Dense Retrieval Method
Jump to navigation
Jump to search
A Dense Retrieval Method is an information retrieval method that uses dense vector representations to match queries with documents through semantic similarity computation.
- AKA: Dense Passage Retrieval Method, Neural Retrieval Method, Embedding-Based Retrieval Method.
- Context:
- It can typically encode Text Sequences with neural encoder models.
- It can typically compute Cosine Similarity between query vectors and document vectors.
- It can typically perform Approximate Nearest Neighbor Search for efficient retrieval.
- It can often utilize Bi-Encoder Architectures for independent encoding.
- It can often employ Contrastive Training for representation learning.
- It can often apply Hard Negative Mining for discrimination improvement.
- It can often integrate Vector Databases for scalable indexing.
- It can range from being a Single-Vector Dense Retrieval Method to being a Multi-Vector Dense Retrieval Method, depending on its representation granularity.
- It can range from being a Symmetric Dense Retrieval Method to being a Asymmetric Dense Retrieval Method, depending on its encoder configuration.
- It can range from being a In-Domain Dense Retrieval Method to being a Zero-Shot Dense Retrieval Method, depending on its training coverage.
- It can range from being a Monolingual Dense Retrieval Method to being a Cross-Lingual Dense Retrieval Method, depending on its language capability.
- ...
- Examples:
- Foundation Dense Retrieval Methods, such as:
- Domain-Specific Dense Retrieval Methods, such as:
- Architecture-Based Dense Retrieval Methods, such as:
- ...
- Counter-Examples:
- Sparse Retrieval Method, which uses term-based rather than dense representations.
- Lexical Retrieval Method, which matches keywords rather than semantics.
- Boolean Retrieval Method, which uses exact matching rather than similarity.
- See: Bi-Encoder Model, Information Retrieval Method, Vector Database, Contrastive Learning Technique for Legal Text, Neural Information Retrieval, Semantic Search, Embedding Model.