Knowledge-Driven Modulation of Neural Networks for Predictive Analytics

In this paper, we propose a novel approach to Next-Activity Prediction (NAP) that combines the strengths of symbolic and neural networks. Our approach leverages attention mechanism, which is an emerging technology in the field of Neural Networks (NN), to model relationships in data and make predictions more accurate.
The traditional NAP approaches do not consider the possibility of having a background knowledge that can correct predictions in exceptional circumstances. However, our proposed method takes into account this critical aspect, providing a more robust and reliable approach. We use an emerging technology based on attention mechanism to build a Symbolic-Neural Network (SNN) system for NAP. This attention mechanism goes beyond the classical dense layers, convolutional layers, and LSTM cells, introducing a new general computing mechanism to model relationships in data.
Our approach represents an advancement over the state-of-the-art for several reasons. Firstly, it considers the possibility of having background knowledge that can correct predictions in exceptional circumstances. Secondly, it uses attention mechanism, which has shown promising results in text generation. Finally, our approach provides a more robust and reliable prediction by taking into account the background knowledge.
In summary, this paper proposes a novel approach to NAP that combines the strengths of symbolic and neural networks using attention mechanism. Our proposed method is more robust and reliable, as it considers the possibility of having background knowledge that can correct predictions in exceptional circumstances. This approach represents an advancement over the state-of-the-art for several reasons, including its ability to consider background knowledge and its use of attention mechanism.