A lot of us have probably heard about tools, platforms or applications used for sentiment analysis. For the last couple of years, companies offering such services have emerged from nowhere storming the doors of small and medium companies, huge corporations, as well as pr agencies. But does it really work as it should? Can you really let the computer analyse the conversation?
I have been testing several platforms, both paid and free, for several months now, trying to answer the questions above. The results so far, apart from waisted time and dollars, should not surprise anyone. Most of the conversations that were picked as negative by various applications, were in fact positive. The most surprising result was the number of neutral voices. It was extremely large in comprison with positive and negative sentiment. Does that mean, people really don’t care about your products or services? Well, not really. It only means that the computer doesn’t know whether it should be negative or positive. More than 80% of neutral responses were either positive or negative, but you can’t expect the programme to understand sarcasm, irony, metaphore, humor, or even a game of words, can you?
To give you the example:
Customer A: I tell u, this product is the best!
Customer B: oh yes, and I bet u’re the best worker in the company where they produce it.
In the above example, computer would count customer B as a positive voice. Why? It would take positive words like; ‘yes’, ‘the best’, and add it to words refering to ‘the company’ and ‘produce’ - the result - positive voice. The reality is, computer didn’t understand that Customer B was sarcastic, and that the comment was in fact negative.
So, does that mean we should not use these platforms or applications? The answer is - we should. But add one more ingredient when using it - your amazing brain. Double check every opinion, comment and every conversation. Remember that computer can analyze the chemical structure of a raindrop, but only YOU can put face in the rain and tell what it feels like.