People are complicated and nuanced – which is why social analytics tools are so crucial to understanding their wants, needs, and interests. Technology like Next Generation Artificial Intelligence (AI) makes that job easier. Here’s why you need it and how it works.
Diversify for Strength
Because there’s so much to learn about your audience – what they’re about, how they view the world, how they perceive your brand, etc. – it’s best to use a combination of tools to gain those insights.
Social listening, social monitoring, image analysis, competitive analysis, and more all have unique strengths that bring you the specific data needed for varying brand goals. The Next Gen AI technology powering these facets of social analytics tools works similarly.
AI is still a developing field, so there’s no one way to get all you need from it. For our purposes, we use three distinct types of AI tech to gather the highest level insights for brands. Because we believe transparency matters, we wanted to explain a bit about each:
Expert Systems
An expert system provides a way to encode human-designed algorithms – for example, determining the sentiment of a statement like “I love Netflix.”
It’s obvious to a human, but the algorithm has to be programmed to recognize the emotion of the verb “love” as positive, and to understand it’s directed at the object of the verb, in this case “Netflix.”
The advantage of expert systems is their high precision based on human knowledge of linguistics, which can be encoded into the algorithms they use. This is best for uncovering sentiment towards a particular object and sentiment drivers overall.
There are inherent challenges, however, as well. It takes time for linguistic experts to do their part. And lower recall means the system will miss some results.
Machine Learning
Machine learning works with algorithms that are not explicitly designed by humans, but are optimized via trial and error by a computer program.
For example, how do you teach a machine to distinguish between a bird and a sheep?
You can program in a set of rules for the machine to follow – if you know what the distinctions are yourself. But what if you don’t?
Machine learning enables you to introduce a dataset of animals already labeled as being either a sheep or a bird. From there you could use your intuitions about the types of features that differentiate all kinds of animals – like fur, number of legs, size, color, etc.
The information for a given animal might look like:
- Covering = Wool
- Legs = 4
- Weight = 60kg
- Color = Black
Feeding this data into a machine learning algorithm would teach it how to classify animals based on the data received. For example, if 90% of your labeled birds had feathers, the algorithm would learn that having feathers is strong evidence the animal it’s attempting to classify is a bird.
Google Photos results for “bird” include paper bats
When the foundation is laid properly, machine learning is best for high recall document classification – like language detection or ad/spam detection.
But just like with expert systems, there are drawbacks. Machine learning takes a lot of human effort to ensure the data being processed is accurate. For instance, in the example above, if the algorithm for birds was trained using only flamingos, it might learn that birds are pink. As a result, it would not classify penguins or ravens as birds – even though they are.
Because of the effort required to expand what needs to be learned at any time, machine learning is less flexible than expert systems.
Deep Learning
Deep learning is a specialized form of machine learning that transforms layers of data. So, while the classification result of a “traditional” machine learning algorithm might represent a single layer based on the specified features input, a deep learning algorithm will consist of many layers – hence the “deep” qualifier.
This allows for a more complex data process, but with less human manipulation.
With the bird versus sheep algorithm above, for example, instead of defining number of legs or, the presence of beak, fur, feathers, you could provide the algorithm with information like the shape of the feet, the head, and the texture of the body. The algorithm would then learn how the combination of these features add up to 4 legs and wool.
This circumvents the need to predefine the features we think might be necessary to differentiate our categories.
The disadvantages aren’t very surprising. More complex algorithms take longer to be “trained,” and require specialized GPU (graphics processing unit) machines.
Their ability to learn their own features makes debugging the algorithms a challenge. With expert systems and machine learning, it’s easy to identify the cause behind why the system classified an animal as a bird or sheep.
In deep learning, we don’t know what features the algorithm is using, which can make it difficult to correct, and even require changing the structure of the algorithm. Understanding how deep learning algorithms make their decisions is still an active area of research.
Still, it is useful for finding patterns in large datasets, like assigning “meaning” to words or sentences in pre-trained vectors and some image analysis tasks.
How Do These Technologies Work Together?
What makes Next Gen AI best in class is the combination of these AI technologies. Together, however, they power NetBase’s language analytics – the heart of sentiment analysis – and other social analytics modalities to provide comprehensive data insights.
Here are some ways Next Gen AI breaks down social data.
Classification of known categories – aka, labeling data with a set of known categories. For example:
- Classifying an object mentioned as positive, negative, or neutral
- Identifying a post author as male, female, or neither
- Identifying geographical location by city, DMA, state, and/or country
- Recognizing known logos within an image
Classification provides the surface analytics that tell you a post may have value to your brand.
Discovery of novel insights – or going beyond known categories and letting AI do the work of finding useful insights below the surface, like:
- Sentiment drivers, i.e., why sentiment is expressed towards an object in a sentence
- Trending items likes hashtags and themes
- Conversation clusters, or posts grouped together naturally by topic, without pre-defined themes
This is the data that clues you in to potential actions you can take to connect with an audience authentically.
Learning, to customize results of Classification and Discovery – which is the best part. Expanding on the NetBase AI’s pre-trained social content, your feedback allows the system to adapt to your brand’s specific use cases.
For example, you can change the sentiment of a sentiment driver to teach the system how to correctly classify similar posts. Or give a label to a set of clustered posts – the system will learn to correctly label similar posts going forward.
Individuality Matters for Brands as Well as Consumers
The beauty of Next Gen AI is the way it allows brands to adapt the way analytics are processed – so they provide the intel you need to compete. Just as consumers want to be seen as individuals, your brand’s needs are unique to your goals.
Meeting those unique needs takes an assembly of top-notch tools, driven by technology that can keep up with the pace of change in your vertical, on social, and in the world at large.
As AI continues to evolve, so will the way we use it. But no matter what, we’ll always ensure your results are fast, accurate, transparent, and easy to access. Anything else couldn’t be considered Next Generation.
Want to see our Next Gen AI in action? Schedule a demo with a member of our awesome team.
