Expansion de requête

The Query Expansion component is designed to paraphrase a given input text into multiple alternative queries. Its primary purpose is to enhance semantic search within a knowledge base by generating various rewordings of the original query, thereby increasing the likelihood of matching relevant information during retrieval processes.

What Does the Component Do?

This component takes an initial input message and produces several alternative queries that express the same intent in different ways. By leveraging a language model, it can generate these paraphrased queries, making downstream search or information retrieval tasks more robust and comprehensive. This is particularly useful in AI workflows that involve searching large corpora or knowledge bases, as it mitigates the issue of semantic mismatches between user queries and stored documents.

Inputs

The component supports the following inputs:

Quick Input Reference

• Input: The text to paraphrase (required)
• LLM: Choose which AI model to use (optional)
• Chat History: Contextual history to help model generate better queries (optional)
• Number of Alternatives: How many variations to produce (required)
• Include Original Query: Whether to keep the original in the output (optional)
• System Message: Instructions for the model (optional, advanced)

Outputs

The output is a message object that includes all generated alternative queries (and possibly the original, if specified). These alternatives can be used directly in downstream search or analysis components.

When and Why to Use

• Improve Search Recall: By generating semantically diverse queries, you increase the chances of retrieving all relevant documents, even if they use different terminology.
• Handle Ambiguity: Useful when user queries are vague or underspecified; alternative queries can cover multiple interpretations.
• Enhance Conversational AI: Integrate into chatbots or virtual assistants to interpret user intents more flexibly.
• Preprocessing for QA Systems: Before passing user input to a question-answering model, expand the query to cover more possible answers.

Example Use Case

Suppose a user asks, “How do I reset my password?” The Query Expansion component might generate alternatives like:

• “What steps should I follow to change my password?”
• “How can I recover a forgotten password?”
• “Instructions for resetting a password”
• “How to update account password?”

These expanded queries can then be used to search a knowledge base, improving the chances of finding the most relevant articles.

Summary Table

This component is a valuable building block for any AI workflow that requires robust, context-aware query expansion for improved search and retrieval capabilities.