Skip to main content

🔍 Dimension Members Retriever

In the XpertAI platform, Dimension Member Retrieval is a critical capability module within the ChatBI agent orchestration system. By dynamically synchronizing and vectorizing dimension member information in the semantic model, it provides agents with enhanced contextual awareness and intent recognition, significantly improving the accuracy and user experience of conversational data analysis.

In simple terms, dimension member retrieval enables agents to “know the possible dimension values” and “understand the semantic meanings of these dimension members in different business contexts.”

1. Synchronization and Vectorization of Dimension Member Information​

To enable dimension member retrieval, the synchronization and vectorization of dimension data in the semantic model must first be completed. The process is as follows:

1. Dimension Member Synchronization (Sync)​

After creating or updating the semantic model, manually trigger the dimension member synchronization task, which includes:

  • Refreshing the dimension member list.
  • Uploading and embedding dimension members for subsequent processing and retrieval.

2. Vectorization Processing (Embedding)​

After synchronization, the platform performs semantic vectorization of dimension members, with the following steps:

  • Utilizing built-in or user-defined vector models (e.g., OpenAI Embedding, Tongyi Embedding, etc.).
  • Converting each dimension member (along with its aliases and business tags) into semantic vectors.
  • Storing the vectorized results in the platform’s embedded vector database (support for FAISS, Qdrant, Milvus, etc., in development).

2. How to Use in the ChatBI Toolset​

Once vectorization is complete, dimension member information is integrated into the ChatBI agent’s toolchain via the Dimension Member Retriever tool, with the following functions:

Agent Invocation Process:​

  1. Intent Recognition Phase: The agent determines whether the user’s query contains ambiguous dimension descriptions (e.g., “East China,” “XX customer,” “holidays”).
  2. Retrieval Invocation Phase: The Dimension Member Retriever performs semantic vector matching to identify the closest matching items from the dimension member database.
  3. Result Lookup Phase: The matched dimension values are used as filtering conditions for query construction and chart generation.

Application Examples:​

  • User asks: “How are sales in East China this month?”
    • The agent matches “East China” → region = 'East China' via the Retriever.
  • User asks: “Which product categories sold well during holidays?”
    • Matches “holidays” dimension members → Automatically applies the corresponding holiday time period for analysis.

🧠 ChatBI performs fuzzy recognition and contextual inference based on dimension business tags, pinyin, English aliases, etc.

3. Business Value and Advantages​

FeatureBusiness Advantage
Dynamic Dimension Member SyncEnsures real-time updates, reducing missed data.
Vectorized RecognitionImproves natural language recognition accuracy and supports fuzzy matching.
Multilingual UnderstandingSupports international scenarios, recognizing dimension names in multiple languages.
Business Semantic AwarenessIntelligently understands business aliases, abbreviations, and classification tags.
No Rule MaintenanceEliminates the need for manual mapping tables compared to traditional keyword matching.

4. Usage Notes​

  1. Data Source Access Permissions: Ensure the data tables providing dimension members have full read permissions.
  2. Data Volume and Performance: For datasets with millions of dimension members, configure vector extraction frequency and index partitioning as needed.
  3. Language Consistency: In scenarios with mixed Chinese and English, set multilingual tags or alias fields in the semantic model.
  4. Regular Update Mechanism: For business dimensions (e.g., customers, products), set a regular sync schedule (e.g., daily or hourly).
  5. Privacy Compliance: If dimensions include sensitive information like customers or users, ensure proper desensitization and access control.

5. Future Plans​

In future versions, we plan to:

  • Support user-defined semantic tags for dimension members.
  • Introduce weights and popularity metrics for dimension members to enhance matching accuracy.
  • Provide a dimension member management backend for manual editing and review.