• John Ennis

Eye on AI - July 22nd, 2022

Welcome to Aigora's "Eye on AI" series, where we round up exciting news at the intersection of consumer science and artificial intelligence!

 

This week, we’ll be addressing the differences between statistical and AI model predictions, including how AI can better predict purchasing behavior and enhance live video streaming viewership.


Enjoy!


How Statistical & Machine Learning Models Predict Purchasing Behavior



To begin, let’s take a look at some key differences between statistical and machine learning (ML) models, and the key differences in how each predicts purchasing behavior. According to the article “Artificial intelligence predicts what items land in your grocery cart,” Dr. Lawrence Yang, associate professor of information technology and cybersecurity at Missouri State University, believes the issue with statistical models is that they become flawed when too many dependent variables are added to the models.


“People have been researching [cross purchases in business decisions] for years using economic models such as regressions and other statistical models,” writes Missouri State’s editorial team. “But these statistical models are flawed… machine learning can calculate solutions with multiple variables without sacrificing reliability.”

Because variables are constantly changing, producing accurate, real-time predictions using statistical models is nearly impossible. With machine learning, models can incorporate new variables almost instantaneously without sacrificing reliability.


As an example, Yang points to the purchasing patterns of cake frosting. As you might expect, purchases of frosting tend to increase as the cost of frosting decreases. However, when looking at how a drop in frosting prices might affect the purchasing patterns of other items, outcomes are trickier to predict for statistical models.


“When looking at four variables – like cake mix, frosting, detergent and softener – the traditional econometrics model could perform similarly to AI,” continues Missouri State’s editorial team. “... However, adding four more variables overwhelmed the econometrics model. It took two weeks to evaluate the relationship between those eight variables. Using those same eight variables, AI and machine learning performed the analysis in only two hours.”

More nuanced variables make predictions even more difficult for statistical models. For instance, if someone were interested in looking at how things like location, weather, diet, wealth, etc. affect frosting purchases when frosting prices drop, results from statistical models could take ages to produce. ML models are able to consider all of these variables so long as relevant data sets are available, making ML’s predictions both more accurate and made in near real-time.


Using AI to Enhance Live Streaming Viewership



Speaking of real-time, another unique use case for AI is increasing the viewership of live-stream videos. A live-stream video, as the name implies, is a video that is aired live as it’s created. These types of videos are common among brands and influencers on social media platforms like Facebook and TikTok. According to the article “How Artificial Intelligence is Regulating Live Video Streams,” AI can be used to help streamers determine the best times to maximize exposure for their live streams, share analytics on how people respond to content, and suggest high-impact content for them to produce.


“TikTok, which lets users create short videos that can be shared with friends or posted on other social media platforms, is popular among young people,” writes Readwrite contributor Suvigya Saxena. “To keep up with the demand for new videos, ByteDance has developed a system that leverages AI to learn from user preferences and provide them with relevant content suggestions.”

Sharing content analytics, suggesting display times, and recommending high-impact content are all factors that AI can incorporate in other areas of content production as well. Think of a retailer trying to show the best products at the best time for engagement, or a publication company hoping to share more relevant articles to its users to reduce churn. These types of AI applications have grown in popularity among brands looking to increase personalization capabilities, and are now expanding into other areas of content production.







Other News




 

That's it for now. If you'd like to receive email updates from Aigora, including weekly video recaps of our blog activity, click on the button below to join our email list. Thanks for stopping by!