yothinS/Advertisement_specialist

Performance Marketing Intent & Diagnosis Dataset

A dataset for intent classification and problem diagnosis in digital marketing and performance advertising.

---

Overview

Item	Details
Size	877 records
Categories	26 classes
Languages	Thai (th), English (en)
Primary task	Text classification (multi-class)
Scope	Campaign diagnostics, funnel strategy, metrics analysis, and A/B testing

Links and contact

• Repository: -
• Point of Contact: -

---

Dataset Structure

Data Instance

{
  "id": "sample_001",
  "text": "[Example text, e.g. 'CTR dropped sharply after running the campaign for 3 weeks. What is the cause?']",
  "label": "ad_fatigue"
}

Data Fields

Field	Type	Description
id	string	Unique identifier for each record
text	string	A question, problem symptom, or campaign context
label	string / ClassLabel	Analysis or strategy category (26 classes)

---

Class Distribution

The dataset contains 877 records distributed across classes as follows:

Rank	Label	Records	Share (%)
1	creative & messaging	159	18.13
2	roas_analysis	67	7.64
3	cvr_analysis	51	5.82
4	ctr_analysis	50	5.70
5	audience_strategy	41	4.68
6	campaign_diagnosis	40	4.56
7	cpc_analysis	40	4.56
8	funnel_strategy	39	4.45
9	funnel_diagnosis	36	4.10
10	experimentation	30	3.42
11	competitor & market	27	3.08
12	conversion_strategy	27	3.08
13	ad_fatigue	24	2.74
14	customer_acquisition	22	2.51
15	creative_diagnosis	21	2.39
16	audience_diagnosis	21	2.39
17	cpa_analysis	21	2.39
18	campaign optimization	20	2.28
19	retargeting_strategy	20	2.28
20	consideration_strategy	20	2.28
21	awareness_strategy	20	2.28
22	ab_testing	20	2.28
23	attribution_analysis	16	1.82
24	cpm_analysis	15	1.71
25	retention_ad_strategy	15	1.71
26	landing_page_diagnosis	15	1.71
	Total	877	100.00

---

Dataset Creation

Curation Rationale

To provide a domain-specific dataset for fine-tuning or evaluating Large Language Models (LLMs) and classifiers so they understand the context of performance marketers, such as fixing expensive ad costs, analyzing funnels, and improving ad creatives.

Source Data

• Data collection: [Specify the source, e.g. real cases, synthesis via expert prompting, or marketing forum threads]
• Language: [Thai / English / Mixed]

Annotations

• Annotation process: [Specify whether experts reviewed the data (human-in-the-loop) or categories were assigned semi-automatically]

---

Considerations for Using the Data

Imbalance and Rare Classes

The largest class, creative & messaging, makes up 18.13% of the data, while the smallest classes, such as cpm_analysis, retention_ad_strategy, and landing_page_diagnosis, have only about 15 records each (1.71%).

When training or fine-tuning a model, consider:

• Using stratified splitting when creating train / validation / test sets
• Using a loss function that handles imbalance, such as focal loss or class-weighted cross-entropy
• Applying data augmentation to minority classes

---

Additional Information

Licensing Information

This dataset is released under the [MIT / CC-BY-4.0] license.

Citation Information

@misc{marketing_diagnosis_dataset,
  author    = {[Yothin Sangarun]},
  title     = {Performance Marketing Intent & Diagnosis Dataset},
  year      = {2026},
  publisher = {Hugging Face}
}