BanglaNostalgia Dataset

A labeled Bangla text dataset for nostalgia detection from user comments. BanglaNostalgia is a 20K-row Bengali text dataset for binary nostalgia detection. It includes raw and cleaned text, timestamp, and labels. Suitable for Bangla NLP benchmarking, emotion analysis, and low-resource text classification research.

Dataset Snapshot

• Total rows: 20,000
• Main file columns: 5
• Missing values: 0 in all columns
• Label distribution:
  • label = 1: 16,997 rows (84.98%)
  • label = 0: 3,003 rows (15.02%)
• Language: Bangla (Bengali)

Files

• bengali_nostalgia_labeled.csv
  • Full dataset in one file.
  • Columns: id, raw_text, label, text, reference_time
• splits/bengali_nostalgia_labeled_part001.csv ... splits/bengali_nostalgia_labeled_part020.csv
  • 20 split files, each with 1,000 rows.
  • Columns: id, text, label, clean_text, reference_time

Important Column Mapping

Between the main file and split files:

• Split text = Main raw_text
• Split clean_text = Main text

Use this mapping to avoid confusion when switching between the full file and part files.

Column Dictionary

• id: Unique comment identifier.
• raw_text: Original raw Bangla text.
• text: Cleaned/normalized version of the text.
• clean_text: Same semantic role as text (used in split files).
• label: Binary target label for nostalgia classification.
• reference_time: Timestamp string (ISO-like format) associated with the comment.

Label Definition

Current labeling is binary:

• 1: Nostalgia-related expression
• 0: Non-nostalgia / other expression

If your labeling policy differs, update this section before publishing the Kaggle description.

Recommended Use Cases

• Bangla nostalgia detection
• Bangla emotion/affect classification research
• Low-resource NLP benchmarking
• Text preprocessing and normalization studies for Bangla social text

Quick Start (Python)

import pandas as pd

df = pd.read_csv("bengali_nostalgia_labeled.csv")

X = df["text"]          # cleaned text
y = df["label"]         # binary label

Data Quality Notes

• No duplicate id values in the main file.
• No exact duplicate rows in the main file.
• Class imbalance is significant (~85/15), so use stratified splits and class-aware metrics.
• reference_time can support temporal analysis, but verify whether it should be used as a model feature for your task.

Suggested Evaluation Metrics

Because of class imbalance, prioritize:

• Macro F1
• Class-wise Precision/Recall
• PR-AUC (for positive class)

Avoid reporting only accuracy.

Limitations

• Potential label subjectivity (nostalgia can be context-dependent).
• Source-domain bias (platform and audience effects).
• Temporal/cultural bias across years and events.

Ethical and Responsible Use

• Do not attempt user re-identification.
• Use for research/benchmarking, not for sensitive user profiling.
• Respect platform terms if data originates from public comments.

4) File Information

• bengali_nostalgia_labeled.csv: Main file (20,000 rows). Columns: id, raw_text (original), label (0/1), text (cleaned), reference_time (timestamp).

• splits/bengali_nostalgia_labeled_part001.csv to splits/bengali_nostalgia_labeled_part020.csv: Split part file (1,000 rows each). Columns: id, text (same as main raw_text), label (0/1), clean_text (same as main text), reference_time.