---
title: Bengali Text Classification Overview
url: https://www.emergentmind.com/topics/bengali-text-classification
type: topic
---

# Bengali Text Classification Overview

Bengali Text Classification is the task of assigning structured labels (topic, sentiment, style, intent, emotion, etc.) to unstructured Bengali-language texts, including documents, sentences, or utterances, under resource and morphological challenges characteristic of the language. The field spans classical supervised learning, deep learning, and recent advances in self-supervised and generative modeling, as applied to newspaper categorization, opinion mining, code-mixed sentiment, intent detection, authorship attribution, and communal-violence identification.

## 1. Datasets and Annotation Paradigms

Bengali text classification research has established a range of datasets differing in size, source, domain, and label structure. Major contributions include:

- **News Classification:** Potrika (664,880 manually labeled articles, 8 classes) [2210.10903], BAN-ABSA headlines dataset (9,014 headlines, each with topic and sentiment labels) [2511.18618], Kaggle Prothom Alo (437,948 articles, 9 classes analyzed) [2601.12132], and the Seven-Class corpus (212,184 news/blog articles) [2308.13545].
- **Sentiment and Emotion:** User reviews from Daraz (15,194 manually reviewed entries, mapped to 2 or 9 sentiment classes) [2411.19584], UBMEC for six-way emotion (13,072 samples) [2210.06405], five public Bangla sentiment datasets (SAIL, ABSA, BengFastText, YouTube, CogniSenti) [2011.10106], code-mixed BnSentMix (20,000 annotated social/e-commerce texts, 4 sentiment classes) [2408.08964].
- **Intent and Question Classification:** BNIntent30 (4,433 utterances, 30 intent classes) [2312.10679], two-stage question typing pipeline [1912.00127].
- **Document and Hate Speech:** Large genre-diverse document collections (376k+ for document classification, 35k for hate speech) [2004.07807], communal violence detection (Tasnim et al. 12,791 curated instances) [2506.19831].
- **Stylometry and Authorship:** 450 balanced literary stories for three-way authorship [1208.6268], 90 samples for fine-grained stylometry [1210.3729].

Annotation strategies vary from manual majority voting (Potrika, BAN-ABSA, Daraz) to machine-guided labeling (LDA topic induction [2210.10903]), code-mixed filtering with mBERT for code-switch detection [2408.08964], and expert panel refinement for subjective or subjective/violent content [2506.19831]. Notable is the use of multi-label classification for noisy labels and the prevalence of stratified splits to mitigate class imbalance.

## 2. Feature Engineering and Embedding Approaches

The progression from lexical to distributed semantic representations in Bengali classification mirrors state-of-the-art multilingual NLP:

- **Bag of Words and TF–IDF:** Classical frequency-based vectors and term weighting remain competitive; SVM on normalized TF–IDF yields up to 92.6% F1 on 12-class news categorization [1701.08694], with similar effectiveness in web document classification [1410.2045].
- **Distributional Embeddings:** Word2Vec and GloVe support deeper learning models; doc2vec is effective for ML baselines [2210.10903]. FastText—particularly BengFastText, trained on 250M Bengali articles—outperforms static embeddings on both classification and analogy tasks [2004.07807].
- **Sequence and Subword Models:** Trainable embedding layers (128 to 300-dimensions) are central for large corpora [2308.13545]. Subword tokenization underlies transformer input, mitigating OOV and morphological sparsity [2011.04446], essential for code-mixed and morphologically complex content [2408.08964].
- **Hybrid and Lexicon Features:** Domain-adapted lexica, such as the 1,500-word sentiment polarity dictionary, produce rule-based pseudo-labels that, when fused with neural models, enhance performance on nuanced sentiment categories [2411.19584].

## 3. Model Architectures and Optimization Strategies

A spectrum of model architectures has been systematically benchmarked and extended to Bengali:

- **Classical Learners:** SVMs, logistic regression, SGD classifiers, random forests, and k-nearest neighbors, with doc2vec or TF–IDF features, remain robust for high-dimensional structured text [1701.08694, 1410.2045].
- **Deep Neural Networks:** CNNs, (Bi)GRUs, and (Bi)LSTMs are dominant for sequential modeling, typically initialized with pre-trained embeddings [2210.10903, 2011.10106, 2004.07807]. Hybrid frameworks (e.g., CNN + BiLSTM + Attention) outperform single-branch baselines for headlines and sentiment [2511.18618].
- **Generative Models:** Deep generative approaches (LSTM-VAE, AC-GAN, AAE) provide compressed, discriminative latent spaces for classification tasks, with AAE features yielding ≈98.4% F1 in seven-way news discrimination and approaching BERT-level performance at a fraction of the vector size [2308.13545].
- **Transformers and Pretrained Language Models:** mBERT, XLM-RoBERTa, BanglaBERT, and Qwen variants set new performance ceilings in all major subfields. Monolingual BanglaBERT and multilingual XLM-RoBERTa consistently surpass earlier models by 5–29% accuracy across tasks [2011.04446, 2506.19831]. LLMs with instruction-tuning, LoRA adapters, and 4-bit QLoRA quantization (e.g., Qwen 2.5-7B, LLaMA 3.x) have achieved up to 72% accuracy and F1≈74.2% on balanced nine-class news categorization [2601.12132].
- **Adversarial and Semi-supervised Learning:** GAN-augmented BERT (GAN-BnBERT) introduces a generator-discriminator structure that modestly but consistently improves intent classification (Δaccuracy +0.68pp over standard BERT) and smoother convergence [2312.10679].

## 4. Task-Specific Adaptations and Multilingual Challenges

Bengali text classification confronts low-resource constraints, morphological complexity, code-mixing, and class imbalance:

- **Code-mixed Sentiment:** BnSentMix demonstrates that fine-tuned transformers (English BERT, mBERT, XLM-RoBERTa, BanglaBERT) perform on par (Acc/F1 ≈ 69.8%) for code-mixed Bengali–English sentiment, with "mixed" labels being hardest to accurately identify [2408.08964].
- **Fine-grained and Multi-label Tasks:** Multi-label and hierarchical variants (LDA+KNN for topic, hybrid lexicon-BERT cascades for nine-class sentiment [2411.19584]) address nuanced opinions, aspect-based sentiment, and ambiguous utterances.
- **Commonsense and Contextual Cues:** Integrating explicit linguistic knowledge (e.g., Bengali WordNet-based sense definitions [1508.01349]) or detailed stylometric markers (n=76 in stylometry [1210.3729]) improves disambiguation tasks.
- **Mitigating Imbalanced and Noisy Data:** Undersampling/oversampling, weighted loss, and paraphrastic augmentation reduce bias towards majority classes [2511.18618, 2506.19831], with ensemble voting further enhancing stability and generalizability.

## 5. Performance Benchmarks and Comparative Analysis

Empirical results across major benchmarks consistently establish the superiority of transformer-based architectures, but also provide strong baselines from classical and hybrid models. Illustrative summary:

| Task/Corpus                                    | Best Model                      | Accuracy (%) | Macro-F1 (%) | Reference           |
|------------------------------------------------|----------------------------------|--------------|--------------|---------------------|
| 9-class News (Kaggle Prothom Alo, balanced)    | Qwen 2.5-7B + QLoRA             | 72.0         | ~74.2        | [2601.12132]        |
| 8-class News (Potrika)                         | GRU + FastText                  | 91.83        | ~90          | [2210.10903]        |
| 7-class News (Fang et al.)                     | AAE (32-dim), BERT (768-dim)    | 98.4 (AAE)   | 99.1 (BERT)  | [2308.13545]        |
| 5-class News (BengFastText)                    | BengFastText + MConv-LSTM       | —            | 87.1         | [2004.07807]        |
| 30-class Intent (BNIntent30)                   | GAN-BnBERT                      | 96.73        | 96.7         | [2312.10679]        |
| 4-class Headline (BAN-ABSA, imbalanced)        | BERT-CNN-BiLSTM                 | 81.37        | 81.54        | [2511.18618]        |
| 6-class Emotion (UBMEC)                        | mBERT                           | 61           | 71.03        | [2210.06405]        |
| Sentiment, 9-way (Daraz reviews)               | BSPS→BanglaBERT Hybrid Pipeline | 89           | 89           | [2411.19584]        |
| Code-mixed Sentiment (BnSentMix, 4-class)      | BERT / XLM-RoBERTa              | 69.8         | 69.1         | [2408.08964]        |
| Communal Violence (Ensemble, 4-class)          | BanglaBERT Ensemble             | —            | 63           | [2506.19831]        |

Metrics are macro-averaged where reported and reflect held-out or test splits. Classical models (linear SVM, kNN on Doc2Vec) remain competitive (F1 ≈ 87–90%) on balanced news tasks [1701.08694, 1410.2045]. Adversarial, generative, or hybrid approaches close the gap with resource-intensive models without proportional memory overhead.

## 6. Interpretability, Error Analysis, and Future Directions

Interpretation of Bengali text classification models increasingly employs representation probing (cosine similarity of word embeddings to diagnose class confusion), LIME for local token-weight explanations, and error cluster analysis. Common error types include:

- Confusion among semantically or temporally proximate labels (e.g., time vs. date vs. distance in intent) [2312.10679]
- Overlap in embedding space between communal and non-communal lexicon, leading to high-confidence misclassifications [2506.19831]
- Borderline polarity (Slightly Positive/Neutral/Negative), especially in fine-grained sentiment [2411.19584]
- Underperformance on minority or mixed sentiment classes due to intrinsic imbalance [2408.08964]

Research priorities include large-scale domain-adaptive pretraining for Bengali, smarter data augmentation (context-aware paraphrase, back-translation), margin-based and contrastive objectives for separation of confusable classes, hierarchical and multitask approaches, and further integration of lexicon-driven and pretrained semantic representations.

Ensemble meta-models, parameter-efficient fine-tuning (LoRA/QLoRA), and hybrid rule–neural pipelines are emerging as best practices for competitive accuracy under computational constraints. Benchmarks, datasets, and codebases are increasingly open-source, supporting reproducibility and rapid field advancement.

Source: https://www.emergentmind.com/topics/bengali-text-classification