10x popular : social media trends, office dogs & UX
Brands on social media: what your followers find important
Of course, it could be that someone shares this post because he or she is positive about it. But if no further opinion is shared in the message, I would rather find these messages in the 'neutral' segment.
Learning to interpret
The context of a post is also important. Suppose you receive the following two short comments under a Facebook post:
'Everything'
'Absolutely nothing!'
You would say that the first is positive and the second is negative. But what if these are two answers to the question: 'What can be improved next time at our event?'. Then the meaning suddenly changes. The analysis tool and/or employee should therefore include this information in the qualification.
Objective vs. subjective measurement
You have roughly 2 different forms of sentiment analysis: automated and manual. In other words, objective and subjective.
1. Objective measurement
A standard automated sentiment analysis is based on a list of Dutch words and word combinations. Word combinations then mean, for example, 'not good'. The word 'good' is of course positive, but in combination with 'not' it is negative.
In short, you can set up an automated sentiment analysis in three ways :
Rule-based : the sentiment is based on manually set rules
Automatic : the system determines the sentiment using machine learning
Hybrid : a combination of the two above
But even if you choose the combined Hybrid version: an automated sentiment analysis does not take into account double negatives, sarcasm, irony, emoticons, cynicism and comparisons. In addition, words in street language or youth language sometimes have a different meaning, such as 'cruel' or 'sick'. These words can be used both negatively and positively. Incidentally, more and more experiments are being done with this. Such as this experiment with Netflix reviews, in which different emotions in text are increasingly being recognized better.
Automated sentiment analysis can thus make a rough estimate of whether a message is positive, neutral or whatsapp number list negative. But it also has limitations.
In a manual sentiment analysis, an employee determines whether a message is positive or negative. Correctly interpreting double negatives, emoticons and comparisons is no problem.
But sarcasm, irony and humor are also difficult for a human to judge due to the absence of face-to-face contact. Of course, you cannot see whether someone is rolling their eyes, raising their voice or standing with their arms crossed. In addition, different people in different situations interpret messages in different ways.
Different emotions of people.
But, what now?
Manually classifying all messages? That is not possible. Only automatic sentiment analysis? You cannot draw good conclusions from that. A combination of these two can provide the solution, but it is not a panacea either. For now, it is a good idea, if you have the capacity for it, to combine automatic and manual measurement and not only focus on the negative sentiment.
How long will it take before a system comes onto the market that interprets sentiment (including videos, audio, images and emoticons) 100 percent correctly? Or is this a process that a system or robot will never be able to completely take over from humans?