VIETNAM NATIONAL UNIVERSITY HO CHI MINH CITY UNIVERSITY OF INFORMATION TECHNOLOGY ADVANCED PROGRAM IN INFORMATION SYSTEMS HA THI KIEU OANH - 19521983 APPLYING AI MODELS TO JOB SEARCHING BACHELOR OF ENGINEERING IN INFORMATION SYSTEMS THESIS ADVISOR DR. DANG KHANH HUNG HO CHI MINH CITY, 2024 ACKNOWLEDGMENTS To complete this thesis, I would like to express my profound gratitude to those who have contributed and supported me throughout the research and execution of this project. First of all, I express my deepest gratitude to Dr. Dang Khanh Hung, who spent a lot of time guiding and advising me throughout the implementation process.
The teacher's support as well as his enthusiastic support helped me complete the project smoothly. Second, throughout the course of this academic journey I would like to express my especially rich gratitude to Dr. Cao Thi Nhan as my advisor. Thank you for accompanying me throughout my university journey.
Next, I would like to send my most sincere thanks to my friends and family. Your encouragement, sharing, and help have given me the motivation to persevere through difficult and tiring times. Most especially, I am very grateful that when I was depressed, my family always stood by me, supporting and encouraging me to overcome it. To achieve the best research results, | must also thank the lecturers of the Department of Information Systems.
Their responsiveness to my questions demonstrated a commitment to academic excellence that went above and beyond, significantly enhancing my understanding of the subject. I would like to thank everyone, whether directly or indirectly, who helped and contributed to the success of this thesis. Your valuable support has helped me complete this research project in the most successful and complete way. TABLE OF CONTENTS ABSTRACT ou.
The problems and its significance. Ăn TT TH nu HH ngờ 12 1. Structure of the the€SIS.- --G- G ST TH nh ng tr 13 Chapter 2 : LITERATURE REVIEW. SG TH SH SH SH 11H 111011111118 ng ng, 14 Chapter 3 : BACKGROUND AND RELATED WORKS.
SH TH HH HH HH HH. Input Embedding oe eee eecccesceesecesceeeseceeeceaeeesceceaeeeseeceaeeseneeeaeeenaeeeaeeees 21 3. _ The difference between RoBERTa and BERT in the embedding process. 'Tokeۖ1Zaf1OTI.
--- E611 11111 10111 ng ng re 25 3. LG ng TH nu HH nu Hết 26 3.- Ăn TH TH TH TH HH HH HT ch TH 29 3. SH HH TT TT TT HH HH HH TT Tà 31 3. Ăn TT TT TH HH HH HH ch TH 31 3.
SẶ Ăn TH HH HH nh tt 32 3. Get elements (Jobs' name, Descriptions, SkIÏÏs. Front-end WebsIf€.á- ch HT TH TH TH HH HH TT ch nh ng 38 3. Back-end WebsIfe.
- ch TH ng HH tt nh 40 3. Back-end Model S€TV©T.- ó- <6 + 1E vn ng nh nh 42 Chapter 4 : SYSTEM ANALYSIS AND DESIGN. _ Training Model ArchIf€CfUTC. System database đ€SIØI.
- -G- (1g ng nh ng 54 Chapter 5 : IMPLEMENTATION AND DEPLOYMENTT. User Ïnf€TÍaC€.- Ăn TH TH TH TH HH HH HH nh ng 58 5. Ăn nh TT TT TH HH Hà HH TT TT 63 5. Ăn TH TH TH TH HH HH HH nh th cư 65 5.- -ó- Ác 1v HH nh nh tt 67 5.
68 Chapter 6 CONCLUSIONS AND FUTURE WORKS. - 11T TT TH TH HH HH TH TT TH 71 6. Future Research Directions: 00. eee esceseeseeseeseeseesecseceseessesseeeeeeaeeseeseeaees 72 REFERENCES.- Án TH HT HT HT TH TT HT Tàn nh TH nhiếp 73 APPENDIX A GUIDE TO SET UP OUR MODEL AND SERVER.‹- 75 APPENDIX B SOME IMAGE ABOUT WEBSTTE.
sóc ssreeeree 79 APPENDIX C EXAMPLE CV WE CAN USE. ce ccceccesseeseeseeneceeeeeeseeeeeeseeseeaeeaes 82 LIST OF FIGURES Figure 1: Visualization of the KNN algorithm.-- -- 55s «<< ++se+ses2 16 Figure 2: The transformation of vectors using soffmax. ---‹---««++- «+2 18 Figure 3: Comparison between cross-entropy function and squared distance. 19 Figure 4: Work embeddIng.-- << + x13 9111 91 9 vn ng rệt 21 Figure 5: Data preprocessing in BERT.-- --- c6 31133 EESerrreersee 23 Figure 6: VnCoreNLP taSK.
s1 vn ng ng ng ng trệt 26 Figure 7: The mechanism of operation of the Transformer Encoder. 28 Figure 8: Deep Averaging n€fWOTK. -- «+ + x11 29 Figure 9: The architecture of the Transformer model.-‹---««+««++s+++ 30 Figure 10: Crawl job from TopCV, TopDey.:eeceeceeseesceeseesseeseeeeeeseenneeneeens 37 Figure 11: Job Dataset (Kag8Ì€).- -- cv ng ng 38 Figure 12: ReactJS ÏOEO. - -- cv TH kg 39 Figure 13: NodeJs OgO.
Ăn TH ng Hà 4] Figure 14: Flask with Python. ceceecssesscssecseesecsseessesecsseeesesseeeseessessaeeneeeas 42 Figure 15: Training model ArchIf†eCfUTe.- - -- «+ + «+ +s£++se£+sesseeeesees 44 Figure 16: System ATrCHf©CfUTC.-- -- G21 132211883 11 113 111 8 111 8 111g rey 46 Figure 17: Skills description Of CV.- c1 1111 11111 11111811118 111g re, 48 Figure 18: Skills Of CT. kh TH HH nu nhờ 49 Figure 19: Predict Job TUfiCfIOH.- -s - <6 1% 311 83111 1 911 11v ng rệc 49 Figure 20: API for getting job recommendafIOTS.-- «5s ««++sx++s +2 49 Figure 21: Accuracy for 2 language predicted jobs model.-- ---««- 50 Figure 22: Database ÏesIgn.-- 5< 1k1 ng ệt 54 Figure 23: The results of analyzing CÝY.- c1 v1 vn ng key 56 Figure 24: Get Job recommendafIOTIS.-- s6 + + +svEEseeEseeeseeereeee 57 Figure 25: Application ATCHIf€CfUTG.- - - 5 1 1E VE+SEEEEeeEseekesrkrreee 58 Figure 26 : LOIN PØ.- - ch TH HH HH nhờ 59 Figure 27: Sign Up Page. cece cecsscssecsseesecssecsseeseesecseesesssesseeeaeeeseeseeeeseaeeeeees 59 Figure 28: Home Page.
eee eessecesseccesseecssseeesseeecsseeessaeecssseecesaeecssaeeesseeeesaeees 60 Figure 29: Home Page (VietnarmeS€). vn key 60 Figure 30: Uploading CV. 62 Figure 32: Result when upload process đOne.- «+5 ++++++s+svesseesxs 62 Figure 33: Job ID€SCTIPDfIOH.-- (G1192 119 911119 11 vn ng key 63 Figure 34: U00 8v 07. 65 Figure 36: Careerbuilder W€DSIfC.- ---- 5 11v ng ng ng ng 67 Figure 37: Our Home Page WebsIfe.- ----- -- <1 1v 9 ng ng 68 Figure 38: Login Page.
79 Figure 39: Home pagel c.e- 80 Figure 40: User Profiloedbc. eo 81 Figure 41: ori: Ce Ƒ ee \ n. 82 Figure 42: Youn: CCS ee 82 Figure 43: Format8) s1. 82 LIST OF TABLES Table 1: Example Result LIST OF ABBREVIATIONS Abbreviation Full form Al Artificial Intelligence K-NN K-Nearest Neighbors GMM Gaussian Mixture Model BERT Bidirectional Encoder Representations from Transformers NLP Natural Language Processing DAN Deep Averaging Network ABSTRACT This topic research the application of artificial intelligence (AI) models to the job search process, in order to improve the efficiency and accuracy of finding suitable jobs.
In this research, we built and deployed a job search system based on advanced AI models, K-Nearest Neighbors (k-NN). This system not only helps users find jobs that match their skills and experience, but also recommends potential career opportunities based on data analysis. Our model is designed to help job seekers, educators, career counselors, and employers by offering tailored career advice. It looks at current job market trends and your specific skills to suggest jobs that fit you best.
By focusing on what you can do rather than just your job history, this tool offers a clearer and more accurate way to explore career options. We tested the system on a large data set, including thousands of job and candidate resumes, to evaluate the system's effectiveness and accuracy. Test results show that the system has the ability to accurately predict and provide useful suggestions to users, helping them save time and effort during the job search process. The results obtained from this study not only demonstrate the application potential of AI in the field of job search but also open up new research directions for the development of smarter career support systems in the future.
Context Using AI to personalize the job search experience: AI models will be used to provide job seekers with a more personalized job search experience, tailored to their skills, experience, and their preferences. This may include recommending suitable jobs, providing advice on writing resumes and cover letters, and assisting with interview preparation. Automate recruiting tasks: AI models will be used to automate many time- consuming recruiting tasks, such as screening resumes, evaluating candidates, and scheduling interviews. This will free up time for recruiters so they can focus on more important aspects of the recruiting process, such as building relationships with candidates and making hiring decisions.
Using AI to improve diversity and inclusion: AI models can be used to help eliminate bias in the hiring process and ensure that all candidates have an equal chance of being hired use. This can be done by removing biased language from job descriptions and resumes and using AI algorithms to evaluate candidates more objectively. Using AI to create more efficient labor markets: AI models can be used to create more efficient labor markets by connecting job seekers with suitable employers. This can be done by using data about job seekers' skills, experience and interests to match them with the most suitable jobs.
Overall, applying AI models to the job search industry has the potential to revolutionize the way people search for and get jobs. By personalizing the job search experience, automating recruiting tasks, improving diversity and inclusion, and creating more efficient labor markets, AI can help people find jobs that suit them and maximize their potential. The problems and its significance - Problem: Bias: AI models are built on available data, and that data may contain built- in biases and prejudices in society. This can lead to AI models making unfair decisions against job seekers, such as discriminating based on race, gender or age.
Transparency: The inner workings of many AI models are difficult to understand, making it difficult to know exactly why the model made a particular decision. This can cause difficulties for job seekers if their resume is rejected by AI and they don't understand why. Privacy: AI models often request large amounts of personal data from job seekers, such as work history and skills. It is important to ensure that this data is collected, stored and used securely and in compliance with privacy regulations.
Automation and job loss: The rise of AI in recruiting could lead to the automation of many recruiting tasks currently performed by humans. This could lead to job losses in human resources and related fields. - Importance: Efficiency: AI is capable of processing large amounts of data quickly and accurately, which can help improve the efficiency of the job search process. For example, AI can help screen resumes and match candidates with more suitable jobs.
Personalization: AI can be used to provide job seekers with a more personalized job search experience. For example, AI can recommend jobs that match a job seeker's skills, experience, and interests. Wider reach: AI can help connect job seekers with employers globally, which can open up more job opportunities for job seekers. e Minimize bias: When designed and used properly, AI can help minimize bias in the hiring process by eliminating subjective factors that can lead to discrimination 1.
Motivation - For job seekers: e Increase job search efficiency: AI can help job seekers find and apply for jobs that match their skills, experience and interests more quickly and easily. This can save job seekers time and effort and help them increase their chances of finding the right job. e Improve application experience: AI can help job seekers create CVs and cover letters tailored to each position applied for, and practice answering interview questions. This can help job seekers be more confident and perform better during the application process.
e Expanding job opportunities: AI can help job seekers connect with potential employers globally, which can open up new job opportunities that they may not have known about. - For employers: e Reduce recruitment time and costs: AI can automate many time-consuming tasks in the recruitment process, such as screening CVs, scheduling interviews, and evaluating candidates. This can help recruiters save time and money, while allowing them to focus on more important aspects of the recruitment process.