Our projects

Our project entailed the development of an AI-powered anti-fake detection service, designed to scrutinize news content and identify instances of misinformation based on specific manipulation criteria. This advanced tool analyzes various elements within news reports to detect common signs of fake news, such as:

• Excessive Emotional Content: Detecting news that triggers strong emotional reactions, often used to cloud judgment or manipulate readers.
• Sensationalism: Identifying when stories are exaggerated to create buzz or attract more attention than the facts warrant.
• Complex Terminology Misuse: Spotting the abuse of jargon or complex language that is not understood by the general population, potentially misleading readers.
• Repetitive Statements: Recognizing when a specific statement is repeated, either word-for-word or paraphrased, to unduly emphasize a point or idea.
• Loud Slogans: Identifying catchy but empty slogans used to captivate and misguide the audience.
• Grouping and Separation: Detecting divisive tactics that categorize people into groups, often used to foster conflict or bias.
• Half-truths: Spotting instances where partial truths are presented, misleading readers about the full context.
• Concept Substitution: Recognizing when terms or concepts are swapped to distort the truth.
• Emotional Adjective Use: Identifying the use of emotionally charged adjectives to bias the perception of the news.
• Pseudo-experts: Detecting scenarios where opinions from so-called experts are presented as undeniable facts.

By leveraging AI and machine learning algorithms, our service meticulously scans and cross-references these criteria against news content to determine its authenticity and reliability. This technology is crucial for combating the spread of misinformation, helping maintain a well-informed public and preserving the integrity of news dissemination.

The development of our fake news detection service involved creating a comprehensive web solution utilizing a variety of advanced technologies. We employed Python for its versatility and strong support for data analysis and machine learning. PyTorch was chosen for developing and training the AI models due to its efficiency in handling deep learning tasks. For natural language processing (NLP), libraries like gensim, spacy, NLTK, and transformers were integral in analyzing and understanding the textual content of the news.

To manage the flow of data and ensure robust communication between different components of the system, RabbitMQ and WebSocket were implemented. Selenium played a crucial role in automating data collection, while MySQL and Elasticsearch were used for database management and efficient search functionalities, respectively. The entire front-end interaction was built using JavaScript to provide a dynamic and responsive user experience.

Three AI models were trained and launched as part of this service, each meticulously designed to analyze news content in real-time, published on various online platforms. These models evaluated the news based on ten specific factors to determine their authenticity.

A significant part of developing this tool was gathering large datasets to train our Large Language Model (LLM). We undertook extensive parsing of diverse news sources, structuring the data properly for effective machine learning. This process was crucial in creating a robust model capable of accurately detecting fake news. Additionally, a dedicated team of 10 news editors was involved in manually labeling the news content, providing the necessary training labels for the machine learning models. This collaboration ensured that the AI models were well-trained to identify nuances and patterns in fake news, making our detection service reliable and effective in combating misinformation.

In the design of our fake news detection service, the primary focus was on crafting a highly functional and user-friendly admin panel, as the service does not feature a public-facing website or interface. This admin panel is the central hub from which users can manage the analysis process, entering the URL of the news article to be evaluated and subsequently receiving a detailed report along with a percentage indicating the likelihood of the news being fake.

The design process for the admin panel was approached with meticulous attention to clarity and efficiency. We aimed to create an interface that streamlined the user's workflow, allowing for easy submission of news URLs and clear, intuitive access to the resulting reports. Given the complex nature of the analysis being performed, it was crucial that the panel remained straightforward to navigate, ensuring that users could effortlessly manage and interpret the results.

Key design considerations included a clean, minimalistic layout that prioritized functionality and avoided any unnecessary visual clutter that could distract from the main tasks. The interface was structured to facilitate quick actions and easy access to historical data and analytics, empowering users to make informed decisions based on the AI-driven insights provided.

To support these design goals, we employed a coherent color scheme and typography that enhanced readability and guided the user's attention to the most critical information. Interactive elements were designed to be responsive and informative, providing immediate feedback and detailed explanations of the analysis outcomes to aid in understanding the assessment of news authenticity.

The resulting admin panel emerged as a powerful tool in the fight against fake news, embodying a design that marries simplicity with sophistication, ensuring that users can leverage the full capabilities of the AI analysis without the need for extensive training or technical expertise.