📘 Overview

KBCMultiHopQAGeneratorBatch is a batch-based multi-hop question-answer pair generation operator designed to automatically generate questions and answers that require multi-step reasoning from given textual data. By invoking a Large Language Model (LLM), this operator transforms raw text into structured QA data, suitable for constructing complex QA datasets or enhancing knowledge bases.

---

init Function

def __init__(self,
  llm_serving: LLMServingABC,
  seed: int = 0,
  lang="en",
  prompt_template = None
):

Initialization Parameters

Parameter	Type	Default	Description
llm_serving	LLMServingABC	Required	The LLM service instance used for inference and generation.
seed	int	0	Random seed to ensure reproducibility of the generation process.
lang	str	"en"	Language setting that specifies the output language for QA pairs (e.g., "en" or "zh").
prompt_template	PromptABC	Text2MultiHopQAGeneratorPrompt	Prompt template object used to construct the input for multi-hop QA generation.

---

Prompt Template Description

Template Name	Purpose	Applicable Scenario	Key Features
Text2MultiHopQAGeneratorPrompt	Generate multi-hop QA pairs from text	Scenarios that require constructing complex reasoning questions from long passages	Built-in template that guides the model to generate the question, reasoning steps, final answer, and supporting facts, ensuring structured and logical output.

---

run Function

def run(
    self,
    input_key: str = 'chunk_path',
    output_key: str = 'enhanced_chunk_path',
    storage: DataFlowStorage = None,
):

Parameters

Name	Type	Default	Description
storage	DataFlowStorage	Required	DataFlow storage instance responsible for reading and writing data.
input_key	str	"chunk_path"	The input column name that contains the path to JSON or JSONL files with text chunks to process.
output_key	str	"enhanced_chunk_path"	The output column name that will store the path to the enhanced files containing generated QA pairs.

---

🧠 Example Usage

self.knowledge_cleaning_step4 = KBCMultiHopQAGeneratorBatch(
    llm_serving=self.llm_serving,
    lang="en"
)
self.knowledge_cleaning_step4.run(
    storage=self.storage.step(),
)

---

🧾 Default Output Format

The operator reads the text files specified by input_key, generates QA pairs, and writes the enriched content back to the same file.

Field	Type	Description
text	str	The original context text.
qa_pairs	list	A list of generated multi-hop QA pairs, each containing the question, answer, reasoning steps, and supporting facts.

Example Input (chunk_path file content)

{
  "cleaned_chunk": "The Eiffel Tower is located in Paris, the capital of France. The Louvre Museum, also in Paris, is the world's largest art museum."
}

Example Output (after operator execution)

{
  "cleaned_chunk": "The Eiffel Tower is located in Paris, the capital of France. The Louvre Museum, also in Paris, is the world's largest art museum.",
  "qa_pairs": [
    {
      "question": "In which country is the world's largest art museum located?",
      "reasoning_steps": [
        {"step": "The text states the Louvre Museum is the world's largest art museum."},
        {"step": "The text also states the Louvre Museum is in Paris."},
        {"step": "Paris is identified as the capital of France."}
      ],
      "answer": "France",
      "supporting_facts": [
        "The Louvre Museum, also in Paris, is the world's largest art museum.",
        "Paris, the capital of France."
      ],
      "type": "Geography"
    }
  ]
}