100,000+ Movie Reviews from Kazakhstan: Russian, Kazakh, and Code-Switched Texts

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, short

Summary

A new publicly available corpus of 100,502 movie reviews from Kazakhstan, collected from kino.kz and spanning 2001–2025, covers 4,943 unique film titles. This multilingual dataset primarily features Russian reviews, alongside Kazakh and code-switched texts. Reviews are manually annotated for language and sentiment polarity, with 11,309 also including explicit user-provided ratings. Researchers defined two sentiment tasks: three-way polarity classification and five-class score classification. Benchmarking compared classical BoW/TF–IDF baselines against multilingual transformer models like mBERT, XLM-RoBERTa, and RemBERT. Experimental results indicate that transformer models consistently outperform classical baselines on polarity classification. However, score classification remains challenging under leakage-controlled evaluation due primarily to severe class imbalance and subtle distinctions between adjacent rating levels.

Key takeaway

For NLP engineers developing sentiment analysis models for diverse language data, especially those involving code-switching, you should consider the new Kazakhstani movie review corpus. This dataset offers a valuable resource for training and evaluating multilingual models on Russian, Kazakh, and code-switched texts. Use transformer models for robust polarity classification, but be prepared to implement advanced techniques to mitigate class imbalance and subtle distinctions when tackling fine-grained score prediction.

Key insights

This new multilingual Kazakhstani movie review corpus enables sentiment analysis research, highlighting transformer efficacy for polarity but challenges for fine-grained score classification.

Principles

• Multilingual transformers excel in polarity classification.
• Fine-grained score classification is difficult with imbalance.
• Code-switched text presents unique NLP challenges.

Method

The method involves collecting 100,502 reviews, manual annotation for language and sentiment, defining two sentiment tasks (polarity and score), and benchmarking classical BoW/TF–IDF against mBERT, XLM-RoBERTa, and RemBERT models.

In practice

• Utilize mBERT, XLM-RoBERTa, or RemBERT for polarity.
• Address class imbalance in score classification.
• Explore code-switching in multilingual NLP.

Topics

• Movie Reviews
• Sentiment Analysis
• Multilingual NLP
• Code-Switching
• Kazakh Language
• Transformer Models
• Dataset Creation

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• KazakhOCR: A Synthetic Benchmark for Evaluating Multimodal Models in Low-Resource Kazakh Script OCR — Current MLLMs struggle with OCR and language identification for low-resource Kazakh scripts, especially Arabic, performing worse than traditional OCR.
• Teaching Machines to “Feel” Tweets — Domain-native pre-trained models are crucial for accurate sentiment classification in ambiguous social media language.
• Extending Item Response Theory for Efficient and Meaningful Multilingual Evaluation — Multilingual-IRT extends Item Response Theory to efficiently evaluate LLMs across languages, identifying translation errors and cultural bias.
• TeamHerald@CHIPSAL 2026: Hate Speech Detection and Sentiment Analysis of Nepali Memes using Transformer-based Architectures and Ensemble Learning — Ensemble strategies for meme analysis perform differently for binary vs. multi-class tasks, requiring task-specific aggregation.
• SHIFT: Semantic Harmonization via Index-side Feature Transformation for Multilingual Information Retrieval — SHIFT is a training-free indexing method that corrects language bias in multilingual retrieval by adjusting document embeddings.
• Zero-Shot to Full-Resource: Cross-lingual Transfer Strategies for Aspect-Based Sentiment Analysis — Cross-lingual transfer strategies for ABSA vary in effectiveness based on model architecture and resource availability.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.