Designing a Query Neural Network

Abstract

This paper presents the Query Neural Network (QNN). The QNN embeds a set of rules in an Artificial Neural Network (ANN) which answers queries in a backward chaining style. This work can be classified in the research field of 'Integration of Machine Learning and Reasoning'. This field has shown the advantages when combining hand-built-classifier and empirical learning. Furhter, a current line of research in this area is the study of the integration of goal-directed reasoning with backward chaining into an ANN (D'Avila Garcez et al., 2019). While there are already some tools that implement goal directed reasoning, none of them can do this in propositional logic, which also can handle negations and hard rules. Moreover, the QNN tries to close this research gap. The Design Science Research methodology was chosen as an appropriate strategy to design, implement, and evaluate the QNN. First, a concept of the QNN was created. Afterwards, it was implemented within a python program . To evaluate the QNN a sample data from FHNW from the application process for the master's degree in business information systems were used to test if the QNN meets its testing criterias. It was evaluated if the goal-directed reasoning works accurately and the QNN meets its requirements. The evaluation results showed that the QNN meets its requirements.