Implementation:Datajuicer Data juicer QueryIntentDetectionMapper

Knowledge Sources	Datajuicer_Data_juicer
Domains	Data_Processing, Mapping
Last Updated	2026-02-14 16:00 GMT

Overview

Concrete tool for detecting user intent in queries provided by Data-Juicer.

Description

QueryIntentDetectionMapper is a mapper operator that classifies the user intent of queries by predicting an intent label and confidence score using a Hugging Face text-classification model. It optionally translates Chinese queries to English before classification using a translation model. The predicted intent label and score are stored in the sample metadata fields. Samples that already contain intent annotations are skipped. Requires CUDA acceleration and operates in batched mode.

Usage

Use when you need to automatically tag query intent in training data to enable downstream filtering or analysis by intent category.

Code Reference

Source Location

• Repository: Datajuicer_Data_juicer
• File: data_juicer/ops/mapper/query_intent_detection_mapper.py

Signature

@OPERATORS.register_module("query_intent_detection_mapper")
class QueryIntentDetectionMapper(Mapper):
    def __init__(
        self,
        hf_model: str = "bespin-global/klue-roberta-small-3i4k-intent-classification",
        zh_to_en_hf_model: Optional[str] = "Helsinki-NLP/opus-mt-zh-en",
        model_params: Dict = {},
        zh_to_en_model_params: Dict = {},
        *,
        label_key: str = MetaKeys.query_intent_label,
        score_key: str = MetaKeys.query_intent_score,
        **kwargs,
    ):

Import

from data_juicer.ops.mapper.query_intent_detection_mapper import QueryIntentDetectionMapper

I/O Contract

Inputs

Name	Type	Required	Description
hf_model	str	No	HuggingFace model ID for intent classification (default: bespin-global/klue-roberta-small-3i4k-intent-classification)
zh_to_en_hf_model	Optional[str]	No	Translation model from Chinese to English (default: Helsinki-NLP/opus-mt-zh-en)
model_params	Dict	No	Model parameters for the classification model
zh_to_en_model_params	Dict	No	Model parameters for the translation model
label_key	str	No	Key name in meta field for the output label (default: query_intent_label)
score_key	str	No	Key name in meta field for the output score (default: query_intent_score)

Outputs

Name	Type	Description
meta[label_key]	str	Predicted intent label
meta[score_key]	float	Confidence score for the predicted intent

Usage Examples

process:
  - query_intent_detection_mapper:
      hf_model: 'bespin-global/klue-roberta-small-3i4k-intent-classification'
      zh_to_en_hf_model: 'Helsinki-NLP/opus-mt-zh-en'

Related Pages

• Environment:Datajuicer_Data_juicer_Python_Runtime_Environment