Лучшие темы прошлых семестров 

• Transposable element annotation with graph-based methods
• Исследование игры LLM в мафию с использованием графовых эмбеддингов
• Создание персонализированных маршрутов для туристов на примере города Москва

Graph and Text-based Recommendation System for City Tourists and Residents

Система рекомендации для туристов и жителей города на основании анализа графов и текстов

Description

With increasing amounts of user-generated content, reviews, and social network interactions available online, providing personalized recommendations for city exploration becomes crucial. Traditional recommendation systems often rely solely on collaborative filtering or content-based filtering, neglecting the complex relationships between users, places, events, and contextual text data. This project proposes a hybrid recommendation system leveraging both textual analytics and graph analysis. Text analytics will extract user sentiment, interests, and contextual details from reviews and social posts. Simultaneously, graph analysis will exploit connections among users, attractions, businesses, and events. Combining these insights, our system aims to provide precise and personalized recommendations for tourists and local residents.

Playing Mafia with LLMs

Играем в мафию LLMками

Description

"Mafia" is a popular social deduction game that involves deception, argumentation, and collaboration. This project explores how Large Language Models (LLMs) can simulate human-like behavior in such an interactive, high-stakes setting. Each LLM instance will take the role of a player—Mafia, Detective, or Civilian—and participate in structured dialogues representing day and night phases of the game. The goal is to investigate the models' reasoning, bluffing, persuasion, and lie-detection capabilities in a competitive multi-agent environment.

Transposable element annotation with graph-based

methods

Description

Transposable elements (TEs) are mobile DNA sequences that can significantly im-
pact genome structure and evolution. They are classified into Class I retrotransposons,

which use RNA intermediates, and Class II DNA transposons, which move directly
through DNA intermediates. Accurate annotation of TEs is challenging due to their
non-coding regions, high sequence repetitiveness, and variability.
To address these challenges, a method utilizing machine learning models, including

Graph Neural Networks (GNNs), has been developed for the classification and anno-
tation of TEs. This approach aims to capture the intricate relationships within TE

sequences, leading to improved accuracy in identifying and classifying these complex
genomic elements.