Person-Job Fit model based on sentence-level representation and theme-word graph

Previous studies on Person-Job Fit mainly focused on the use of large amounts of data to train word-level embedding to model with the Joint Feature Extractor. Therefore, in order to reduce the computational load and text truncation problems caused by word-level representation and better integrate the idea of Click-through Rate prediction, a Person-Job Fit model based on sentence vector and theme-words knowledge graph (KG-DPJF) is proposed. The model uses a Bi-directional Long Short-term Memory Neural Network to extract theme-words from long text of recruitment resume and requirements and construct a knowledge graph. The model uses the Bert pre-training model to code the input sentences, and uses a multi-level attention mechanism to calculate the correlation among candidate resumes, historical resumes and recruitment requirements features, then inputs the correlation and weighted output to the classifier to predict the matching degree. Finally, the experiment is tested in the industry data set, and the KG-DPJF method achieves nearly 9% and 5% performance improvement compared with method based on logistic regression and single convolution neural network feature extraction method, and shortens training time by about four times compared with word-level embedding method.