DrivelologyMultilabelClassification#

Overview#

Drivelology Multi-label Classification evaluates models’ ability to categorize “drivelology” text into rhetorical technique categories: inversion, wordplay, switchbait, paradox, and misdirection. Each text may belong to multiple categories.

Task Description#

Task Type: Multi-label Text Classification
Input: Drivelology text sample
Output: One or more technique categories
Domain: Linguistic analysis, rhetorical technique detection

Key Features#

Five rhetorical technique categories
Multi-label classification (multiple categories per sample)
Tests understanding of linguistic creativity mechanisms
Requires recognition of humor and irony techniques
Detailed category definitions provided

Evaluation Notes#

Default configuration uses 0-shot evaluation
Metrics: F1 (weighted, micro, macro), Exact Match
Aggregation method: F1 weighted
Categories: inversion, wordplay, switchbait, paradox, misdirection

Properties#

Property	Value
Benchmark Name	`drivel_multilabel`
Dataset ID	extraordinarylab/drivel-hub
Paper	N/A
Tags	`MCQ`
Metrics	`f1_weighted`, `f1_micro`, `f1_macro`, `exact_match`
Default Shots	0-shot
Evaluation Split	`test`
Aggregation	`f1_weighted`

Data Statistics#

Metric	Value
Total Samples	600
Prompt Length (Mean)	1637.18 chars
Prompt Length (Min/Max)	1580 / 2041 chars

Sample Example#

Subset: multi-label-classification

{
  "input": [
    {
      "id": "0e8acb03",
      "content": [
        {
          "text": "#Instruction#:\nClassify the given text into one or more of the following categories: inversion, wordplay, switchbait, paradox, and misdirection.\n\n#Definitions#:\n- inversion: This technique takes a well-known phrase, cliché, or social script a ... [TRUNCATED] ... e should be of the following format: 'ANSWER: [LETTERS]' (without quotes) where [LETTER]S is one or more of A,B,C,D,E.\n\nText to classify: 後天的努力比什麼都重要，所以今天和明天休息。\n\nA) A. inversion\nB) B. wordplay\nC) C. switchbait\nD) D. paradox\nE) E. misdirection"
        }
      ]
    }
  ],
  "choices": [
    "A. inversion",
    "B. wordplay",
    "C. switchbait",
    "D. paradox",
    "E. misdirection"
  ],
  "target": "AB",
  "id": 0,
  "group_id": 0,
  "metadata": {
    "text": "後天的努力比什麼都重要，所以今天和明天休息。",
    "label": [
      "inversion",
      "wordplay"
    ],
    "target_letters": "AB"
  }
}

Note: Some content was truncated for display.

Prompt Template#

Prompt Template:

{question}

Usage#

Using CLI#

evalscope eval \
    --model YOUR_MODEL \
    --api-url OPENAI_API_COMPAT_URL \
    --api-key EMPTY_TOKEN \
    --datasets drivel_multilabel \
    --limit 10  # Remove this line for formal evaluation

Using Python#

from evalscope import run_task
from evalscope.config import TaskConfig

task_cfg = TaskConfig(
    model='YOUR_MODEL',
    api_url='OPENAI_API_COMPAT_URL',
    api_key='EMPTY_TOKEN',
    datasets=['drivel_multilabel'],
    limit=10,  # Remove this line for formal evaluation
)

run_task(task_cfg=task_cfg)