Sentiment analysis can provide a rich seam of data which can be a great way of unearthing real customer insights like never before. However, as with any analysis, it’s not as straightforward as it seems. Without the right analytical minds behind it, the results are all too easy to misinterpret, to draw incorrect conclusions from and, ultimately, to make poor decisions on as result.
These tips are designed to help marketers understand what good sentiment analysis looks like.
1. Context is key
If your intention is to elicit a specific emotional response, simple sentiment analysis tools won’t take that into consideration. Words that appear to express negative sentiment are tagged as such in lexicons and used to score overall sentiment. For example, you might want to elicit a “crying” response, but this would be scored as negative, rather than positive. So it’s important the word lexicon correctly scores according to the context of your brand or campaign. If not, you’ll end up with an overall score that isn’t an accurate reflection of what you’ve achieved.
2. Standard analytical packages can’t cope
Having accepted that off-the-shelf tools don’t always take account of context, if you choose to conduct your own analysis, standard analytical packages can’t cope with the large volumes of unstructured data. And, if you want to be sure your conclusions are robust and valid, only an analyst with a solid mathematical background can help avoid the common mistakes related to things like the number of observations of a particular sentiment and statistical significance.
3. Know what you’re measuring
Having clear objectives means that when you set out, you will gather the right data needed to measure against these objectives. And when considering big data sources such as Twitter, which generates over 8TB of data each day, this becomes even more important.
4. Data processing at a serious scale
There are several Firehose providers available that allow full access to all Twitter data. These remove the restrictions set by Twitter’s API, but if you want to use them, you’ll need expert application programmers who can interface with these providers.
5. Prepare your data for analysis
What do the emoticons mean? To get a true picture of what people are saying, you’ll need to convert and interpret any rogue characters, de-dupe and prepare the data for analysis. Tools like MapReduce and Hadoop can help speed up the process, but it takes specialist skills to do this.
6. There’s breadth to Twitter data
The whole ethos behind Twitter is to share information. With sentiment analysis, anything that spreads and shares the original sentiment, including retweets, will need to be considered as part of your analysis.
7. Reach≠ impact
Sentiment tools often measure reach with the assumption that a large following means greater influence. This isn’t necessarily the case. A negative tweet seen by a large number of people won’t necessarily mean the attitude of those viewing it will also be negative. Consider weighting sentiment and the impact this may have on overall sentiment.
8. Don’t let verified accounts skew your results
Verified accounts are often associated with organisations or celebrities and can have a huge impact on your analysis. This is due to their disproportionately large followings and, if their tone of voice is always positive (or negative or sarcastic), once they are re-tweeted it can have an impact.
9. …or brand tweets
Owing to the fact that brand tweets tend to be positive, they can also skew any sentiment analysis. The impact of these brand tweets should be considered when undertaking your analysis.
10. Comparisons can produce neutral outcomes
Where tweeters have compared one brand with another, a neutral result can be the outcome when using standard sentiment analysis tools. Even if there was a clear expression of positive sentiment about one company and negative sentiment about another, the negative sentiment will cancel out the positive. Being able to spot this at the data processing stage is important in establishing a robust foundation for analysis.
Like all analysis, sentiment analysis is not as easy as it first appears. This is where experts with the right analytical minds come in. If you’re relying on the conclusions of the analysis to make decisions, you need confidence in the way your data is being treated. Follow these ten tips and you’re already on your way to ensuring your analysis is painting a true picture of customer sentiment.