The Innopolis University team has developed an algorithm that automatically changes pronouns in police documents.
The new algorithm was created during the Digital Breakthrough hackathon. Employees of the Center for Artificial Intelligence of a Russian IT university won by developing a solution for the Department of Information Technologies, Communications and Information Protection of the Ministry of Internal Affairs of the Russian Federation. For the 1st place in the competition, they received 750 thousand rubles.
The hackathon participants were required to develop an algorithm that conducts a linguistic analysis of documents, after which it transforms the first-person narration into a third-person text: for example, from the combination “I saw that Ivanov approached me” to “He saw that Ivanov approached him. ” The Innopolis University team trained the neural network on a 12 GB news bulletin dataset. It marks the belonging of the predicate to the desired subject, determines the form of the word and morphological categories.
The neural network results were used to write a Python algorithm based on heuristics and rules of the Russian language. The solution considers the special specifics of departmental texts (which are not always easy to read) and can also process literary texts. The algorithm processes pronouns, verbs, prepositions, quotes, direct speech, determine the belonging of pronouns to names, and highlights characters, dates, amounts of money, locations.
Employees of a Russian IT university proposed implementing the algorithm into a complex service with a web interface. The user can insert text, upload text files of different formats, audio files for speech recognition, and images with text. Also, a plug-in for LibreOffice with the selection of changed parts of the text is developed.
“The developed solution was tested by experts from the Ministry of Internal Affairs of the Russian Federation. The algorithm has shown excellent results using their examples. In 48 hours, we developed a cross-platform autonomous product that is ready for implementation in the department and can relieve law enforcement officers from routine tasks, ”said Semyon Kiselev, team leader, an employee of the Center for Artificial Intelligence of Innopolis University.