Common Query Pipeline

The Common Query Pipeline Plugin in OVOS is a specialized pipeline component designed exclusively for handling general knowledge questions. It processes utterances that resemble questions—typically starting with interrogatives like what, who, how, when, etc.—and queries a set of registered general knowledge skills to find the most accurate factual answer.

Unlike conversational or chit-chat pipelines, this plugin focuses strictly on fact-based question answering. It does not generate answers or perform retrieval-augmented generation (RAG). Instead, it relies on a reranker module to evaluate candidate answers from all queried skills and selects the most relevant and factually accurate response.

Purpose

  • Handle only question-like utterances (e.g., “What is the tallest mountain?”, “Who wrote Hamlet?”).
  • Query multiple general knowledge skills to obtain candidate answers.
  • Use a reranker mechanism to evaluate and select the most confident and factually accurate response.
  • Provide a robust fallback for answering factual queries outside of high-confidence intent matches.

Pipeline Stages

This plugin registers a single pipeline:

Pipeline ID Description Recommended Use
common_qa Common Query matches Only as good as the common query skills you install

How It Works

  1. Question Detection: The pipeline filters incoming utterances to only process those that appear to be questions, based on interrogative keywords.
  2. Parallel Skill Querying: The plugin sends the utterance to all registered common query skills capable of answering general knowledge questions.
  3. Candidate Collection: Each skill returns zero or more candidate answers along with confidence scores.
  4. Reranking: A reranker component evaluates all candidate answers across skills to identify the best response, focusing on factual accuracy and confidence.
  5. Answer Delivery: If a suitable answer is found, it is returned to the user; otherwise, the query is passed on or marked as unanswered.

Installation

The Common Query Pipeline Plugin is included by default in ovos-core, but can also be installed independently:

pip install ovos-common-query-pipeline-plugin

Configuration

"intents": {
  "common_query": {
    "min_self_confidence": 0.5,
    "min_reranker_score": 0.5,
    "reranker": "ovos-flashrank-reranker-plugin",
    "ovos-flashrank-reranker-plugin": {
      "model": "ms-marco-TinyBERT-L-2-v2"
    }
  }
}
  • min_self_confidence: Minimum confidence required from the skill answer itself before reranking.
  • min_reranker_score: Minimum reranker score threshold to accept an answer.
  • reranker: The reranker plugin to use (must be installed separately).
  • Model: Choose a suitable reranker model based on accuracy and device constraints.

Performance Considerations

  • The plugin’s response time depends on the slowest queried skill — the latency of installed common query skills affects overall speed.
  • Enabling rerankers, especially on resource-limited hardware (e.g., Raspberry Pi), may add noticeable latency.
  • Timeout (default 2 seconds) ensures responsiveness but might cause some slow skill answers to be discarded.
  • Tune confidence thresholds and reranker settings according to your hardware capabilities and user experience goals.

Example Usage Scenario

User says: "When was the Declaration of Independence signed?"

  • The utterance is detected as a question.
  • The plugin queries ovos-skill-wolfram-alpha and ovos-skill-wikipedia.
  • Each skill returns candidate answers with confidence scores.
  • The reranker evaluates answers and selects the most reliable response.
  • The selected answer is delivered back to the user.

Important Notes

  • No generation or RAG: The plugin only retrieves answers from skills; it does not generate or synthesize new content.
  • No chit-chat: This pipeline is strictly for general knowledge queries, not for casual conversation or small talk.
  • Reranker-based selection: The reranker improves the quality of responses by ranking answers from multiple sources.
  • Skills required: Ensure that relevant common query skills (e.g., knowledge bases, encyclopedias) are installed and enabled.