Implementation:Datajuicer Data juicer TextActionFilter

Knowledge Sources	Datajuicer_Data_juicer
Domains	Data_Quality, Filtering
Last Updated	2026-02-14 16:00 GMT

Overview

Concrete tool for filtering data samples based on action verb count provided by Data-Juicer.

Description

TextActionFilter is a filter operator that keeps samples containing at least a minimum number of action verbs. It extends Filter and uses the two-phase compute_stats/process pattern. It uses a spaCy NLP model (supporting English and Chinese) to parse text and identify action verbs by checking both POS tags (VERB) and specific fine-grained tags (VV, VB, VBP, VBZ, VBD, VBG, VBN). The count is cached under num_action in the sample stats. Useful for filtering datasets for narrative or instructional content where the presence of actions indicates dynamic, descriptive text.

Usage

Import when filtering based on action verb count. Configure in YAML or Python.

Code Reference

Source Location

• Repository: Datajuicer_Data_juicer
• File: data_juicer/ops/filter/text_action_filter.py

Signature

@OPERATORS.register_module("text_action_filter")
class TextActionFilter(Filter):
    def __init__(self, lang: str = "en", min_action_num: int = 1, *args, **kwargs):

Import

from data_juicer.ops.filter.text_action_filter import TextActionFilter

I/O Contract

Inputs

Name	Type	Required	Description
lang	str	No	Language for action detection: "en" or "zh" (default: "en")
min_action_num	int	No	Minimum number of actions to keep sample (default: 1)

Outputs

Name	Type	Description
samples	Dict	Filtered samples with num_action stat computed

Usage Examples

YAML Configuration

process:
  - text_action_filter:
      lang: en
      min_action_num: 1

Python API

from data_juicer.ops.filter.text_action_filter import TextActionFilter
op = TextActionFilter(lang="en", min_action_num=1)

Related Pages

• Environment:Datajuicer_Data_juicer_Python_Runtime_Environment