Giving our clients Facebook and Messenger moderation was just the beginning! Now, we would like to introduce you to our newest feature: sentiment analysis in Sotrender!
What is sentiment analysis?
Social media managers, moderators, and CS specialists have to be in the know about the company’s image online. One way of doing so is by tracking reactions and feelings about posts on social media. This is essentially what sentiment analysis is: it analyzes the emotional charge of your users’ comments.
So what do we mean by “sentiment analysis” from Sotrender’s perspective and what does it do?
- Automatic analysis of a piece of text to identify emotional charge (“negative”, “neutral”, or “positive”). This includes labeling an entire group of comments.
- Task prioritization by filtering comments
- Ability to monitor company image
- Ensure quality of moderator or CS specialist responses
But don’t worry, we will explain each of these points in detail.
We instinctively know that specific words have an emotional charge. Sentiment analysis recognizes and quantifies emotional states that are directed at a post or comment. Posts containing words such as “thank you” and “awesome” would be labeled as positives. Whereas, posts containing words such as “hate” and “low quality” would clearly be labeled as negative.
Our model is based on an algorithm that automatically comes up with how each word in the sentence contributes to the sentiment of the whole text. It is worth noting that the understanding of sentiment differs between domains and our model is focused on analyzing sentiment of social media.
Here’s an example of how it can work for emojis, with a model prepared by our R&D team:
It’s important to note what exactly confidence means here. The more data is used to train our model, the more confident it is about specific labels. Ultimately, we choose the label with the highest confidence.
You’re not a robot, though. You can’t analyze so many comments at a time and calculate the sentiment efficiently on your own. However, AI and machine learning models can help businesses detect the emotional charge of a comment relatively quickly (with the right amount of training and quality datasets).
Before the model even starts the assessment of sentiment, it detects whether the text is in Polish or English. Of course, if there is a demand for it, we can create a model per language. When using the application, you may see the label “undefined” because our sentiment model can only assign sentiment to comments written in these two languages.
Sotrender’s model will label this text content as “negative”, “neutral” or “positive”. This makes it easier to gather statistics about your comments in Sotrender’s moderation analytics report. The more content (also known as training data in the Machine learning domain) you put into the model, the more confident the model becomes about its choices. It will eventually learn to differentiate sentiment of content even better than it did before.
What is the benefit of using sentiment analysis in Sotrender?
There are several reasons why Sotrender’s sentiment analysis is a must-have. We’ll walk you through each reason why you should be using our tool to analyze and manage your comments on Facebook.
Automation is the future
Social media managers lose time on marking whether a comment was positive, negative, or neutral. So wouldn’t it be perfect if a tool did that for you? That leads me to my next point:
Sentiment analysis in Sotrender generates an automatic assessment of the users’ comments. In a short period of time, you will see the full interpretation of comments on the right side. However, natural language processing models can take some time to learn the context of certain words. This is why it’s helpful to both us at Sotrender and to you if you correct the interpretations.
Changing sentiment manually is straightforward. Simply select the accurate label from the dropdown menu. The model will start to learn why one object is positive in one setting, and negative in another. Eventually, it also means that you can teach the model slang and jargon that will be easier to recognize later on.
Remember, as the comments keep growing, the model keeps learning. It will become more accurate at detecting the subtleties of sentiment in natural language.
For now, we have only prepared an English and Polish sentiment analysis capability. This is definitely a plus, even for English speakers who use a different dialect and ways of expressing themselves.
Sentiment of multiple comments
To properly understand a conversation, you need to see the full context. Usually, you will see users posting multiple comments back and forth. Those comments are likely to have a similar sentiment (either all positive or all negative). This aspect of our sentiment analysis allows you to save time analyzing each individual comment.
Once a cluster of comments has one or more comments that are negative, the “group” will be automatically labeled as negative. Even if you had multiple positive or neutral comments, the negative comment has more weight, and therefore changes the sentiment of the comment group. This design allows you to save time and prioritize which thread of comments you need to address urgently.
Here is how the group sentiment is weighted, from heaviest to lightest impact:
Because the neutral tag is the lightest, it has the least impact. This means that in order for a group of comments to be labeled as neutral, all of the comments in the group must be neutral as well.
So how can you use group comment sentiment to your advantage? Let’s say you want to announce a product launch on your Facebook Page. You can combine the results of Sotrender’s social media analytics and the moderation analytics report to check whether the product was met positively or not.
When you have many users writing to you, it can be difficult to decide which user you should address first. If there is a negative comment, you should probably direct your attention to the user so that you can change it to neutral at the very least. In fact, according to SmartInsights, 37% of users expect a response in under 30 minutes after they have contacted you. What’s more, 31% of users expect to be answered in under 2 hours. It won’t look good if you ignore their complaint while you’re answering all the positive comments.
Since we like making your job easier, we can help social media managers prioritize tasks with comment filters. We’ve talked a bit about filters when we introduced Facebook moderation in Sotrender if you want to refresh your memory.
Essentially, our users will be able to filter comments according to sentiment, meaning that you can address the negative and most critical comments as soon as possible. This feature is important when it comes to preventing a full-blown crisis from escalating.
Those of us working with social media know that your company could be one negative comment away from being involved in a crisis. One user’s comment could set off a chain reaction of negativity that is difficult to moderate. To further emphasize this point, consider these statistics:
- 49% of clients that had a positive experience with your brand are willing to talk about it on social media
- According to HelpScout, after a bad experience with customer service, 33% of Americans would think about changing companies
- One negative experience is enough for 51% of customers to never want to purchase anything from your company again
In such situations, you would benefit from having access to a social inbox with automatic sentiment analysis to manage user comments and your own responses.
Monitoring company image
To excel at community management and social customer care, you have to know what your customers think of your brand. You might be used to reading comments individually and checking the number of reactions. This is definitely not the best way to make data-driven decisions, nor is it the most efficient.
With the help of a SaaS tool like Sotrender, you will keep track of what your customers are saying about you. As a company, you will also keep track of your employees’ performance.
For the sake of openness and providing the best social customer care to your users, you need to manage your moderators. Sotrender’s team can provide you with a report where you can check the sentiment per moderator. You might have a moderator/ CS specialist who is particularly good at dealing with disgruntled customers, or one that is capable of answering in multiple languages. Whatever the case may be, you can organize your moderation team more efficiently to suit their talents and your followers’ needs.
By checking in on your moderators every once in a while, you’ll know which moderators are providing excellent social customer service. You will also be able to hold an employee accountable if a customer feels that they were wronged. This shows your audience that you are serious about providing excellent social customer service and you care about their feedback.
Sotrender’s sentiment report
There is something else to consider that our report can answer. Are you sure that you have the right number of moderators or CS specialists? Are all of them working on weekdays, or do you have someone working weekends?
You should check when most of your customers are reaching out to you so you can decide whether it’s worth it to invest in more moderators over the weekend or different time zones. On the other hand, you might find that you have an adequate number of moderators and that your plan is working out accordingly. Still, you won’t know until you check. 😉
As you can see, there are many positives to using Sotrender’s moderation tool for sentiment analysis. If you are looking for a way to prioritize your workload and manage it better, our tool can definitely help you. You can always test out our app for free for 2 weeks without having to enter any credit card details. 😉
Can you think of some other ways to use sentiment analysis for your business? Let us know!