0Pricing
Vector Databases: Pinecone, Weaviate & pgvector · レッスン

セマンティック検索とハイブリッド検索

ベクトルの類似度とキーワード照合を組み合わせた高度な検索技術を実装し、より優れた結果を得ます。

「セマンティック検索とハイブリッド検索」はCoddyKit上の無料Vector Databases: Pinecone, Weaviate & pgvectorレッスンです。 これはレッスン1/4です。 下記で完全なレッスンを無料で読むことができます。その後、ブラウザ内の組み込みコードエディタと24時間対応のAIチューターでハンズオン演習できます。 これはVector Databases: Pinecone, Weaviate & pgvector学習パスの一部であり、ウェブとCoddyKitアプリ全体で進捗が同期されます。 Vector Databases: Pinecone, Weaviate & pgvectorコースには全4レッスンが含まれています。

このレッスンの一部はまだ翻訳されておらず、英語で表示されています。

Beyond Basic Searches

Welcome! In this lesson, we'll dive into advanced search techniques in Weaviate. Moving past simple vector searches, we'll explore how to combine different methods for incredibly precise results.

We'll cover:

  • Pure semantic search
  • Traditional keyword (BM25) search
  • The power of hybrid search

Semantic Search: Meaning First

Semantic search finds items based on their meaning, not just exact words. It uses vector embeddings to represent data, measuring "distance" to find similar concepts. Weaviate uses .with_near_text() for this.

Try this example:

import weaviate
import os

# Connect to your Weaviate instance
# Ensure WEAVIATE_URL is set (e.g., "http://localhost:8080")
client = weaviate.Client(
    url=os.getenv("WEAVIATE_URL", "http://localhost:8080")
)

# Make sure you have an 'Article' class with 'title' and 'content' properties
# and some data imported for this to work!

query_concept = "latest advancements in technology"

response = client.query.get(
    "Article", # Your class name
    ["title", "content"]
).with_near_text(
    {"concepts": [query_concept]}
).with_limit(2).do()

print("Semantic Search Results:")
for item in response["data"]["Get"]["Article"]:
    print(f"- {item['title']}")

Keyword Search Fundamentals

While semantic search is powerful, sometimes you need to find exact keywords. This is where traditional keyword search comes in. Weaviate supports this using the BM25 algorithm.

BM25 (Best Match 25) is a ranking function used by search engines to estimate the relevance of documents to a given search query. It's great for precision when you know exactly what words you're looking for.

Keyword Search with BM25

You can perform keyword searches in Weaviate by combining a .with_where() filter with a text search, and asking for the _additional {score} to see BM25 relevance.

Here's how to search for articles containing specific keywords:

import weaviate
import os

# Connect to your Weaviate instance
client = weaviate.Client(
    url=os.getenv("WEAVIATE_URL", "http://localhost:8080")
)

# Make sure you have an 'Article' class with 'title' and 'content' properties
# and some data imported for this to work!

keyword_query = "AI" # Search for articles containing "AI"

response = client.query.get(
    "Article",
    ["title", "content", "_additional {score}"] # Request BM25 score
).with_where({
    "path": ["content"], # Search in the 'content' field
    "operator": "Like",
    "valueText": f"*{keyword_query}*" # Wildcard search
}).with_limit(2).do()

print("Keyword Search Results:")
for item in response["data"]["Get"]["Article"]:
    print(f"- {item['title']} (BM25 Score: {item['_additional']['score']:.2f})")

Why Hybrid? Limitations

Both semantic and keyword searches have strengths and weaknesses:

  • Semantic: Great for conceptual understanding, but can miss exact terms.
  • Keyword: Excellent for exact matches, but struggles with synonyms or nuanced meaning.

Imagine searching for "best car for family trips." Semantic search might show SUVs, while keyword search might only show articles with "family" and "trip." What if you want both?

Introducing Hybrid Search

Hybrid search combines the strengths of semantic (vector) search and keyword (BM25) search. It retrieves results based on both conceptual similarity and exact term matching, then intelligently fuses them.

This leads to more comprehensive and relevant results, especially for complex or ambiguous queries.

Weaviate's `with_hybrid`

Hybrid search combines semantic and keyword strengths. Weaviate's .with_hybrid() operator makes this easy. It takes both a query and an alpha parameter to control the blend:

  • alpha = 0: Pure keyword
  • alpha = 1: Pure semantic
  • alpha = 0.5: Equal blend (default)

Experiment with this:

import weaviate
import os

# Connect to your Weaviate instance
client = weaviate.Client(
    url=os.getenv("WEAVIATE_URL", "http://localhost:8080")
)

# Make sure you have an 'Article' class with 'title' and 'content' properties
# and some data imported for this to work!

query = "AI tools for data analysis" # Hybrid query text
alpha_value = 0.7 # 0.7 for more semantic weighting

response = client.query.get(
    "Article",
    ["title", "content", "_additional {score, id}"] # Request score & ID
).with_hybrid(
    query=query,
    alpha=alpha_value
).with_limit(3).do()

print(f"Hybrid Search Results (alpha={alpha_value}):")
for item in response["data"]["Get"]["Article"]:
    # The 'score' here is the hybrid score
    print(f"- {item['title']} (Score: {item['_additional']['score']:.2f})")

Understanding Result Fusion (RRF)

When you perform a hybrid search, Weaviate needs a way to combine the rankings from both the semantic and keyword searches into a single, unified list. This is often done using an algorithm like Reciprocal Rank Fusion (RRF).

RRF is a clever method that assigns a score to each document based on its rank in the individual search results. Documents that rank highly in both semantic and keyword searches will get a significantly boosted final score.

Benefits of Hybrid Search

Hybrid search offers several advantages:

  • Improved Relevance: Catches both exact matches and conceptually similar items.
  • Robustness: Performs well even with short, ambiguous, or rare queries.
  • User Satisfaction: Leads to more comprehensive and helpful search results.

It's a crucial technique for building advanced search experiences in AI applications.

Test Your Knowledge

Hybrid search combines semantic and keyword search. Which parameter in Weaviate's .with_hybrid() operator controls the balance between these two search types?

Summary of Advanced Search

Great job! You've mastered advanced search techniques in Weaviate. We explored:

  • Semantic Search: Based on meaning and vector similarity.
  • Keyword Search: Using BM25 for exact term matching.
  • Hybrid Search: Combining both for superior relevance, controlled by the alpha parameter.

These powerful tools will help you build more intelligent and robust search applications. Keep experimenting with different query types and alpha values!

よくある質問

「セマンティック検索とハイブリッド検索」レッスンは無料ですか？

はい。「セマンティック検索とハイブリッド検索」の完全なテキストはこのウェブで無料で読めます。インタラクティブに演習し（組み込みコードエディタと24時間対応のAIチューター）、Vector Databases: Pinecone, Weaviate & pgvectorコースの残りをアンロックするには、CoddyKit PROにアップグレードしてください。 Vector Databases: Pinecone, Weaviate & pgvectorコースには全4レッスンが含まれています。

「セマンティック検索とハイブリッド検索」で何を学びますか？

ベクトルの類似度とキーワード照合を組み合わせた高度な検索技術を実装し、より優れた結果を得ます。 ブラウザで直接実行するハンズオンコードでVector Databases: Pinecone, Weaviate & pgvectorを演習し、24時間対応のAIチューターがレッスンを進める中での質問に答えます。

Vector Databases: Pinecone, Weaviate & pgvectorを始めるのに経験は必要ですか？

事前経験は必要ありません。CoddyKitのVector Databases: Pinecone, Weaviate & pgvectorは初級者から上級者向けに構成されているため、ここから始めるか最初から始めて、自分のペースで進むことができます。 これはレッスン1/4です。

「セマンティック検索とハイブリッド検索」レッスンにはどのくらい時間がかかりますか？

ほとんどのCoddyKitレッスンは約5～10分かかります。各レッスンはコンパクトでインタラクティブなので、着実に進歩し、ウェブとアプリ全体で正確に前回の場所から再開できます。

このVector Databases: Pinecone, Weaviate & pgvectorレッスンでコードを書いて実行できますか？

はい。すべてのVector Databases: Pinecone, Weaviate & pgvectorレッスンに組み込みコードエディタが含まれているため、ブラウザでリアルコードを書いて実行し、即座のAIフィードバックを取得できます。ローカル設定は不要です。

このコースのすべてのレッスン

  1. セマンティック検索とハイブリッド検索
  2. Weaviateモジュールの利用
  3. バックアップと復元の戦略
  4. Weaviateのマルチテナンシー
← Vector Databases: Pinecone, Weaviate & pgvectorに戻る