Function code evaluation method and device based on neural network

The invention discloses a function code evaluation method and device based on a neural network. The method comprises the following steps: acquiring a target code snippet from a standard code for realizing a specific function, and acquiring a sample code snippet from a realization code of a corresponding function written by research and development personnel; through the same feature extraction algorithm, feature data of program capacity and key program elements in the code snippets are extracted to form seven-dimensional feature vectors, and tag feature vectors and sample feature vectors are correspondingly obtained after normalization processing; for the constructed neural network model, using a label feature vector as a label value, and using a large number of sample feature vectors as input data for training; and evaluating the probability that the to-be-tested code can realize the corresponding function by using the trained neural network model. According to the method, the digitalization and characteriza