Introduction to Vector Search
Get an understanding of vector search concepts, use cases and the components in its workflow.
What is vector search?
Vector search is also known as semantic search or similarity search. It uses vector representations to retrieve a dataset's most relevant (or similar) items. For example, given the phrase “delicious Italian cuisine,” a vector search might return closely related phrases like “tasty pasta dishes” or “authentic Mediterranean food” based on the semantic similarity, even if they don’t share exact keywords.
In traditional keyword-based search, textual data is typically indexed using inverted indexes, which map words to the documents or passages containing them. This approach does not consider the semantic relationships between words and documents. Vector search, on the other hand, operates in a high-dimensional vector space, where each item (document, image, or any other data point) is represented as a dense vector embedding. These vector embeddings capture the semantic relationships between items, allowing for more accurate and relevant search results.
Aspect | Vector Search | Traditional Search |
Performance | Uses vector representations and similarity calculations | Uses keyword matching and inverted indexes |
Semantic Understanding | Can understand context and meaning | Limited to exact keyword matches |
Data Representation | Dense vector embeddings | Sparse, term-frequency based representations |
Query Flexibility | Can handle natural language queries effectively | Works best with specific keywords |
Computational Complexity | Generally more computationally intensive | Usually faster for exact matches |
Data Types | Can handle various data types (text, images, audio) | Primarily designed for text |
Applications of vector search
Vector search has numerous applications across various domains.
Question answering
Question answering systems use vector search to retrieve the most relevant passages or contexts needed to answer a given query. The application can ...