Knowledge Graph Prediction using Negative Statements: an Approach Based on Entity-nearest Neighbor Count Algorithm

Abstract

This paper presents our contribution to knowledge graph predictions using negative statements (NEGKNOW) challenge. This contribution consists of the definition of the Entity-nearest Neighbor Count (E-NNC) Algorithm. In this algorithm, we consider that if two entities are in relation, then, they linked to least one common entity. The list of common entities between two entities are called their common neighbors. Thus, the algorithm calculates the common neighbor between two entities. The E-NNC algorithm defined in this work was applied for the three tasks of the NEGKNOW challenge. These tasks consists of predicting if there is an interaction between two proteins (Task A), a protein and a disease (Task B) and a gene and a disease (Task C). The organizers of this challenge provided the train and the test set. The algorithm assessed on the train set to evaluate its performance. For task A, the algorithm proves to be powerful because we obtained an accuracy of 0.9. For Task B, an accuracy of 0.9 and 0.5 for task C.