Business creates special channels for working with clients (hotline, website forms, chatbots, etc.), but if a customer is disappointed, they will definitely go to social media – the most habitual place for most of us.
Perhaps the post will go unnoticed, or maybe it will pick up a lot of reposts, comments, and likes – causing anger. In this case, silence can be costly.
Reputation is easy to lose, but difficult to recover.
To prevent such splashes of negativity, we use social media monitoring software, almost all of which are able to recognize the sentiment of mentions.
Read on to learn how to work with sentiment and what other benefits we can gain using social media sentiment analysis.
What is sentiment analysis?
Sentiment analysis is a fundamental social media monitoring metric that helps analyze the online landscape quickly and effectively.
This social listening feature presents an instant overview of all mentions in the monitoring stream, regardless of one's level of experience with analytics software.
Using this function, you can view positive and negative conversation trends, see how the target audience reacted to the latest activation from a competing brand, and quickly respond to a negative comment from an influencer.
The below graph shows the basic sentiment of social conversation over time.
Examples of how to measure sentiment of social conversations
1. The distribution of sentiment in dynamics: discussions over the topic of alternative energy and electric transport, as seen below.
2. Sentiment in trends: These are trends detected in the industry of alternative energy and electric transport. Mentions containing negative sentiment are marked red, positive-sentiment mentions are green, and neutral mentions are blue. (Here, blue highlighted news “Volkswagen will be spending $800 million in Chattanooga, Tennessee. It will be making Electric Cars”)
3. Sentiment in sources: Same topic.
4. Sentiment by authors: This screenshot covers the topic of Tesla activities and products.
5. Sentiment in links and hosts: Tesla topic.
Present-day sentiment analysis provides an opportunity not only to assess the sentiment of brand mentions, but also get a whole range of additional tools that simplify communication with target audience, establishing contacts, sharing information, searching for influencers, supplying them information, and engaging in brand promotion.
Understanding social media sentiment
Mention sentiment data can enhance your social media efforts.
There are a few important things to keep in mind while working with social media sentiment:
1. The basic idea is simple: Sentiment can be positive, negative, and neutral. But the impact of each type of sentiment is different.
Remember the rules of human psychology: People are more likely to share something negative than something positive.
If you have more negative than positive mentions under the “Analytics” tab of your social media monitoring system, this is not a communication crisis, but rather a working process that you just need to resolve efficiently and quickly.
At a minimum, classify negative messages into categories, determine the types of negativity (constructive, emotional, trolling), draw up a disaster plan (scheme for the transfer of mentions requiring reaction to the appropriate specialists of your company), and develop scenarios (not scripts) of communication with dissatisfied customers.
Negative customer feedback is an opportunity to find out about problems directly – bypassing opinion polls and costly analytical research. In social networks, users say everything they think, not what you want to hear. This is an opportunity to make your products, services, and your business better.
For example, the Lego brand has long been analyzing its tone of voice and users’ mentions sentiment in social media. Lego has reached the level of full predominance of positive mentions over negative ones.
2. Some important numbers to keep in mind: the post author's Facebook friend count and the number of their Twitter followers and YouTube subscribers.
These are the first people to see the author's posts, and they're the ones in the author's social circle and immediate sphere of influence.
3. If you found a positive mention about your brand, take a closer look at the publisher's social media following. If the blogger has more than 1,000 followers or subscribers, be thankful for this unsolicited quality promotion. But if the same blogger left you a negative review, go back to point No. 1 and respond immediately.
4. You can dive deeper into the analytics and examine whether the source of the mention contains your target audience. For example, the social media groups, forums, discussion boards, etc. where the post was published. If you find a negative comment about a laptop in a cooking group, it might not affect your reputation too much, but that doesn't mean you should ignore it.
In order to best apply items No. 2 and No. 3 to your workflow, set up rules for a smart notification that can be sent to you as soon as the mentions are published.
How to use sentiment analysis to monitor social media
Sentiment analysis is an area of computational linguistics that deals with the selection of emotionally-colored vocabulary or emotional evaluation from texts.
At the moment, there are two most commonly used approaches:
- Based on machine learning methods, and
- Based on the use of sentiment dictionaries.
Recent developments in artificial intelligence (AI) have improved its methods of automatic sentiment detection, but AI still falls short when it comes to identifying the sentiment specifically as it pertains to the main subject of the mention. The same words can carry negative or positive connotations in different contexts.
For example, "Fairy is great for washing dirty pots," is a positive mention, while, "Coca-Cola is great for washing dirty dishes," is a negative comment about the beverage, as Coca-Cola isn't interested in being associated with a household cleaning product.
Advanced monitoring software works with an object-oriented model that uses the latest developments in big data and deep neural networks.
Training these models requires a dataset of more than 2 million marked-up mentions from a variety of fields – such as taxi services, online retail, finance, pharmacology, news media, politics, sports, food and beverage services, consumer goods, and more.
By constantly training the algorithm, data scientists achieve the accuracy of up to 90 percent in identification of positive and negative sentiment. In marked-up topics that contain a large data set of mentions with correctly assigned sentiment, that accuracy increases to over 95 percent.
At the heart of the model are deep neural networks with recurring memory layers, as well as layers that identify which part of the post needs to be analyzed based on the context, subject, and topic.
Thanks to a large training data set, developers are able to teach this model to understand the specifics of social media communication in order to accurately determine sentiment for different brands according to their unique demands.
With the solid foundation of a marked-up dataset, even tricky cases like the Coca-Cola example don't pose a problem – the model will "understand" the context.
In order to help the algorithm improve its sentiment detection accuracy, it is necessary to mark up some mentions with corresponding sentiment manually. This is especially important for new trends, emerging industries, and new word combinations.
Developers of monitoring software have made great strides in social media sentiment detection, but detection of more complex sentiment – such as sarcasm or irony – remains a challenge. (Despite the fact that scientists are constantly finding new methods to solve this problem. Recently, for example, MIT researchers stated that they taught the neural network to understand sarcasm through the connection between a certain language style and emojis.)
Sentiment in posts without copy
It's impossible to detect sentiment in posts without any copy. While image recognition tools can spot a logo in a wordless social media post, the context in which the logo appears can only be detected by manually viewing the photo or using Visual Insights – a graph of objects and settings that most frequently appear with a brand's logo. But this topic is subject for a separate post.
Users don't always explicitly mention your brand: Sometimes, they simply post a picture. Social media users can post photos of your brand in a negative context, modify branded images, or publish photos of counterfeit or pseudo-counterfeit products. Finding and mitigating these types of publications is especially important for online brand protection. So, you should be happy if the only non-verbal humor at the expense of your brand is limited to these kinds of images.
Nevertheless, technology is constantly improving, and the future of monitoring systems is to recognize the sentiment not only of the texts of users' messages, but audio and video.
Massive data scientist teams will likely solve these problems by teaching machines to understand all the intricacies of human speech in a not-so-distant future.
Ready to learn more about understanding your social audience? Find our the best social media marketing strategies you can implement in 2019.