Analysis of social media streams is usually restricted to just fundamental analysis and count-based metrics, i.e. how many people tweeted about that event, brand or product, who tweeted the most, and where those tweets are coming from etc. This is similar to scratching the surface and missing out on those high-value insights waiting to be discovered. So what should you do to help a brand take full advantage of the captured data?
Simple. Carry out sentiment analysis
MonkeyLearn defined sentiment analysis as a natural language processing (NLP) technique used to determine whether data is positive, negative or neutral. Sentiment analysis is often performed on textual data to help businesses monitor brand and product sentiment in customer feedback and understand customer needs.
Why sentiment analysis
Social media is an effective tool for reaching out to new customers and engaging with current ones. Positive customer feedback and social media posts encourage other customers to purchase from your company. The opposite is also true. Negative social media posts or reviews can be costly for your company.
Understanding how your customers feel about your brand or products is critical. This data can assist you in improving the customer experience or identifying and correcting problems with your products or services.
A positive customer service experience can make or break a business. Customers expect their concerns to be addressed quickly, efficiently, and professionally. Sentiment analysis can assist companies in streamlining and improving their customer service experience.
How customers feel about a brand can impact sales, churn rates, and how likely they are to recommend this brand to others.
In summary, sentiment analysis lets you understand how customers feel about that product, event, brand etc.
It is not just enough to run a count-based analysis. That's just scratching the surface. Consider the two scenarios below.
Mr A tells you Mr B, and Mr C tells him something about you. Intuitively you might want to ask what he said even if you eventually don't ask. You would intend to know if what was said about you was either good or bad.
Datafest Africa held an event recently, and I'm sure some are either in the process of scraping the data or have already done it. They should have at least 1000 tweets. That's a minimum of 1000 feedback. Reporting 1000 people tweeted about the tweet doesn't aid much with process improvements. It doesn't say much about what they did right. Those tweets could either be positive or negative. If I were on the team, I would want to know people's perceptions, not just the number-based metrics.
For instance, the team wants to hold the next event in another location. You can back up your choice of location with data to the group by saying we had lots of positive feedback from this location based on tweets made in that location during the previous event. I'm not sure of the level of data you can scrape from Twitter, but I'm assuming the tweeter's location would be one.
With sentiment analysis, you can take your insights a level higher.
The question remains, I don't know python, so how can I do this?
The good news is you don't have to worry about any piece of code.
Excel have an Add-in that lets us do you do this. Let's go into Excel to do this.
Go to the insert tab.
Select Get Add-ins
Search for Azure Machine Learning and add the add-in
Once added, you will have a pop-up section on the right side of your Excel.
Select Text Sentiment Analysis. The image below comes up.
The input range is the column containing the feedback/comments, while the output is where you want the analysis result. You get two results (two new columns)
Sentiment — This could either be positive, negative or neutral
Score — Ranges from 0–1. The more positive a comment, the number tends towards and vice versa.
Let's do a quick demo in excel.
Note that no model is 100% accurate. If a model is 100% accurate, there has been data leakage. For example, classification models aim to minimize false positives and negatives as much as possible and not eradicate them.
I changed the comments, and we have a more accurate sentiment.
I want you to take your analysis a step higher and not just leave it at count-based metrics.