4 Ins and Outs of 2024: Vector Database Edition

chloewilliams

Chloe Williams

Posted on January 30, 2024

4 Ins and Outs of 2024: Vector Database Edition

If your recommendations are still similar in 2024, you’re doing it wrong (respectfully).

Here’s a summary of what’s in and out in the world of vector databases in 2024.

Image description

1. One type of recommendation is out, a balanced set is in

❌ Out: offering only 1 type of search like Top-K. Don’t get me wrong, top-K is core to vector search but sometimes it gives recommendations of items that are too similar, affecting recommendation quality. Just because someone listens to Adele on sad days, that doesn’t mean they want to listen to her all the time.

✅ In: The inclusion of range search ensures a more ✨balanced✨ set of results by allowing you to define a distance range for vector similarity. Balanced recommendations help prevent recommending things that are too similar or too disparate.

Image description

2. Extra steps to normalize vectors are out, using cosine similarity is in

❌ Out: Taking an extra step 👎to normalize a vector to measure similarity (identifying sentences or phrases that convey similar meanings to each other) or relatedness in various domains.

✅ In: Cosine similarity allowing you to easily normalize a vector in one step 👍

3. Updating data in multiple steps is out, using Upsert to do it seamlessly is in

❌ Out: Inefficiently updating data in a database in a frustrating 😡two-step process: delete, then insert. This can’t ensure data atomicity and operational convenience.

✅ In: Upsert simplifying the update process: if data does not exist in the system, it inserts it; if it exists, it updates it. 😀

Image description

4. The cloud provider triad (AWS, GCP, and Azure) is in

❌ Out: Vector databases not being available on the cloud provider triad: AWS, GCP, and Azure
✅ In: Vector databases that are available on 3 major cloud providers 🌟(including GCP Marketplace) and 8 regions across North America, Europe, and Asia 👀 AKA Zilliz Cloud.

A world with a vector database being available on all 3 cloud platforms is no longer imaginative. It’s true with Zilliz Cloud. <br>

If you’re interested in trying range search, upsert, and cosine similarity (on any of the 3 cloud platforms), get started here.

Image description

There’s also a live webinar covering these features and more on February 1st. Register here.

Image description

There will also be a live Q&A on Discord. Join us here.

Image description

💖 💪 🙅 🚩
chloewilliams
Chloe Williams

Posted on January 30, 2024

Join Our Newsletter. No Spam, Only the good stuff.

Sign up to receive the latest update from our blog.

Related

4 Ins and Outs of 2024: Vector Database Edition
vectordatabase 4 Ins and Outs of 2024: Vector Database Edition

January 30, 2024