Can Large Language Models Make Even Better E-commerce Recommendations?
E-commerce giants, like Amazon and Walmart, have long recognized the power of product recommendations to create a convenient shopping experience and increase sales.
These recommendations often appear in sections with names like:
Frequently bought together
Related products
Customers also purchased
What customers buy after viewing this item
Many of these product suggestions are possible because of advances in machine learning over the past two decades.
So, with the massive success of ChatGPT and other AI systems based on large language models, is it possible to make leaps forward in how we suggest products to online consumers?
In this article, we will explore how traditional algorithms work and how AI-driven models could potentially provide even more customized recommendations for a next-generation shopping experience.
How traditional algorithms work
Imagine you're scrolling Netflix, which pioneered the implementation for many of the algorithms we use today. How does Netflix know what movies or shows to suggest to you? There are a couple methods.
One way is what we call collaborative filtering.
Think of it like this: Netflix keeps track of what everyone watches and how they rate those shows. Let's say you've watched and loved a lot of sci-fi shows. Netflix sees that other users who also love sci-fi have been raving about this new space exploration series. Based on this, Netflix suggests this series to you. In essence, this method looks at patterns in viewing behavior across users and recommends shows accordingly.
The other technique is content-based filtering.
This one is a bit different. Instead of looking at what other users are watching, Netflix focuses on the content of the shows you already like. For example, if you've been watching a lot of action-packed superhero shows, Netflix might recommend another show that's filled with epic fights and superpowers. In this approach, it's all about matching the characteristics of the show (like genre, actors, or theme) with what you've shown interest in before.
These methods have been amazingly effective, but they aren't perfect.
For example, they don't always take into account that you might be in the mood for a comedy movie after binge-watching a heavy drama series, or that you usually watch action movies on weekends but prefer documentaries on weekday evenings. However, adding more data can also degrade performance, so there is always a delicate balancing act when trying to make an algorithm more sophisticated.
Applying AI to the process
Just like how in our Netflix example where the app knows what movie to recommend next, e-commerce stores can use artificial intelligence, specifically language learning models (LLMs), to suggest products that match your tastes. These AI models, when tweaked just right, can grasp your preferences in a more comprehensive way and make product recommendations that are super personalized to you.
AI-driven models, especially LLMs, have the potential to change the way product recommendations are made by taking a more holistic approach to personalization. By fine-tuning generative AI models, retailers can create a recommendation engine that understands individual customer preferences and makes suggestions accordingly.
For example, consider a customer shopping for a new smartphone on a major online retailer's website. A traditional recommendation algorithm might suggest popular phone cases or screen protectors based on what other customers purchased alongside similar devices.
An AI-driven model, however, could take into account the customer's brand preferences, previous device purchases, and other items in their cart (e.g., a wireless charger). The AI model might then recommend a phone case that is not only compatible with the chosen smartphone but also aligns with the customer's aesthetic preferences and complements the wireless charger.
These AI-driven recommendations can consider:
Items in the cart: AI algorithms can analyze the combination of items in the cart and suggest complementary products, creating a seamless shopping experience.
Past purchases: By examining a customer's purchase history, AI models can offer recommendations that align with the user's previous preferences and needs.
Customer meta-data: AI-driven models can analyze customer data, such as demographics, browsing behavior, and location, to tailor recommendations further.
Here is a simplified example of a fine-tuned LLM that looks at some items in a cart, understands their relationship, makes an assertion about the use case, and recommends another product:
What it could mean for business
Let's talk business. Imagine you're running an online store. Having AI make personalized suggestions is a total game-changer, for a few reasons.
More sales: With AI, you can show customers items they're more likely to want, which could mean more stuff going from their cart to checkout. It's like Netflix suggesting a show you're probably going to binge-watch. You didn't know you wanted it, but there it is!
Better customer relations: Giving shoppers a more personalized experience makes them feel special, and that can build a strong bond over time. They're more likely to come back for more, which is a big win for your business.
Speedier shopping: AI can help shoppers find what they need faster, which makes their shopping trip easier and more efficient. Less time scrolling and searching, more time enjoying their purchase.
For the people shopping on your site, AI recommendations can make their experience smoother and more fun. They get spot-on product suggestions without having to hunt for them, saving time and taking the stress out of shopping. Plus, it makes them feel good about your store and more likely to come back and shop again.
Picture this: a customer is on your site looking for outdoor furniture. The AI remembers that they bought a patio umbrella a while back, and right now, they're checking out outdoor chairs. With all this info, the AI can suggest a set of chairs that not only match the umbrella's style and color but also fit their budget and comfort requirements. It's like Netflix knowing you're into feel-good comedies and recommending one that just came out — a perfect fit!
The next level: multi-modal models
AI is just starting to go mainstream, and it's already cooking up some new tricks, like multi-modal models. These are like the Swiss Army knife of AI. They can handle different types of data, like text and pictures, to make even better suggestions.
Imagine an online shop using this AI to analyze the words and images of products on their site. This way, the AI gets the full picture of each product's style, color, and pattern, and can suggest items that really match your vibe. By looking at all this data together, it gets a better understanding of what you're into and can make your shopping trip even more personalized.
AI-powered recommendations in the future
The basic "you might like this because others bought it" recommendations are getting a makeover with AI models, like the ones we use here at Entry Point AI, and the promising new multi-modal models.
LLM's are changing the game by considering a bunch of factors like what's in your cart, your past buying habits, and your customer profile. This is revolutionizing how products are suggested and setting the stage for a shopping experience that feels like it was made just for you.
For businesses, using AI for recommendations has a lot of perks. More sales, customers who stick around, and a smoother shopping process. For shoppers, it means an easy, enjoyable shopping trip that's tailored to what they want.
Looking ahead, we're excited about the potential of even more advanced models that can analyze both words and images. The future of online shopping looks bright, with AI tech leading the way. By embracing these new tools, online stores can stay on the cutting edge and keep delivering top-notch shopping experiences.
The future is here, and it's personalized.