Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Semantic Search
Search method that understands the intent and contextual meaning of user queries rather than relying solely on exact keyword matching. It uses artificial intelligence techniques to interpret concepts and relationships between terms.
Transformer Models
Deep learning architecture based on attention mechanisms that captures long-range contextual dependencies in texts. These models form the foundation of modern semantic search and natural language understanding systems.
Semantic Attention
Mechanism allowing models to weight different parts of a text differently based on their relevance to the global context. Semantic attention helps identify the most important concepts in a query or document.
Hybrid Search
Approach combining traditional keyword search (sparse retrieval) with semantic search (dense retrieval) to optimize precision and recall. This method leverages the strengths of each technique to provide more relevant results.
Vector Indexing
Process of organizing and storing embeddings in optimized data structures for fast similarity searches. Vector indexing is crucial for maintaining high performance in large-scale semantic search systems.
Dense Retrieval
Search method using dense embeddings to find documents semantically similar to a query, as opposed to sparse retrieval based on term occurrences. It excels at understanding context and abstract concepts.
Sparse Retrieval
Traditional search technique based on the presence and frequency of exact keywords in documents, represented by sparse vectors. It remains effective for specific queries with precise terms.
Semantic Distance
Quantitative measure of the semantic gap between two concepts or texts in a vector space, often calculated using Euclidean distance or cosine similarity. It allows quantifying conceptual proximity independently of the words used.
Vector Queries
Search queries transformed into numerical vectors to enable semantic comparisons with indexed documents. This approach allows finding relevant results even without exact keyword matches.
Semantic Vector Space
Multidimensional representation where concepts and words are positioned according to their mutual semantic relationships. In this space, geometric proximity between vectors corresponds to meaning similarity.
Semantic Recontextualization
Process of adapting the meaning of a term or phrase based on the overall context of the document or conversation. This technique is essential for understanding nuances and ambiguities in natural language.