Abstarct: Abstract In recent years, the online recruitment and job searching industry has grown significantly, helping businesses screen a wide range of candidates with different qualifications and capabilities before conducting in-person interviews. On the other hand, applicants can now reach more companies in different geographical locations and send them their resumes, which in turn makes job hunting more competitive among all applicants. Job recommendation systems, along with textual skill extraction algorithms, automatically extract skills from job descxriptions and resumes, recommending the most suitable jobs to applicants. However, such systems come with various challenges, such as flaws in the writing process of job descxriptions and resumes. For instance, some skills may not be mentioned by the writer for various reasons. Also, the lack of a comprehensive standard template for writing resumes and job descxriptions can lead to some information being overlooked. In such cases, automated recommendation systems cannot provide desirable results. Considering these issues, the overall goal of the designed system is to extract as many explicit and implicit skills as possible, without relying on the textual structure of the resume and job descxriptions, and then suggest jobs baxsed on them. In addition, to help applicants in the competitive job search environment, we designed a system that, baxsed on the applicants’ field of profession and their current skills, identifies other important and similar skills to help them better navigate their career path and plan more intelligently. To achieve the goals of this thesis, we developed a system comprising three parts in extracting job skills textual data and improving the job offer. For the first part, we extracted both explicit and implicit skills from job descxriptions and resumes texts, and considering the density of skills in each resume label, we calculated an importance score for each skill in that label. At the end of this part, the most suitable jobs are suggested for each resume by matching their skills while taking their importance into consideration. To evaluate the relevance of the job recommender system results, we used the nDCG metric and obtained an accuracy of 87.27% for the entire resume dataset. In the second part, the skills extracted from the resumes were categorized baxsed on their field of activity (also referred to as their label) and converted from their textual format into a vector using a word embedding model. Then, 10 binary deep learning models were trained to predict each label of the resumes baxsed on their skills. In the third part, we suggested skills for each resume baxsed on their label and the label-baxsed skill categories that were created in the previous part. Then, we re-implemented the job recommendation process on the resume, adding the new suggested skills to them. Finally, to further improve the results, we excluded the suggested skills that negatively impacted the accuracy of job offers. As a result, the system managed to improve the average accuracy of job offers by 33% for a set of 1,000 random resumes with initially poor job offer accuracy.