<![CDATA[Social assistance needs system accurate recommendations for ensure distribution appropriate target. Research This aims to implement Neural Collaborative Filtering (NCF) to recommend recipient help social based on integration of dynamic parameters of poverty data. The NCF method was chosen Because his ability combines Generalized Matrix Factorization (GMF) and Multi-Layer Perceptron (MLP) to catch non-linear relationship between data. The dataset is taken from 845 recipients assistance in Cijulang Village, District Ciamis, with criteria covering employment, income, health, and family history assistance. The preprocessing stage includes data cleaning, label encoding, one-hot encoding, and data splitting (training-validation 80:20). The NCF architecture is built with embedding layer (dimension 32), hidden layer MLP (128-64-32 neurons), and output layer that combines GMF and MLP. Evaluation using Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). The results show that the model achieves RMSE 0.63 and MAE 0.47 on the training data, but overfitting occurred with a validation RMSE of 1.40 and MAE of 1.24. Analysis indicates the need for hyperparameter optimization (e.g., regulation, dropout rate) for an increase in generalization. Findings This prove NCF potential in increase accuracy recommendation help social, at the same time highlight importance data handling no balance and sparsity in context poverty. Implications study covers improvement transparency distribution assistance and reduction jealousy social through recommendation data -based. This study gives contribution methodological in NCF adaptation for sector public.]]>
