本頁說明 AlloyDB AI 向量搜尋策略，並解釋各策略的適用時機。根據預設，AlloyDB 會使用 K 近鄰搜尋 (KNN) 尋找與查詢類似的向量。向量索引會實作稱為「近似最鄰近」(ANN) 的搜尋策略。建立向量索引時，AlloyDB AI 會使用 ANN，效能優於 KNN。請注意，選取向量索引時，您需要平衡查詢延遲和召回率。
[[["容易理解","easyToUnderstand","thumb-up"],["確實解決了我的問題","solvedMyProblem","thumb-up"],["其他","otherUp","thumb-up"]],[["難以理解","hardToUnderstand","thumb-down"],["資訊或程式碼範例有誤","incorrectInformationOrSampleCode","thumb-down"],["缺少我需要的資訊/範例","missingTheInformationSamplesINeed","thumb-down"],["翻譯問題","translationIssue","thumb-down"],["其他","otherDown","thumb-down"]],["上次更新時間：2025-09-04 (世界標準時間)。"],[[["\u003cp\u003eAlloyDB AI uses k-nearest neighbors (KNN) search by default to find vectors similar to a query, but when a vector index is created, it uses Approximate Nearest Neighbor (ANN) for better performance.\u003c/p\u003e\n"],["\u003cp\u003eRecall measures how effectively a search retrieves all relevant items, with KNN achieving 100% recall by using brute force, while ANN, used with indexes, might have a lower recall rate.\u003c/p\u003e\n"],["\u003cp\u003eQuery latency measures the speed at which search results are generated.\u003c/p\u003e\n"],["\u003cp\u003eKNN is recommended for applications requiring high accuracy and when dealing with fewer than 100,000 vectors, whereas ANN is preferred for low latency and when handling over 100,000 vectors.\u003c/p\u003e\n"],["\u003cp\u003eCreating a vector index is recommended by Google to optimize performance of vector searches.\u003c/p\u003e\n"]]],[],null,["This page describes AlloyDB AI vector search strategies and explains\nwhen to use each strategy. By default, AlloyDB uses k-nearest neighbors\nsearch (KNN) to find vectors that are similar to a query. Vector indexes\nimplement a search strategy called Approximate Nearest Neighbor (ANN). When you\ncreate a vector index, AlloyDB AI uses ANN, which provides better\nperformance than KNN. Keep in mind that, when you select a vector index, you\nneed to balance query latency and recall.\n\n*Recall* measures how effectively a search retrieves all relevant items for\na given query. For example, imagine you have 100 embeddings, each one\nrepresenting an entity in your database. You query your embeddings with a\ntarget vector and limit it to 10 results. A KNN vector search finds the 10\nexact closest vectors using a brute force calculation method, which\nresults in 100% recall. AlloyDB AI uses this method by default\nif no vector search index is created or chosen.\nWhen you create a vector index in AlloyDB for PostgreSQL, it typically uses ANN,\nwhich might partition\nvectors according to similarity to facilitate faster retrieval. As a result,\nusing ANN, the 10 vectors returned in the earlier example might not be exactly\nthe 10 vectors that are closest in\ndistance. If only 8 out of the 10 retrieved vectors are the closest in space\nto your query vector, then your recall is 80%.\n\n*Query latency* defines how fast the search results are generated. For\nexample, latency is calculated based on the time spent on a search to\nreturn the vectors after you submit a query.\n\nChoose your search strategy\n\nWhen you perform vector search in AlloyDB, choose one the following\nsearch strategies:\n\n|-------------------------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|---------------------------------------------------------------------------------------------------------------------------------|\n| Search Strategy | Description | Use Cases |\n| K-nearest neighbors (KNN) | An algorithm that finds the k-nearest neighbors data points to a given query data point. When you perform a vector search without creating an index, a KNN search is performed by default. To further improve the performance of KNN search, add your embedding column, and other columns related to your query, to the column store in the [columnar engine](/alloydb/docs/columnar-engine/about). You can [add the columns manually](/alloydb/docs/columnar-engine/manage-content-manually) or [add the columns using auto-columnarization](/alloydb/docs/columnar-engine/manage-content-recommendations). | - Your application is very sensitive to accuracy and you need the exact closest matches. - You have fewer than 100,000 vectors. |\n| Approximate Nearest Neighbors (ANN) | An algorithm that finds approximately the closest data points. ANN divides existing customer data points into small groups based on similarities. | - Your application requires low latency. - You have more than 100,000 vectors. |\n\nGoogle recommends that you create a vector index to optimize performance on your\nvector search queries. For more information about how the ANN index is used for\nsimilarity searches, see [Create indexes using ScaNN](/alloydb/docs/ai/store-index-query-vectors?resource=scann).\n\nTo accelerate your filtered KNN search, use the [columnar engine](/alloydb/docs/columnar-engine/configure).\n\nWhat's next\n\n- [Create indexes and query vectors using ScaNN](/alloydb/docs/ai/store-index-query-vectors?resource=scann)\n- [Tune vector query performance](/alloydb/docs/ai/tune-indexes)"]]
