Embedding models are machine learning models that represent data (such as text, images, or other information) in a continuous, low-dimensional vector space. In other words, they transform objects into vectors.
Each object type has its own specialised embedding model.
Examples of embedding models are:
- For images: CLIP, DINOv2, ResNet, etc.
- For text: GloVe, Text Embedding 3 Small and Large, Qwen3 Embedding 8B, Mistral Embed 2312, Gemini Embedding 001, etc.
- For audio: Wav2Vec, VGGish, OpenL3, Speech2Vec, VQ‑VAE, YAMNet, etc.
Objects are fundamentally data, which passes through multiple layers within an embedding model. Each layer extracts progressively more abstract features.
For example, in images, early layers detect basic features, let’s say edges. As the images pass through deeper layers, the embedding model recognises more complex features, such as objects.
In text, the first layers may detect individual words. Deeper layers will detect context and meaning.
Text-Based Embedding Models
Embedding models transform raw text, such as a sentence or a paragraph, into a fixed-length vector of numbers that captures its semantic meaning. These vectors allow machines to compare and search text based on meaning rather than exact words. In practice, this means that texts with similar ideas are placed close together in the vector space. For example, instead of matching only the phrase “machine learning”, embeddings can find documents that discuss related concepts even when they use different terms.
The steps that an embedding model performs are:
- Vectorisation: The model encodes each input string as a high-dimensional vector.
- Similarity scoring: The model compares vectors using mathematical metrics to measure how closely related the underlying texts are.
The most common similarity metrics these models use for comparing embeddings are:
- Cosine similarity measures the angle between two vectors.
- Euclidean distance gauges the straight-line distance between points.
- Dot product assesses how much one vector projects onto another.