srjecsSRJECSSRJECS2788-9408https://doi.org/10.47310/srjecs.2026.v06i01.004Use Recommendation Systems to Analyse Resumes Using Deep Learning Analysis of a Resume Classification DatasetLuay IbrahimKhalafDepartment of Computer Science, College of Computer Science and Mathematics, Tikrit University, Tikrit, IraqThis article focuses on the application of recommendation systems within the realm of internet application technology. Recommendation systems leverage artificial intelligence and ML to create algorithms to predict user preference for products and services through acquisition (recruitment) of employment. In the area of ​​recruitment, the ATS reads resumes, categorizes candidates and makes decisions based upon data extracted from these resumes with the use of natural language processing, artificial intelligence and other advanced technologies. The article discusses both theoretical and practical applications of recommendation systems within the framework of learning models for resume analysis. Additionally, the article demonstrates how learning models using recommendations can be combined with deep learning techniques (the foundation of many recommendation systems) to improve the effectiveness of resume matching and to aid recruiters in identifying candidates quickly. Recommendation algorithms along with learning models were used to create a hybrid method of analyzing resumes based upon a dataset of 13,389 resumes that had been classified into their respective categories. The new model for tracking systems is a hybrid combination of content-based matching, BERT (Bidirectional Encoder Representations from Transformers) sentence inclusion and machine learning classification strategies to enhance the recruitment process by improving the accuracy of how a system matches candidates. The following four sections describe the process by which the authors leveraged the learning models to enhance their recruiting systems: (1) The authors used BERT to evaluate the difference between BERT and two other learning models (TF-IDF and Support Vector Machine or SVM) by using accuracy, F1, precision and recall metrics, (2) Section II analyzes the recruitment process in more detail, providing a detailed analysis of error (source of errors) in the recruitment system, as well as the implications of noise present in OCR (Optical Character Recognition), (3) The authors created a candidate matching system by creating hybrid candidate matching tools, incorporating both similarity-based matching and ML as a means to improve match results, (4) The authors develop a path for identifying named entities as a method for extracting skills from the body of text of a resume, (5) The authors presented data analysis results and created visual representations of their results using PCA (Principal Component Analysis) and various other techniques and (6) The authors provide additional support for their analysis by creating a version that analyzes Arabic resumes using AraBERT (a BERT based model for Arabic). System trials were conducted and the BERT model achieved an F1 score of 94%, while the TF-IDF+SVM model achieved an F1 score of 85%. These figures demonstrate the effectiveness of the transformer-based models. The ATS Production framework, used globally for automating recruitment processes, was also developed.