Turkish Sentiment Analysis (Keras LSTM + BERT)

2026-02-26 · Source: NLP on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Software Development & Engineering · Depth: Intermediate, quick

Summary

An end-to-end natural language processing (NLP) pipeline has been developed for Turkish sentiment analysis, integrating automated web scraping, custom Long Short-Term Memory (LSTM) model training, and a BERT-based Transformer model. The project begins with data preprocessing using Pandas and Regex for cleaning, followed by tokenization and padding. The custom LSTM architecture features a 32-dimensional embedding layer, a 64-unit LSTM layer with 20% dropout, and a Softmax output for multi-class classification (Positive, Negative, Neutral). A hybrid validation system combines the custom model, fine-tuned on niche data, with the pre-trained BERT-base-Turkish-Sentiment model from Hugging Face, using confidence thresholding to reduce false positives. Real-time data acquisition is enabled by a BeautifulSoup web scraper that bypasses bot detection. The entire system is presented through a user-centric desktop application built with CustomTkinter, offering features like batch analysis from URLs, real-time single-entry prediction, and a professional dark-themed UI.

Key takeaway

For NLP Engineers building sentiment analysis systems, integrating a custom LSTM model with a pre-trained Transformer like BERT-base-Turkish-Sentiment can significantly enhance prediction reliability, especially when combined with confidence thresholding. Consider developing a dynamic web scraper to move beyond static datasets and provide real-time analysis capabilities, and wrap the solution in a user-friendly GUI like CustomTkinter for broader accessibility.

Key insights

A hybrid NLP pipeline combines custom LSTMs and pre-trained BERT for robust Turkish sentiment analysis with real-time data acquisition.

Principles

• Combine custom models with pre-trained Transformers for reliability.
• Implement dropout for LSTM models to prevent overfitting.

Method

The method involves Regex-based data cleaning, tokenization, LSTM training with embedding and dropout layers, hybrid validation with BERT, and real-time web scraping for dynamic data input.

In practice

• Use BeautifulSoup for dynamic web scraping.
• Employ CustomTkinter for accessible desktop GUIs.
• Apply confidence thresholding for hybrid model predictions.

Topics

• Turkish Sentiment Analysis
• NLP Pipeline
• LSTM Networks
• BERT Transformers
• Web Scraping

Code references

• Kubraakk/CustomerSentimentAnalysis

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

Related on AIssential

• An End-to-End Hybrid Framework for Rumour Detection in Low-Resources Algerian Dialect — A hybrid framework combining transformer embeddings with classical classifiers enables effective rumour detection in low-resource Algerian dialect.
• Python Sentiment Analysis: From Simple Tools to BERT — Effective sentiment analysis requires matching the right Python tool to text complexity and business needs.
• From Raw Text to Machine Learning: A Complete NLP Pipeline Explained — An NLP pipeline systematically transforms raw text into numerical data for machine learning models through cleaning, preprocessing, and vectorization.
• Disaster Question Answering with LoRA Efficiency and Accurate End Position — A LoRA-optimized Japanese BERT-Bi-LSTM hybrid system delivers high accuracy and efficiency for disaster-specific question answering.
• Adapting Multilingual Embedding Models to Turkish via Cross-Lingual Tokenizer Surgery and Offline Distillation — Efficient adaptation of multilingual embedding models to specific languages is achievable through tokenizer surgery and offline distillation.
• Model in Distress: Sentiment Analysis on French Synthetic Social Media — Synthetic data generation with backtranslation can overcome data scarcity and privacy issues in multilingual sentiment analysis.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by NLP on Medium.