Choosing the right Visual-AI tool for brand value

A.I. in the Visual Age

3.2 billion images are uploaded to social media everyday. That’s astronomical in comparison to just a few years ago. In fact, in 2014, the number of images uploaded daily to social media was just 1.8 billion.

There has been evident changes to the ways in which we consume social media content, with platforms such as Snapchat, Instagram, and Pinterest becoming increasingly popular. With this evident shift from text to visual, we can see how brands’ spending habits are adapting to suit new forms of advertising. In the past, brands spent most of their advertising budget on traditional media, but because of this change in the industry, they have had to review how they market themselves.

There has also been an upsurge in influencer marketing in which sponsorship deals and partnerships are being taken over by social media influencers and popular events, such as sporting events and concerts. However, with this new way of advertising becoming a dominant force in the marketing world, brands have noticed difficulties in finding ways to measure the ROI these sponsorships generate, both offline and online. Brands should be aware of the value of social media and how they can track, measure, and monitor these sponsorships in order to calculate their ROI.

But how can they do this? The answer is: Artificial Intelligence.

So what is Artificial Intelligence (A.I.)?

The definition of A.I. is rather self explanatory. It is the theory, development, and advancement of computer systems that carry out a simulation of human intelligence or behaviors. So, for example, humans have the ability to recognize logos quickly and easily themselves. However, it’s obvious that manually detecting logos at scale using the human eye simply isn’t feasible for brands. What logo recognition can offer Social Media Monitoring companies is a chance to automate those manual processes of recognizing logos for the brands that use their services.

But why do they need logo recognition in the first place? What value would that bring to them? Detecting hashtags and keywords is no longer enough for brands if they want a true picture of their data and insights. Put simply, what logo recognition can do is detect a brand’s logo on social media without the need for those keywords or hashtags, which would solve the issue of monitoring the ROI of sponsorships, as well as monitoring images of brands’ products on social media that users share.

Ok, but is it really worth it? Can’t brands just stick to monitoring text mentions and forget about visual mentions altogether? Well they could, and some brands do, but as a result they are neglecting far more insights than they may realize.

Over 80% of images uploaded to the internet do not contain any textual reference to the brand present within that image. In other words, without some form of logo recognition, they are unable to find the majority of images that contain their brand’s logo as there are no accompanied textual references to the brand that they can monitor. Which means brands are missing out on copious amounts of insights if they are not using logo recognition.

The A.I. Rollercoaster

So what’s the deal with A.I.? Well since its inception, there has been a continuing hype cycle surrounding A.I.’s activity. It appears that every time we encounter a breakthrough, we experience a downfall sometime after. Since investments in A.I. are continuing to increase along with this rollercoaster, these ups and downs A.I. is going through, it begs the question: what’s next for A.I.’s current cycle? Is another crash imminent?

We are constantly being bombarded with A.I. pitches, and if only a small proportion of A.I. companies are delivering a valuable product to tech and analytics companies, then it is likely we will experience another downfall in A.I. sometime in the foreseeable future, meaning a decline in A.I. investments. But what does all of this mean for Social Media Monitoring and other forms of data analytics? Well, it would explain why people make mistakes during the implementation and usage of A.I. tools because there is always so much noise and hype surrounding new developments.

A more scientific approach would be a lot more helpful, both for the analytics companies and for their brands (customers) rather than jumping on the many A.I. bandwagons that may come along. It’s important for them to make sure their logo recognition provider has taken the time to perfect their technology and aren’t falling for the new trends A.I. brings with it. Otherwise, the quality of the technology will not be as good as it could be and instead, they will be over-promising on tasks their technology simply cannot achieve effectively.

A.I. and the Value of Logo Recognition

So how do you ensure that your logo recognition provider is bringing real value to brands? It’s important to establish what your brands want from logo recognition. What insights are important to them? What brands are important for them to detect? Do they only want to detect their own logos or do they also want to detect their competitors’ logos? Complete and accurate insights provided at scale and in real time are what is needed in order for brands to fully benefit from logo recognition.

Brands need to know their audience and fully understand their demographics and psychographics that they want to gain insights from, otherwise their logo recognition provider will be unable to optimize their offering to suit their use case.

There is a difficulty in choosing the right A.I. vendor, however, because A.I. itself is so complex. Many providers will show examples to their clients of what their technology can do face to face, which may seem great on the surface, but oftentimes there is no emphasis put on the quality and accuracy of the technology. There is a much more pragmatic way to go about it in order to combat common mistakes and to ensure the technology being offered is suited to specific use cases.

Establishing a benchmark, i.e. training the system with images containing the logos and images without the logos, is so important. Take both sets of images to your logo recognition vendor and ask them to run the system through the images or through an API and measure the following two metrics: precision and recall.

Precision involves how accurate the detections were, i.e. whether or not the images that were detected actually contain the logos we are looking for.

Recall determines whether or not the technology detected all possible images containing those logos.

Establishing which metric your brands want to maximize will be important at this stage. Collecting a dataset reflecting the desired use case is also important, as well as making sure to go to different vendors to figure out which provider is best for you and your customers. It’s crucial for the sample to mirror what you want the logo recognition to be used for as accurately as possible in order to ensure the most beneficial and valuable results.

Scalability is another metric that is important so that the technology can learn new “concepts”, which would be logos in this case, quickly and be able to deliver the same precision and recall performance. This is vital in order for the system to be able to take on new logos at speed, which means both the vendor and the client need to be able to respond quickly. Finding a simple way to communicate with the A.I. vendor is crucial, but it is also important for the vendor to respond just as fast. Interfacing with the partner to ensure the correct and feasible volume of images and their commercial scalability is so important.

Logo recognition in action

But what will the outcome of logo recognition be? Will it be any different to how it was in the past when visuals weren’t as prominent? The outcome should be the same as it was before with text-only monitoring, but there would be more content to analyze, providing brands with more value thanks to the inclusion of visual.

Let’s look at an example of what logo recognition should look like and the metrics that are important for successful detection results. Taking images of cats as an example, we can see below a benchmarked result containing the relevant metrics we would need for detecting cats within images. In this case, the logo recognition server scanned a number of images in order to determine whether or not these images contain cats.

Looking at these results, we can see that the logo recognition server detected only two images it recognizes as containing cats.

As I’ve said previously, there are two important metrics needed in order for us to determine whether or not this logo recognition vendor is good enough for what we need: precision and recall. Precision involves how accurate the detections were, i.e. whether or not the images that were detected do in fact contain cats. Recall determines whether or not the technology detected all possible images containing cats. Now let’s examine what these specific results mean.

• Why is precision 100%? The images we have detected do contain cats, therefore our precision rate is 100%.
• And why is recall only 50%? We have failed to detect all the images that contain cats as we missed two out of the four images, meaning our recall is at 50%.

It’s important to determine which metric is most valuable for your customers. Is precision more important than recall? Or are both equally important? If they are both necessary for you (which is likely) then this logo recognition provider wouldn’t give you the optimal results.

Another logo recognition provider has scanned the same images and this set of metrics render different results. Let’s take a look at what they mean.

• Why is precision 80%? Although we have successfully detected the images that contain cats, we have also detected an image that does not contain a cat. This is known as a false positive, which decreases the overall precision rate.
• So why is recall 100%? We successfully detected all images that contain cats, despite there being one false positive present in the detection results.

If keeping the number of false positives to a minimum is important to you, then perhaps this logo recognition provider isn’t the right one.

Finally, one more logo recognition provider scans the same images with results containing 100% accuracy. Let’s take a look at what they mean.

• Why is precision 100%? All the images detected contain cats and there are no false positives present.
• Why is recall 100%? We successfully detected every single image containing a cat.

If 90-100% accuracy for both precision and recall is important to you, then this logo recognition provider would be ideal.

To sum up, A.I. is a complex set of technologies where the hype surrounding it can lead to cycles of downfalls. This often makes it difficult for social and digital analytics companies to figure out which logo recognition supplier to buy their technology from and can lead to mistakes during that decision-making process. Logo recognition is an essential tool for Social Media Monitoring companies to provide to their customers if they want them to have the best possible and most accurate results from their data generated from social media.