The invention relates to an ErbB targeted protein family-oriented scoring function construction method based on graph convolution. The method comprises steps of a) constructing ErbB targeted protein data, and constructing a targeted protein data set for four receptor tyrosine kinase ErbB-1, ErbB-2, ErbB-3 and ErbB-4 contained in ErbB; b) expanding the scale of the data set, and expanding the scaleof the data set by adopting methods of original conformation, rotational conformation and optimized conformation; c) constructing an ErbB targeted protein family machine learning scoring function model; and d) training and evaluating the model. The method is advantaged in that an ErbB protein family machine learning molecular affinity evaluation data set is constructed, and a data foundation is laid for a machine learning task of an ErbB protein family; the bottleneck of a classic scoring function is broken through through the GCN technology, performance of the scoring function is improved, and a problem that a machine learning scoring function universal model is low in accuracy is solved.