基于word2vec的程序编译错误信息特征提取方法

Feature Extraction Method of Program Compilation Error Information Based on Word2vec

输入表示为固定长度的特征向量是机器学习算法要求之一。针对编程中的编译错误信息特征,论文提出了基于word2vec模型对编译错误信息进行特征提取。利用滑动窗口取词的方式,建立one-hot字典,结合word2vec中的Skip-gram模型,构建Huffman树,从可变长度的文本中学习固定长度的特征表示。最后使用SVM分类算法进行实验结果的验证。结果表明,该特征提取方法在编译错误信息中有显著的效果。

One of the requirements of machine learning algorithms is to represent the input as a fixed length eigenvector.Aiming at the features of compile error information in programming,this paper proposes to extract the features of compile error information based on word2vec model.The one-hot dictionary is established by using the sliding window to pick words,and the Skip-gram model in word2vec is combined to build the Huffman tree to learn the feature representation of fixed length from the text of variable length.Finally,SVM classification algorithm is used to verify the experimental results.The results show that the feature extraction method is effective in compiling error messages.