We’re using vector search to revolutionize the way that Envato Elements customers find photos.
As part of our ongoing project of improving the search and discovery experience on Envato Elements, we’re excited to announce a whole new way for customers on Elements to find photos in our catalog.
Read on to learn more about what vector search is, and how it benefits both customers and Authors!
What is vector search and how does it differ from lexical or keyword search?
In short: vector search is a new way for search engines to classify and understand information.
In traditional lexical searches, items returned are based on exact or partial matches, synonyms, or similar words to the the search term.
This is great when people search using the same keywords as the ones the items are tagged with. However, it does have some limitations. Some of these include:
- Vulnerability to typos and misspellings
- Ambiguous words or words with multiple meanings (e.g. does someone searching for ‘Jaguar’ want images of the large cat or the car brand?)
- Difficulty parsing longer and compound search terms that require context to understand
[In semantic search] … connections are based on the meaning of the text and involve the study of the relationship between words, phrases, sentences, and their respective meanings.
Shubham Barthwal Introduction to Vector-Based Search, via Medium
In vector search (AKA semantic search), each result is represented by a single point or ‘vector’. Vectors are classified based on their similarity to other vectors.
In this case, items are returned based on their similarity to the meaning of the original search terms, not just when they match from a keyword perspective.
What are the advantages to vector search?
Vector search uses AI to recognize what is in each image and gain a human-like understanding of what a search term means.
If a customer searches for “running shoes”, vector/semantic search is able to consider items that are semantically connected, like “jogging shoes” or “athletic footwear”.
Vector search is also better at parsing longer search terms. An example of this is distinguishing between what the user means when they search “chocolate milk” vs. “milk chocolate”, even though they contain the same keywords.
Why is vector search good for Authors?
The main reason that vector search is good news for Authors is simple. Customers are more likely to download your items if they see them when they’re most relevant.
Our research indicates that more customers will download and license photos when they use vector search. More downloads = more earnings!
As we continue to make improvements to the customer experience on Elements, Authors will benefit.
Where can I try it out for myself?
Vector search is now live for all photo searches on the English version of Envato Elements. If you want to check out more of the updates we’ve made to Elements recently, have a look at our product updates.
What kinds of searches work well for vector search?
Try searching for photos using specific terms for the subjects and styles of the photos you want. Longer and more descriptive searches that include the colors, styles, mood, context and details will help vector search find the best results.
Examples of good searches include:
- “White poodle at the beach”
- “Woman in a car in noir style”
- “Creepy library”
- “Golden forest scape”
- “Orange cat lying on a couch”
- “A man lacing up his runners”
Further reading
- Introduction to Vector-Based Search and Vector Representation Techniques, by Shubham Barthwal
- What is Vector Search, by MyScale
- What is Vector Search? by Dustin Coates
