In this guest blog post, Marcus Pemberton who is Global Head of Agency Account Management at Pulsar, shares how their social listening tool is adapting in an increasingly visual social networking environment.
The nature of how we use social media has changed. Think back to 2010 and how different it used to be – Facebook was king and the long status update, where one outlined thoughts, feelings, experiences and desires, was the norm. Compare that now to the 15 second Instagram story, the lack of text allowed within a Snap, and the rise of meme culture.
Our communications with our social networks have become increasingly visual and less wordy. Therefore, to understand that online behaviour, analytical tools must adapt too.
Pulsar’s New Visual AI
We’ve been leading the charge in visual AI for a number of years, incorporating IBM Watson image tagging and text recognition as standard within our data sets. However, acknowledging the cultural shift, we’re now testing out Clarifai technology that will allow us to delve deeply into the images that people are sharing.
The modules that we’re testing are: Food, Travel, Apparel, Color, Logo Detection, Celebrity, Video analysis and Demographics. For example – we can use the food module to understand exactly what food items people are taking photos of, without them mentioning it in text, allowing us to understand what is considered “Instagrammable”, and therefore giving us an unprecedented insight into consumer behavior. Perhaps more interestingly, we can also use the Demographics module to determine the age and gender of those within the image, and use logo detection to see what brands people find the most visually appealing.
Case Study – Ultimate Vegan Dish
For this exercise we set out to use the power of Vertical AI to run an analysis on over 15,000 vegan dishes, to identify the most common ingredients used. From this information we set out to identify the ultimate vegan 3 course meal recipe.
To begin with we set up search using Pulsar TRAC to pull in publically posted Instagram posts, published globally over a course of a handful of days, which were tagged as #VeganFood or #VeganRecipes. This allowed us to analyze nearly 15,000 individual posts.
Traditionally if we ran this analysis through a general image recognition AI module, it would produce a plethora of image tags which are not associated with our search, such: Plate, Glass, Elephant, Man, Fruit.
However, as here we chose the new Vertical AI image recognition module which has been trained solely to recognize food stuffs, we can get a much more accurate analysis of the image and what it contains.
The AI module picked up over 200 different ingredients and dishes. After hand-picking and eliminating a few false positives (some vegan good is really look similar to their non-vegan counterparts), we selected the top 20 ingredients. And the winners are:
To our dismay, Avocado didn’t top the chart, and surprisingly Chocolate was used in 36% of the dishes. Now using this set of ingredients we came up with a 3 course vegan feast to delight the most hipster of hipsters:
1) Bruschetta starter: Tomatoes, garlic, basil and olive oil topped on a slice of crunchy bread
2) Fresh Gnocchi: Potatoes for the Gnocchi, roasted corn, caramelised onions and pepper with chopped broccoli stems and sprig of basil and parsley on top
3) Spinach side salad: Salad leaves with Spinach, strawberry and balsamic dressing
4) Avo Chocolate Smooth dessert: Chocolate, Avocado, Ice and Bananas for a healthy smoothie
We had to leave out Apple and Rice in this menu, but we think through the power of the new Vertical AI, we have a winning Vegan meal here.