VKontakte Sentiment Analyser

Overview

The VKontakte Sentiment Analyser applies real-time natural-language processing to public posts from Russian news agencies on VKontakte. It tracks how official narratives around the Ukraine conflict are framed within Russian social media. The application fetches posts via the VK API, filters them for relevant war-related keywords, and analyses sentiment using the RuBERT model (sismetanin/rubert-ru-sentiment-rusentiment).

Key Features

• Fetches recent posts from major Russian outlets such as TASS, Kommersant, and RT using the VK API.
• Filters posts for specific user-defined keywords connected to the Ukraine conflict.
• Splits text into individual clauses for fragment-level sentiment scoring to capture nuance.
• Aggregates fragment scores into overall post sentiment, displayed with clear colour-coded categories: blue (positive), grey (neutral), red (negative).
• Allows users to expand any post to view fragment-level reasoning behind each sentiment label.

Technical Details

The backend, built with Python (Flask), manages VK API requests, text preprocessing, and inference via the Hugging Face Transformers library. Posts are cleaned, split by punctuation, and processed through the RuBERT model for polarity prediction. Results are aggregated numerically and returned to the Bootstrap 5 / JavaScript front end for visualisation. The design supports authenticated access tokens stored securely via python-dotenv, and can easily extend to additional platforms such as Telegram or Twitter.

Rationale & Purpose

This project explores how social-media data can enhance open-source intelligence and sentiment mapping within restricted information environments. By monitoring Russian state-affiliated outlets in near real time, the system offers insight into tone shifts, propaganda trends, and audience sentiment surrounding the Ukraine conflict. It demonstrates the integration of language proficiency, machine learning, and data ethics in a single operational tool.

Links