Duration: October 2021 - October 2023

Budget: 178,750.00 €

Scientific Responsible: Ioannis Vlahavas

Category: National Projects

Project's URL: MIA project website

The object of the project is the research for the development of new machine learning algorithms in order to create a library of algorithms for automatic recognition of polarity and emotions, grouping and classification of texts by Media (newspapers, radio, television), online news and social media/networks, for the purpose of providing and interactive presentation to its users. The library will consist of a collection of Artificial Intelligence algorithms that will automate these processes. A test system will also be implemented to test the algorithms and support the library of machine learning algorithms.

MIA (Media Intelligent Analysis):

  1. Will have the ability to automatically recognize polarity in positive, negative, neutral and emotion in 6 categories (pleasure, sadness, fear, anger, surprise and disgust)
  2. Will recognize hate speech and irony in the text
  3. Will be able to sort the texts into predefined categories
  4. Will group the texts based on their content
  5. Finally, it will present the results to users through an online platform and offer them, interactively, the opportunity to explore and control the texts by category of polarity, emotion, irony, hate and group or content.

MIA will be a library of intelligent algorithms as the achievement of features 1-4 will be done using state-of-the-art machine learning algorithms and will offer users the opportunity to discover extracted information from unstructured texts without technical knowledge.

MIA is a library of algorithms that will:

  • Include innovative machine learning algorithms for classifying and grouping unstructured Greek texts from various sources that will evolve into a complete library of algorithms for extracting information from Greek unstructured texts
  • Intelligently and automatically extract information from unstructured Greek texts
  • Be dynamic and respond in real time to users’ preferences and will enable them to correct labels according to their preferences
  • Be flexible and general purpose so that there is no limit to the areas that it can be used
  • Autonomous and can operate without supervision or human intervention
  • Have the ability to automatically retrain algorithms when there is a high error rate