Semantic relationship recognition is the process of document processing and is used to identify the semantic relations contained in the process, which is an important part of the construction of ontology. In the process of constructing petroleum field ontology, the semantic relationship identification is more difficult because the documents in the petroleum field have their unique characteristics. The current semantic recognition algorithm is mainly based on association rules' recognition algorithm, but there is no field-specific orientation. By analyzing the characteristics of petroleum documents, this study proposes a semantic relationship recognition algorithm for petroleum documents based on improved word vector. Based on the Continuous Bag-Of-Words (CBOW) model, this study carries out expanded model training on petroleum terminologies and introduces negative sampling and subsampling techniques to improve the training accuracy and efficiency. Feature vectors are used in training the Support Vector Mechine (SVM) classifier for semantic relationship recognition. The experimental results show that the word vectors trained by this method can accurately identify the semantic relations contained in documents in the petroleum field and have obvious advantages.