基于BERT与LightGBM的人岗匹配模型
A Person-Job Matching Model Based on BERT and LightGBM

在求职招聘市场中，信息不对称导致“逆向选择”，加大了企业招聘和求职者求职的难度。线上招聘平台在疫情时期更加重要，对人岗匹配精度要求更高。传统匹配方式受限，深度学习技术特别是BERT模型和集成模型受到关注。当前学者在研究人岗匹配问题时，采用常见的TF-IDF词向量表示方法和Word2Vec词向量表示方法来对中文文本进行表征，但是由于科学的进步，当下用BERT模型能更好地读取文本语义，因此本文将BERT模型引入到人岗匹配领域中，采取了基于BERT模型的词向量表示和LightGBM模型的人岗匹配方法，以提升匹配精确度和效率，与多种机器学习模型的预测结果相比较之后，最终发现，在这两种方法的结合下，在本文所构建的人才是否投递模型中的精确度达到了0.886，在岗位是否认可模型中的精确度达到了0.926，由这两个模型的效果可以看出BERT模型和LightGBM模型的结合，可以为招聘平台提供精准模型。
In the job recruitment market, information asymmetry leads to “adverse selection”, which increases the difficulty for both enterprises in hiring and job seekers in finding employment. Online recruitment platforms have become even more crucial during the pandemic, placing higher demands on the accuracy of person-job matching. Traditional matching methods are limited, and deep learning technologies, especially the BERT model and ensemble models, have garnered attention. In current research on person-job fit, scholars often represent Chinese text data using common methods such as TF-IDF word vectors and Word2Vec word vectors. However, due to advancements in science and technology, the BERT model is now better at capturing textual semantics. Therefore, this paper introduces the BERT model into the field of person-job fit. This paper proposes a person-job matching method based on the BERT and ensemble models to improve matching accuracy and efficiency. After comparing the prediction results with various machine learning models, it was ultimately found that with the combination of these two methods, the accuracy of the talent submission model constructed in this paper reached 0.886, and the accuracy of the job acceptance model reached 0.926. The effectiveness of these two models demonstrates that the combination of the BERT model and the LightGBM model can provide a precise model for recruitment platforms.