"Any fool can know. The secret is to understand." Albert Einstein
Can't I just do this myself?
Deep learning's dirty secret is that it requires an awful lot of clean data. Typically the engineers with whom we work miss the labelled items and the framework to put them in to train a comprehensive model. We’ve taken millions of retail products, sorted and labelled them by passionate people, to form our Universal Product Ontology
What's so hard about this?
Fine grained classification for ranking is a tough problem. Our deep learning models understand both the words and pictures of retail products to match search engine query data to real product items. Similar.ai makes sense of your data.
Do I need to clean my data?
Your catalogue of products looks perfect to users, so it will look perfect to us too. We all prefer some information expressed with pictures and other information expressed with text. Similar.ai makes sense of your data as it is today. No further enrichment or cleaning needed. If your users love your products, Similar.ai will love them too.
How do I get started?
Send us a product catalogue and we'll add the appropriate labels from our ontology. The more data we have from each product page, the better we do. Just like a person, Similar.ai likes it when a product page has multiple images, text and structured features to help it understands what the product is. You don't need to send us the feed in any particular format, since we natively understand unstructured data about products, so send whatever is simplest. Many of our customers start with a Google Shopping feed. If you don't have a feed to hand, ask us about reading your data from your website.
Isn't everyone doing deep learning these days?
To really understand what you sell, we needed to pair our expertise in image understanding, natural language understanding and ontology with an enormous clean labeled set of data. Our training set contains millions of clothes, home decorations, jewellry, kitchenware, bags, furniture, shoes and other things people wear or buy for their homes, all carefully labelled and enriched by people passionate about the kinds of products you sell, just like you.
Many companies are doing image recognition, trying to work out a rough category for multiple products in a single image. Others try to find the most similar product in an online catalogue to a picture online. Similar.ai is solving a different problem. We find the most common ways people would describe everyday products, such as the things we wear and the things we buy for our homes, to build out our Universal Product Ontology. Then we build machine learning classifiers which leverage a huge set of labelled data to learn how to understand all the images and text on any product page. Coupling both a rich, useful ontology with deep learning classifiers which work at human accuracy and web scale for any retailer, we can enable vastly improved SEO, consistently accurate navigation, semantic search which works for fat tail and long tail queries, and business intelligence which speaks the language of the business.
How do you scale?
Similar.ai has processed millions of items across hundreds of product catalogues.