I see businesses struggle with how to benefit from sentiment analysis. Many reviewers and analysts have covered this topic: see this post from Jason Falls. What are the major problems with automated sentiment analysis? accuracy, reliability and variation in the results from different tools (discussions abound about what is the percentage of accuracy for a given tool) and the presence of noise. My goal in this post is to provide you ways to maximize what you can get from today’s sentiment analysis technology.
Here’s my quick guide on how to effectively use automated sentiment analysis in social media.
- Spend time to research the keywords which are best fit for your business objectives and optimize your search based on the result set that you get.
- Use sentiment analysis to detect change in sentiments over a reasonable duration, say 3-months or longer. Shortcuts don’t work well.
- Competitive Analysis – compare your results against those of your competitors.
- Don’t miss the forest for the trees – sentiment analysis is more valuable in the aggregate. Explore the drill-down capabilities but don’t get lost on individual messages.
- When using a tool, look for the capability to re-classify the sentiment, which should train the tool.
- Don’t underestimate neutral comments – there are plenty of business insights and business opportunities there.
- Source is important: Twitter users tend to complain more. Facebook fan pages tend to have more rich media associated with content.
- Understand the limitations of software-based automation: irony and sarcasm are not well understood by software. Content with mixed opinions are hard to catch.
- Context is key for improving accuracy of sentiment analysis. Optimize for your industry or vertical.
- Combine user and content ranking with sentiments to prioritize your results.
I believe that one of the reason for the disappointment with sentiment analysis is the unrealistic expectations from what the science of today can deliver in terms of understanding human sentiments. With the explosion of user-generated content, Sentiment analysis has become one of the active areas of research in Computer Science and Linguistics (for example, follow this link for a recent paper on sentiment analysis).
I plan to write more detailed posts on the topic of sentiment analysis including the various approaches and options beyond fully automated analysis – call it semi-automated sentiment analysis where the tool does more than 90% of the work in automated fashion. I will also share case studies of how MutualMind has helped its clients with sentiment analysis.
I hope this was helpful – please suggest some additional points and share your experience.
Sentiment without action is the ruin of the soul.
— Edward Abbey