In our early days of working in nutraceuticals, we often wondered how best to harness the power of AI to help make our jobs easier. We know that AI can help us design new health products, supplements, natural medicine, and functional foods; but how exactly can it help us? The typical scenarios would involve using machine learning for regression of numerical data or classification or clustering, and other boring stuff. That’s the stuff all data scientists would start with, but how can we step up our game? How can we make it more human-like in its user experience?
The brightest lightbulb appeared to us when we were chilling post-brainstorming session. We cranked up the 70-inch tv in the office and clicked on Netflix. While browsing (and bickering) for what to watch, our senior developer suddenly stands up in front of the tv and exclaimed “It’s Netflix!!!”
“Yes, we know it’s Netflix, what are you on about?” We chimed in response.
He quickly explained we should have a Netflix-like recommendation engine for our AI platform. It should be able to quickly deduce what is the current intention of a drug design project and then produce recommendations based on that deduction. For example, in a current project, if the AI engine got a few queries on ligands, protein targets and herbal plants that are related to a specific disease, the AI would then suggest other related ligands, peptides and plants. Brilliant!
But, what is a recommendation engine? In this post, we will briefly explain the important tech behind our core AI system.
The Netflix’s Way
The firm made the first pioneering customised movie suggestions in 2000. In response, it established the Netflix Prize in 2006, a competition to develop a system of tailored suggestions based on artificial intelligence, with a $ 1 million grand prize.
The buzz generated from the prize spurred dozens of teams throughout the world to continue working on and developing artificial intelligence algorithms. Many lessons learned from the Netflix Prize helped to develop the recommendation engine, making it one of the best systems of its sort on the market. Generally speaking, there are two algorithms employed by AI engineers to create a recommendation. One is called the collaborative filtering system and the other is the content-based system.
Collaborative filtering is the method of forecasting a user’s interests by detecting preferences and information from a large number of users. This is accomplished by filtering data for information or patterns utilising procedures that include the collaboration of numerous agents, data sources, and so on. The core premise of collaborative filtering is that if customers A and B have similar tastes in one product, they are likely to have similar tastes in other items as well.
Recommendations are generated by content-based systems depending on the user’s preferences and profile. They attempt to match users with things that they have previously enjoyed. The amount of resemblance between goods is often determined by the qualities of products that the user likes. Unlike most collaborative filtering models, which rely on ratings between the target user and other users, content-based models rely only on ratings supplied by the target user. In essence, the content-based method uses several data sources to provide suggestions.
The herbalogi.AI Way
In our company, we develop a hybrid system of collaborative filtering and content-based systems. The engine will learn subtle similarities between ligands, peptides, or plant extracts and group them. Thus, if our scientists querying a particular ligand/peptide/extract, other substances which have been rated to have similarity to it will be recommended. The engine also allows for user ratings where a recommendation can be given relevant ratings, and this will allow the engine to learn about the user. Some users would only prefer recommendations of very closely related information while other users are more adventurous and can accept unexpected suggestions. All features have made available the most advanced AI system that can help us to create the best natural medicine for our customers.