Samsung is living a marketing nightmare with its Galaxy Note 7 phone. After several reports of the phone overheating and catching fire, the company had to issue a global recall. And, as if that was not enough, the phone’s replacements have reportedly been catching fire, too. So much so, that the company has, now, stopped sales of the phone, and urged users to turn off their devices.
I think we can all agree that the situation is pretty dire. Though, you would not know that, based on some sentiment analysis tools. Here is one example:
Admittedly, this is a FREE tool, and there are others in the market which will have higher accuracy rates than this one. However, even those alternatives will struggle to accurately classify the present Twitter conversations about the phone. And one of the key reasons for that is that many of the conversations, such as this one that I spotted yesterday, are heavy with humour and irony:
Here are a couple of examples of humorous tweets being classified as positive by this particular tool:
The other source of errors is the use of terms such as ‘love’ and ‘friends’ which are undoubtedly positive, except that they are not being applied to the object of analysis (i.e., the brand):
I am not picking on this particular tool! Sentiment analysis is complicated, particularly for short segments of text such as Twitter messages. So, what can we do, if we want to know what consumers think about our brand.
The first thing to do, is to carefully test the tool that we are planning to use, specially around its ability to discern irony and identify the object of the conversation. Then, we need to look beyond the scores and heat maps, and see what is it that people are actually saying about us. In particular, we want to look for messages that talk about core reputation elements (e.g., food safety for restaurants, user privacy for health or financial service providers, and so on), or containing emotional responses (e.g., fear). Furthermore, we need to consider the context within which the conversation takes place, such as surges in activity, or what’s happening in the news. And last but not least we need to keep adjusting the dictionaries used to classify the tweets, so that they reflect the specific syntax and style used in social media conversations. In Samsung’s case, the words ‘alert’ and ‘power’ need to be given a negative score.