8 Keys to Meaningful Social Media Sentiment Analysis

The accuracy of social media sentiment analysis tools has improved alongside the maturation of social media industry itself. With any automation technology, a sufficient investment in research and development, as well a series of iterations based on user feedback, will ultimately improve the results of that technology.

Will social media sentiment analysis tools ever have 100% of the accuracy of human analysis of content? Probably not for a long time. Will some of them hit 90% in the next few years?  Quite possibly.

Before even considering an investment in a sentiment analysis tool, your organization needs to have a high enough number of qualified mentions — at least several thousand per month.  Katie Delahaye Paine of ragan.com suggested that the number should be at least 10,000 qualified mentions per month.

Social Media Sentiment Analysis Keys

Ultimately, a social media sentiment analysis tool needs to provide actionable information — but the tool needs to give a sufficient level of confidence to marketers and product managers that their actions are being taken based on as accurate information as possible. Here are some of the key attributes that a sentiment analysis tool should possess in order to provide that level of confidence.

1. Index all corners of the Internet – Consumers and writers spread their praise, their suggestions, their preferences and their criticism all over the Internet. Sentiment should be gathered from not only social media feeds, but from the public Internet, content publishers and other areas. For example, a lot of consumer sentiment is expressed in blog and article comments.

2. Index very large amounts of data – The vendor should have have a highly scalable architecture that can source hundreds of terabytes of information. Naturally, the greater the percentage of qualified mentions that can be indexed the better for accurate social media sentiment analysis.

3. Pre-accumulate information – Data that’s relevant to a customer needs to exist in a data store before a customer even signs up. The customer should not have to wait for data that’s relevant to their brand to start accumulating after they sign up, as this can mean months before a large enough sample size is available and before important trends can be revealed.

4. Account for nuances of language and how people express themselves – One person’s “good” is another person’s “bad”. The adjective “sick” can be a younger person’s highest compliment.  Profanity directed at a brand does not necessarily mean a disdain for the brand.

5. Interpret sentence structure and not just words – People often express more than one thought or opinion in a comment, a post or a Tweet.  The context around a word can give a word either a positive or negative connotation. The ability to analyze sentence structure to infer sentiment is critical.

6. Analyze sentiment on multiple levels – For consumer sentiment to be useful, it needs to be more than just a smiley face or a frown indicator (despite the catchy image used in this post). Relative preferences are important as are emotions and behaviors associated with a brand.

7. Allow for user refinement of queries – a well engineered social media sentiment analysis platform should allow for the exclusion of irrelevant words, phrases and even entire domains from a brand query. If users can fine tune a social media sentiment query engine for their specific needs and projects, the results will be much more accurate and meaningful.

8. Longevity in the field – A brand new startup may claim that it has the best algorithms ever for analyzing social media sentiment. However, no matter what level of intellect and creativity the application’s developers possess, there’s no substitute for product iteration over the course of time.  Many of the most important refinements to a lot of technologies can only happen with the help specific customer feedback over a time period of years, not just months.

While there are a variety of sentiment analysis tools on the market, the best results will come from an application with the attributes listed above.  Are there other important attributes that a social media sentiment analysis tool should have?