Seknd uses machine learning to make personalized skincare recommendations for you. You might ask: How does Amazon know what you want to buy next? How do self-driving cars actually work? What are the mechanisms behind fraud detection? Machine learning and its applications thread our everyday lives, and while the thought might conjure up the image of long, complicated lines of code, it is not as scary or inaccessible as one might think.
Sunny Kim, our CTO and mastermind behind the machine learning platform, says, “We want to help users reach the product that is best for them –– one that is effective and one they will like.” She recognizes that while humans can also effectively recommend products to others, one person can only do so within a limited scope, and hopes to use machine learning to scale up the possibilities of a perfect product “fit.” The company strives to use technology to provide an introduction to and advice for the world of skincare, focusing on a personalized approach that can reach any number of people, and ultimately allow the customer to make the final decision about a product.
While companies like Netflix and Spotify all approach their product recommendation platforms through different angles, they collect and analyze data in similar ways. Sunny says that data generally comes from two different camps: meta data, which is centered around the content itself (what category is it in? Is it a face wash or moisturizer?), and user data, which is further broken down into explicit input from the user (for example, that they have dry skin) and behavioral input (trends we can infer from their activity on the website).
At a high level, Seknd uses collaborative filtering to align this historic data and then predict future trends –– if enough people who like product A like product B, then we can infer that the next person who likes product A will also like product B. Seknd’s platform represents items with high dimensional vectors, called embeddings, which capture complex relationships between different entities. Each of the numbers in the vector captures a different characteristic of the entity it is representing. If a product and a user are close in a vector space, then it’s more likely that the user will rate the product highly.
What is particularly interesting about machine learning and skincare, Sunny says, is the amount of data there is surrounding any given product in itself. Some of the most interesting data points generally come from the ingredients (mostly chemicals); on average, a product consists of 20-30 distinct ingredients, whose specific combination attempts to address a particular goal. Since it’s difficult for any one user to fully comprehend a product’s ingredients and what they do, Seknd hopes to use models to provide the domain knowledge required for this understanding, as well as information about the most effective combinations to address a particular goal or concern. Not only does Seknd determine the effectiveness of a product by understanding its ingredients and their functions, it also sometimes relies on reviews to understand what a customer liked about a product and what it was effective for, though reviews are never 100% accurate due to their subjectivity.
Seknd’s machine learning platform is constantly evolving –– different models add to each other and to the enhance accuracy and precision of its recommendations. Each new data point on user behavior collected from the website makes the model smarter and more competitive, which is why it is difficult to ensure that a program is accurate and effective in its early stages. To tackle this potential drawback, Seknd is focusing on the meta data of the product itself, as well as explicit input from users about their skincare preferences, using these data points to rapidly build upon and enhance its existing platform.
At the end of the day, Sunny recognizes that finding the one perfect product out of millions only happens in an ideal world, and that no platform, even powered by the most advanced technology, is able to make such a recommendation. This is why Seknd focuses on the intuition and tastes of their users –– it should, ultimately, be a trusted seknd opinion that helps YOU select whatever works best for you.