With 15 million people now actively using Twitter in the UK, representing around a quarter of the population, this offers huge potential for marketers to glean customer insight like never before. By mining text in the micro-blogging stream, subjective information can be captured and analysed, and if done right, it can provide valuable information that can be used in business planning to improve customer service, develop propositions and ultimately increase sales.
Sounds easy doesn't it? However, it can be all too easy to draw the wrong conclusions. Here are ten top tips to help marketers understand how good Twitter sentiment analysis works and how to make the most of the data.
1. Context is everything.
Sentiment analysis tools pick up on words that express positive or negative sentiment as tagged as such in the lexicons used to score overall sentiment. But what if your intention was to illicit a negative response and evoke feelings of sadness or to make someone cry? A simple sentiment analysis tool won't have the knowledge of that context or intent, and so while your campaign may have been successful in provoking a 'negative' reaction in terms of emotions felt by the audience, your sentiment analysis tool may not see it that way.
Knowing the context of the response you want to illicit is crucial. You can then make sure your word lexicon correctly scores words as positive or negative according to the context. If attention isn't paid to this detail, the overall score won't reflect the true results of your campaign.
2. Be cautious of standard analytical packages.
As we've established, many off-the-shelf sentiment analysis tools don't take account of context. But, if you do your own analysis, standard analytical packages often can't cope with large volumes of unstructured data. Furthermore, to have confidence in the conclusions you reach, and to be sure they are robust and valid, you'll need an analyst with a proper mathematical background. It takes a trained statistician to avoid the common pitfalls, such as the number of observations and grouping data to ensure statistical significance. Only then can you have confidence that you are taking action on something more than a random result.
3. Be clear on your objectives.
When you have clear objectives, you can make sure you collect the right data needed to measure success against these objectives. When you look at big data sources like Twitter, this becomes even more important. Although it has a 140-character limit, Twitter generates over 8TB of data every day. Collecting data without knowing your objectives, or what you want to measure against, takes a lot of effort while delivering potentially meaningless results.
4. It takes some serious data processing
About that 8TB of daily data - even the subset you want is a lot of data to download and process. To help, there are several Firehose providers available that allow full access to all Twitter data. Using one of these is preferable to trying to pull the data through the Twitter API as they remove some of the barriers set by Twitter's API restrictions. However, you will still need expert application programmers to use the Firehose.
5. Get your data ready for analysis
People use random characters on Twitter, including emoticons. These will all have to be converted and interpreted to get a true picture. You'll also need to de-dupe the data and get it ready for your analytical package. There are tools out there to help you do this, such as Mapreduce and Hadoop; while they can help speed the process up, it takes specialist skills to do so.
6. Your data's got breadth
Twitter's mission is "To give everyone the power to create and share ideas and information instantly, without barriers". Twitter users actively share content and so the original sentiment expressed about your brand can be perpetuated through replies and retweets. All of this content, tweets, retweets and replies will need to be considered as part of your analysis.
7. Are you measuring reach or impact?
It's often assumed that those with a large Twitter following have a greater influence over their followers. But seeing a negative tweet doesn't necessarily mean your attitude will also be negative. Sentiment tools often include measures of reach; however, this isn't necessarily a measure of impact. So, consider weighting sentiment by the number of people that have seen the tweet and the likely impact this may have had on overall sentiment.
8. Consider the impact of verified accounts
Twitter proactively works to establish the authenticity of tweeters and verifies accounts accordingly. These accounts can have a huge impact on your analysis. Firstly, verified accounts are often associated with organisations or celebrities and have very large followings. As such, they can have a disproportionate impact. Secondly, if the organisation or celebrity has a consistent tone of voice and is always positive, negative, cynical, sarcastic, or other 'persona', once re-tweeted it is likely to have an impact on your overall campaign sentiment.
9. ...and brand tweets
As brand messages are generally positive, this can skew any analysis in a similar way to tweets from verified accounts. Consider the impact of removing or keeping brand tweets within your analysis.
10. Beware of comparisons
Where a tweet expresses a comparison of one brand with another, it will often be along the lines of "company X's customer service is great but I hate speaking to company Y" or "company Y's widget is rubbish compared to company X's". Under analysis, the negative sentiment may cancel out the positive, producing a neutral result when in fact the tweet contained a clear expression of positive sentiment towards company X and negative sentiment towards company Y. Spotting and disentangling these comparisons when processing your data is important in establishing a robust foundation for your analysis.
Like any analysis, sentiment analysis is not straightforward and it takes proper analytical minds to avoid the common mistakes around misinterpretation and subsequent potentially poor decision-making. If you're relying on the conclusions of sentiment analysis to make marketing and business decisions, you need to be absolutely confident in how the data is being treated. So check, is the context right? Was anything included that may have skewed the outcome? Did the analysis look at specific words or were entire phrases and sentences considered?
Get it right and sentiment analysis provides a fantastic opportunity to tap into a rich seam of information but be sure your analysis is telling you the true story.
By Nick Evans, Consulting Marketing Practice Director at Jaywing.
PrivSec Conferences will bring together leading speakers and experts from privacy and security to deliver compelling content via solo presentations, panel discussions, debates, roundtables and workshops.
For more information on upcoming events, visit the website.
comments powered by Disqus