Link Prediction Between Food Products With Graph Embedding Using Random Forest Algorithm

In 2020, Islam was recorded as the second biggest religion in the world with 1,9 billion followers. In Indonesia, the Muslim population reaches 263 million people, or 87,2% of the total population. As the Muslim population is growing, the need for halal food products is increasing. Because of the many organizations that created halal certificates in their own country, the source of information regarding halal food products is hard to integrate. The result is obtaining information regarding halal food products is getting harder. To solve this problem, in the previous research, created Linked Open Data Halal (LODHalal) to collect data from a lot of sources and try to integrate those data. But, there is still a lot of redundancy happening and there is still no information about the relationship between food products. So to predict the relationship, this research will create a link prediction for food products. Link prediction is predicting any link between two entities or nodes in a network. Link prediction will be created with data from LODHalal, KlikIndomaret, and HalalMUI. Data will be transformed to vector using graph embedding algorithms which are Node2Vec and RDF2Vec. Those vectors will be used as features for link prediction using Random Forest algorithm. The result is random forest model with Node2Vec has a better performance with 82.7% f1-score and 83.4% accuracy, meanwhile Random Forest model with RDF2Vec has a slightly lower performance with 79.7% f1-score and 80% accuracy. The link prediction model with random forest also resulted in the best performance with halal food product data compared to Gaussian Naïve Bayes, Logistic Regression, and Gradient Boosting Classifier.