening and coding sentiment, accuracy
is still an issue. Speakers from Attensity, TolunaLexalytics, and the American Red
Cross explained the complexities of properly coding sentiment using automated and manual methods. So whether you’re monitoring social media in-house or using a specialized vendor to track your clients’ brands, below are ten issues to pay particular attention to.
  • Sarcasm: is one of the most difficult sentiments for automated tracking to interpret properly. Example: “It was awesome for the week that it worked.”
  • Navel gazing: is when social media tracking turns up items related to your own promotional efforts, and should be filtered out.
  • Neutral sentiment: is similar to the concept of swing voters, and Frank recommended dividing it into specific themes to uncover more detailed opinions.
  • Relative sentiment: is not a classic negative, but can be a negative nonetheless. Example: “I bought an iPhone” is good for Apple, but not for Nokia.
  • Compound or multidimensional sentiment: contain positives and negatives in the same phrase. Example: “I love Mad Men, but hate the misleading episode trailers.”
  • Conditional sentiment: includes actions that may happen in the future. Example: the customer isn’t angry now but says he will be if the company doesn’t call him back.
  • Positive feelings can be unrelated to the core issue. For example, many comments about actors focus on their personal lives, not their acting skills.
  • Negative sentiment is not necessarily bad: This relates to the classic PR dilemma regarding negative publicity. Example: Sarah Palin’s appearance on the Today show generated many negative comments but still drove ratings increases.
  • Ambiguous negative words: Their context needs to be thoroughly understood and tagged accordingly. Example: “That backflip was so sick” is really a positive statement.
  • Beware of Google translate syndrome: Western and Asian sentiment differ greatly, as
  • do the meanings of their emoticons, so they need to be interpreted correctly.

    The Wall Street Journal’s weekend Review section uses NetBase data to track Twitter and Facebook sentiment around timely topics. (see image above) Last week, they analyzed the public’s reaction to the recent high sale price for the famous Munch painting, “The Scream.” While the sale price prompted mixed views, the extreme sentiment expressed in the painting itself was universal.