Sentiment analysis is the analysis of text by natural language processing or computational linguistic to understand subjective opinions. It’s important when looking at social media as it goes beyond the qualitative measures such as engagement.
To give an example; your latest post generates huge numbers of comments which results in a high engagement score. But when you dig a little deeper you find that all the comments are negative, even angry, your post has infuriated your customers. It’s sentiment analysis that can automate this qualitative information source.
It’s a rich source of information for analysing online content about your company particularly ratings and social media. But it’s not easy to do.
If someone writes a post about Amazon are they talking about the retailer or the river? To make a good analysis for Amazon you would need to find a way to exclude posts made about the river. If your company name is also a normal word or family name this gets more complicated.
Even within one language words can have different meanings, depending on the source country; thongs means underwear in the UK, but footwear in Australia. Fanny is problematic in the UK, but not in the US. Schoon means clean to the Dutch, but beautiful to the Belgians. Coger is benign in Spain but obscene in Mexico.
Some population groups use slang that is difficult to untangle; in English “bad” can mean good, “wicked” could be positive and “groovy” could mean outdated.
So understanding the use of some words will require knowing the context, and that’s difficult to automate.
Some posts are clearly negative or clearly positive, but we use sarcasm which is hard to read.
“Restaurant X lost my booking for 10 people, brilliant”
Sometimes the slightest change in the subject of a sentence makes all the difference, this example came up on the wiki page.
“They would not let my dog stay in this hotel”
Which is somewhat negative, but explains hotel policy, and probably wouldn’t deter too many visitors. Changing just one word makes it very negative;
“I would not let my dog stay in this hotel”
Sentiment analysis in languages other than English is less robust, and some social listening tools don’t cope with languages outside Europe at all. The best social listening tools allow you to manually adjust sentiment analysis labels on a post because this is so tricky to analyse
When monitoring a company announcement a while ago, when we expected reactions in English and Dutch, we didn’t wait for the automated sentiment analysis from the social listening tools. We automated the social listening search to match keywords around the announcement and then manually watched the tweets and posts across the screen. This worked as we were covering the short period of the announcement and there were relatively few mentions. It’s not a long term solution.
Sentiment analysis will get better; it’s becoming a vital tool for understanding consumers’ opinions across digital media.
Image: Sentiment Analysis, my image