Organized using attention-need-satisfaction-visualization-action framework for a final coursework in business communication class

Attention

Social media is a fundamental part of people’s life now. Each day, there are 5 hundred millions of tweets sent by billions of twitter users around the world. That is around 6 thousands per second! It’s really a lot, and among this ocean of words, contains a really valuable insight about our customer that have the potential to shape our future.

And that is, what they truly feel about us. If our customers are happy about our coffee, they may say things like “Oh my god this coffee is so amazing” or “This coffee tastes so good”. If they disliked it, they will post negative comments such as “The roast is too dark, didn’t enjoy it” or “This coffee is horrible #badcoffee #ihatemylife”.

If we can analyse thousands, or even millions of tweets like this on a daily basis and get a detailed report about it, we can adjust our efforts in a precision that we’ve never achieved before.

Need

In the past, we collect the opinions from our customers by encourgaing them to scan a QR code on our bag and submit a questionaire. However, from our previous study, we found out that around 95% of our customers did not scan or even notice the QR code. This means that we have lost 95% of the potential feedbacks to our product, which could have been used in our decision making process.

Although we did try to analyse the customer’s sentiment on social medias before, we did that my assigning actual workers to search for tweets and manually connect with them to get their opinion. This created a huge labor overgead for us, and the efficiency was not good either.

Satisfaction

Therefore, we want to introduce machine learning to solve this problem. Current machine learning algorithms can nearly do anything now, and when related to extracting people’s opinion, there are specialised machine learning models that can do “Sentiment Analysis” task.

By using such a model, we can classify the emotion of a tweet into 3 categories, negative, neutral or positive. The model can also extract the keywords related to emotions.

Visualization

For example, here is an experiment we conducted on our customer’s attitude towards our customer service. We found 10,000 tweets and implemented a model to classify their sentiment. As you can see in this bar plot, most of them are satisfied with our customer service. For those who hated our customer service, the biggest issue they pointed out is that the response time is too slow. So, we might need to improve our response speed later.

Action

As you can see, we got a detailed report on the customer’s attitude towards our customer service, and we know what we need to do to improve this situation. I really hope that we can apply this technique to a wider range of our operation, and make it be a really helpful tool for our development.