![]() |
ИСТИНА |
Войти в систему Регистрация |
ИПМех РАН |
||
Building complex event representation and event forecasting are two important problems which are usually solved separately. In this work we propose general approach that allows to incorporate all available information about events, such as numeric, categorical and binary features and even some text descriptions. We propose ways to build latent representation of complex events using one of dimensionality reduction methods (matrix factorization, neural autoencoders) together with forecasting further occurrences of these events. The experimental results compare combinations of different methods for event representation and forecasting.