Hiring teams often receive hundreds of resumes for a single job opening. Reading each one manually takes hours. A smarter approach is to teach a machine to read resumes and pull out the key facts — the candidate's name, job title, skills, education, and employers — automatically.
In this tutorial you will build exactly that. You will train a Named Entity Recognition (NER) model using spaCy — a popular Python NLP library — on a dataset of roughly 200 real resumes. NER is a technique that labels spans of text with categories such as "Person", "Company", or "Location". Here you will define your own resume-specific categories like Companies worked at, Skills, and Degree. Once the model is trained, you will load a brand-new PDF resume, extract its text, and pass it through the model to get a structured summary in seconds.
Prerequisites: Python 3.x, spaCy, PyMuPDF (fitz), and the resume dataset on GitHub.
Understanding the Training Data
Collecting labeled training data is one of the most important steps in any machine learning project. For NER, each training sample pairs a raw text string with a dictionary that marks where each entity starts and ends, and what category it belongs to.
The dataset used here contains around 200 resumes. Each resume was annotated so that spans like a person's name, their employer, or their degree are tagged with the correct label. You can download the full dataset and follow the official spaCy annotation format described in the spaCy training data guide.
Import the standard libraries needed for training:
import spacy
import pickle
import randomLoad the pre-annotated training data from disk. Each entry in the list is a tuple: the raw resume text followed by a dictionary whose entities key holds a list of (start, end, label) triples:
train_data = pickle.load(open('train_data.pkl', 'rb'))
train_data[0]The output below shows the first resume in the dataset together with its entity annotations. Notice that Companies worked at is a custom label, and a single resume can contain many instances of the same label:
('Govardhana K Senior Software Engineer Bengaluru, Karnataka, Karnataka - Email me on Indeed: indeed.com/r/Govardhana-K/ b2de315d95905b68 Total IT experience 5 Years 6 Months Cloud Lending Solutions INC 4 Month • Salesforce Developer Oracle 5 Years 2 Month • Core Java Developer Languages Core Java, Go Lang Oracle PL-SQL programming, Sales Force Developer with APEX. Designations & Promotions Willing to relocate: Anywhere WORK EXPERIENCE Senior Software Engineer Cloud Lending Solutions - Bangalore, Karnataka - January 2018 to Present Present Senior Consultant Oracle - Bangalore, Karnataka - November 2016 to December 2017 Staff Consultant Oracle - Bangalore, Karnataka - January 2014 to October 2016 Associate Consultant Oracle - Bangalore, Karnataka - November 2012 to December 2013 EDUCATION B.E in Computer Science Engineering Adithya Institute of Technology - Tamil Nadu September 2008 to June 2012 https://www.indeed.com/r/Govardhana-K/b2de315d95905b68?isid=rex-download&ikw=download-top&co=IN https://www.indeed.com/r/Govardhana-K/b2de315d95905b68?isid=rex-download&ikw=download-top&co=IN SKILLS APEX. (Less than 1 year), Data Structures (3 years), FLEXCUBE (5 years), Oracle (5 years), Algorithms (3 years) LINKS https://www.linkedin.com/in/govardhana-k-61024944/ ADDITIONAL INFORMATION Technical Proficiency: Languages: Core Java, Go Lang, Data Structures & Algorithms, Oracle PL-SQL programming, Sales Force with APEX. Tools: RADTool, Jdeveloper, NetBeans, Eclipse, SQL developer, PL/SQL Developer, WinSCP, Putty Web Technologies: JavaScript, XML, HTML, Webservice Operating Systems: Linux, Windows Version control system SVN & Git-Hub Databases: Oracle Middleware: Web logic, OC4J Product FLEXCUBE: Oracle FLEXCUBE Versions 10.x, 11.x and 12.x https://www.linkedin.com/in/govardhana-k-61024944/', {'entities': [(1749, 1755, 'Companies worked at'), (1696, 1702, 'Companies worked at'), (1417, 1423, 'Companies worked at'), (1356, 1793, 'Skills'), (1209, 1215, 'Companies worked at'), (1136, 1248, 'Skills'), (928, 932, 'Graduation Year'), (858, 889, 'College Name'), (821, 856, 'Degree'), (787, 791, 'Graduation Year'), (744, 750, 'Companies worked at'), (722, 742, 'Designation'), (658, 664, 'Companies worked at'), (640, 656, 'Designation'), (574, 580, 'Companies worked at'), (555, 573, 'Designation'), (470, 493, 'Companies worked at'), (444, 469, 'Designation'), (308, 314, 'Companies worked at'), (234, 240, 'Companies worked at'), (175, 198, 'Companies worked at'), (93, 137, 'Email Address'), (39, 48, 'Location'), (13, 38, 'Designation'), (0, 12, 'Name')]})Training the NER Model
With the data loaded, you can now build and train the spaCy NER pipeline.
The train_model function below does the following:
- Starts from a blank English model (
spacy.blank('en')). - Adds an
nercomponent to the pipeline and registers every custom label found in the training data. - Disables all other pipeline components so only
nerweights are updated. - Runs 10 training iterations. At each iteration the data is shuffled to prevent the model from learning order-dependent shortcuts.
- Applies a dropout rate of
0.2— dropout randomly zeroes out 20 % of features during each update, which forces the model to learn more robust representations and reduces overfitting.
Create the blank spaCy model and define the training loop:
nlp = spacy.blank('en')
def train_model(train_data):
if 'ner' not in nlp.pipe_names:
ner = nlp.create_pipe('ner')
nlp.add_pipe(ner, last = True)
for _, annotation in train_data:
for ent in annotation['entities']:
ner.add_label(ent[2])
other_pipes = [pipe for pipe in nlp.pipe_names if pipe != 'ner']
with nlp.disable_pipes(*other_pipes): # only train NER
optimizer = nlp.begin_training()
for itn in range(10):
print("Statring iteration " + str(itn))
random.shuffle(train_data)
losses = {}
index = 0
for text, annotations in train_data:
try:
nlp.update(
[text], # batch of texts
[annotations], # batch of annotations
drop=0.2, # dropout - make it harder to memorise data
sgd=optimizer, # callable to update weights
losses=losses)
except Exception as e:
pass
print(losses)Now call the function to start training. Expect this to take several minutes depending on your hardware:
train_model(train_data)The NER loss is printed after each iteration. A steadily decreasing loss means the model is learning to locate entities more accurately:
Statring iteration 0
{'ner': 15037.69734973145}
Statring iteration 1
{'ner': 14219.407121717988}
Statring iteration 2
{'ner': 10966.300542596338}
Statring iteration 3
{'ner': 9323.783655686213}
Statring iteration 4
{'ner': 8716.684299001408}
Statring iteration 5
{'ner': 7539.511302326008}
Statring iteration 6
{'ner': 5957.294807362359}
Statring iteration 7
{'ner': 4725.2797832249025}
Statring iteration 8
{'ner': 5975.5969295029045}
Statring iteration 9
{'ner': 4244.961263079282}The loss falls from roughly 15 000 at iteration 0 to around 4 200 by iteration 9, confirming that the model is converging. The slight uptick at iteration 8 is normal and caused by shuffling.
Saving and Loading the Model
Training a model from scratch takes time. Save the trained model to disk so you can reload it instantly in future sessions without retraining:
nlp.to_disk('nlp_model')Load the saved model back into memory:
nlp_model = spacy.load('nlp_model')Running Predictions on Training Resumes
Before testing on an unseen PDF, verify that the model correctly labels a resume it has already seen. Because random.shuffle(train_data) is called inside train_model, the item now at index 0 will differ from what was there before training:
train_data[0][0]'Vineeth Vijayan "Store Executive" - Orange City Hospital & Research Institute Nagpur, Maharashtra, Maharashtra - Email me on Indeed: indeed.com/r/Vineeth-Vijayan/ ee84e7ea0695181f I have over 2 years of experience working as a "Store Executive" in a reputed big organisation based in India. I am a hard worker and a very good learner. I aspire to grow in my career by joining a good organisation. I am good at work and people management. I possess good computer knowledge. I have underwent training in courses like MS Excel Level 1, 2 & 3. MS Word & Powerpoint. I have experience working in ERP. I am a certified ISO 27001: 2013 Internal Auditor. Willing to relocate: Anywhere WORK EXPERIENCE "Store Executive" Orange City Hospital & Research Institute - Nagpur, Maharashtra - March 2016 to Present Job Profile: - 1. Keep records to maintain inventory control, cost containment and to assure proper stock levels. 2. Rotate stock and arrange for disposal of surpluses. 3. Plan and perform work that involves ordering, receiving, inspecting, returning, unloading, shelving, packing, labeling and maintaining a perpetual inventory of items. 4. Coordinate freight handling, equipment moving and minor repairs. 5. Perform peripheral site-specific duties as required and instructed by supervisors. 6. Monitoring and updating stock details etc. in Mednet store software. Use ABC, VED & FSN Stock Analysis methods when required. 7. Report any stock discrepancy to supervisor for appropriate action. "HR Officer " Gulf Warehousing Company Qatar - Agility Group - January 2012 to July 2015 a multinational company catering Customers in the field of Warehousing, Freight Forwarding, Records Management, Courier and International Relocation Services. Job Profile: - 1. Deliver new hires orientation program to enable new hires to adjust themselves to new working environment and facilitating newcomers joining formalities. 2. Design and develop Training Calendar and budget for all type of training programs, for the year ahead for all designations within the company, according to the needs of the business and Training Need Analysis survey results. https://www.indeed.com/r/Vineeth-Vijayan/ee84e7ea0695181f?isid=rex-download&ikw=download-top&co=IN https://www.indeed.com/r/Vineeth-Vijayan/ee84e7ea0695181f?isid=rex-download&ikw=download-top&co=IN 3. Sourcing and coordinating with external vendors for training that are requested by line managers/departments. 4. Maintaining employee training records and updating it on a regular basis. 5. Assist Supervisors in developing and implementing changes in HR policies and procedures. 6. Coordinate in conducting employee engagement programs. 7. Support departments in complying with ISO Quality Management requirements and facilitate various HR related welfare activities. "Training Coordinator" M-Squared Software & Services - Kerala - Kerala, IN - July 2010 to December 2011 a US based company with headquarters in Nevada, Las Vegas that deals with security gadgets, medical transcription and software development. Job Profile: - 1. Perform regular office administration duties that includes: Office infrastructure management, housekeeping supervision, responding and sending e-mails to head office and clients, monitoring time-in & time-out of employees, managing employee leave status and office cash management. 2. Handle walk-in enquires, attend telephone calls and conduct course induction sessions for new joiners. 3. Train the students in work related processes. 4. Anchoring and conducting course related seminars in well-established educational institutions. 5. Manage training materials that are stipulated within the company regulations. 6. Formulating training programs and sending it to training manager and to all concerned departments for approval. 7. Preparing tools for training. 8. Organize the training that includes: distribution of training materials, monitor trainees in training and support trainers in training. 9. Record training plans and training programs. 10. Maintain employees and students database. 11. Record results of evaluation and training. 12. Archive records of training, including course content, the training, number of students, results, and feedback. 13. Conduct team huddles/counseling sessions for trainees when required. "Manager" Nair IT Pvt. Ltd - Nagpur, Maharashtra - September 2008 to July 2010 Client: Accentia Oak Technologies. Job Profile: - 1. To manage the day-to-day planning, operation and problem-solving of the team to meet the required service level components, quality and productivity targets of the client. 2. Delivery of team statistics, quality and productivity targets & indicators, taking note any downtime during the day and report it to the management level superiors. 3. Operational Management: Managing the floor, adherence to schedule and monitoring of attendance. Submits daily report to the management level superiors. 4. Monitoring the portal periodically, prioritizing jobs according to TAT, and assigning jobs to correct work pool. 5. Coaching and relaying feedback to MTs/Editors. 6. To motivate the team to achieve the client targets. Update the team on the status of work at the start and end of each shift. 7. Compiling reports on team\'s performance and client feedback. "Process Executive" in Infosys BPO Limited, Bangalore CISCO - Bengaluru, Karnataka - January 2006 to July 2008 Job Profile: - 1. To handle end to end order placement process of Cisco. 2. Book orders for customers as per Cisco\'s booking policies. 3. Customer advocacy. 4. Check and avail special discounts, services, licensing, upgrades etc., to the customers. 5. Receive customer feedback on products, services etc., and resolve them if any. 6. To work along with logistic team to ensure that the products reach precise destinations safely. 7. Provide round the clock 24/7 customer service to the client. "Sr. Medical Transcriptionist" DTS Information Systems Pvt. Ltd - Bengaluru, Karnataka - December 2004 to January 2006 in DTS Information Systems Pvt. Ltd, Bangalore (A wholly owned subsidiary of DTS America Inc., Tennessee, USA) from Dec 2004 - Jan 2006. Client: Vanderbilt University Medical Center, Nashville, Tennessee. Job Profile: - 1. To transcribe documents pertaining to patient\'s summary of illness, diseases and diagnostic procedures etc. to avail medical insurance claims as per American transcription guidelines. 2. To ensure that finished end document meets the clients TAT and Quality requirements. 3. To counsel MT\'s regarding their errors etc and if needed conduct team huddles. Computer Exposure Operating System: Windows XP Microsoft Office Package: MS Excel, MS Word, MS Access and Power Point. Experience in working on Oracle. ❖ Completed MS Excel Level 1, 2 & 3 Courses from New Horizons Computer Training Center, Qatar. EDUCATION B.Sc. in Microbiology in Microbiology S.N.G College, Bharathiar University April 2004 SKILLS Storekeeping / Inventory Management / Purchase (2 years), People & Time Management (2 years) ADDITIONAL INFORMATION Competencies ❖ Effective communications skills, excellent interpersonal skills, good time management skills, results-oriented individual and a very good team player. ❖ Sincere, hardworking and willing to take up challenging assignments. Other Qualifications ❖ ISO 27001: 2013 Internal Auditor Certification from Coms Vantage Consultancy, Qatar in October 2014 ❖ ISO 27001: 2013 Awareness Certification from Coms Vantage Consultancy, Qatar in September 2014 ❖ Carriage of Dangerous Goods by Road Certification in 2012 from IRU Academy in tie up with Mowasalat - State of Qatar owned Transport Company. ❖ First class in Hindi Parichaya Examination conducted by Dakshina Bharat Hindi Prachar Sabha. ❖ Second class in Prathamic Examination conducted by Dakshina Bharat Hindi Prachar Sabha.'Pass the resume text to the model and print each entity with its label aligned in a fixed-width column:
doc = nlp_model(train_data[0][0])
for ent in doc.ents:
print(f'{ent.label_.upper():{30}}- {ent.text}')NAME - Vineeth Vijayan
DESIGNATION - Store Executive
EMAIL ADDRESS - indeed.com/r/Vineeth-Vijayan/ ee84e7ea0695181f
DESIGNATION - Store Executive
DEGREE - B.Sc. in Microbiology in Microbiology
COLLEGE NAME - S.N.G College
SKILLS - Storekeeping / Inventory Management / Purchase (2 years), People & Time Management (2 years)The model correctly identifies the candidate's name, job title, email address, degree, college, and skills — seven distinct entity types extracted in one pass.
Testing the Model
Now you will test the model on a completely unseen resume stored as a PDF. The PyMuPDF library (imported as fitz) lets you open PDF files and extract their text page by page.
Install PyMuPDF with pip if you have not done so already:
pip install PyMuPDFOpen Alice Clark CV.pdf, extract all text, and collapse newlines into spaces so the model receives a clean single-line string:
import sys, fitz
fname = 'Alice Clark CV.pdf'
doc = fitz.open(fname)
text = ""
for page in doc:
text = text + str(page.getText())
tx = " ".join(text.split('\n'))
print(tx)Alice Clark AI / Machine Learning Delhi, India Email me on Indeed • 20+ years of experience in data handling, design, and development • Data Warehouse: Data analysis, star/snow flake scema data modelling and design specific to data warehousing and business intelligence • Database: Experience in database designing, scalability, back-up and recovery, writing and optimizing SQL code and Stored Procedures, creating functions, views, triggers and indexes. Cloud platform: Worked on Microsoft Azure cloud services like Document DB, SQL Azure, Stream Analytics, Event hub, Power BI, Web Job, Web App, Power BI, Azure data lake analytics(U-SQL) Willing to relocate anywhere WORK EXPERIENCE Software Engineer Microsoft – Bangalore, Karnataka January 2000 to Present 1. Microsoft Rewards Live dashboards: Description: - Microsoft rewards is loyalty program that rewards Users for browsing and shopping online. Microsoft Rewards members can earn points when searching with Bing, browsing with Microsoft Edge and making purchases at the Xbox Store, the Windows Store and the Microsoft Store. Plus, user can pick up bonus points for taking daily quizzes and tours on the Microsoft rewards website. Rewards live dashboards gives a live picture of usage world-wide and by markets like US, Canada, Australia, new user registration count, top/bottom performing rewards offers, orders stats and weekly trends of user activities, orders and new user registrations. the PBI tiles gets refreshed in different frequencies starting from 5 seconds to 30 minutes. Technology/Tools used EDUCATION Indian Institute of Technology – Mumbai 2001 SKILLS Machine Learning, Natural Language Processing, and Big Data Handling ADDITIONAL INFORMATION Professional Skills • Excellent analytical, problem solving, communication, knowledge transfer and interpersonal skills with ability to interact with individuals at all the levels • Quick learner and maintains cordial relationship with project manager and team members and good performer both in team and independent job environments • Positive attitude towards superiors & peers • Supervised junior developers throughout project lifecycle and provided technical assistancePass the extracted text to the trained model and print the predicted entities:
doc = nlp_model(tx)
for ent in doc.ents:
print(f'{ent.label_.upper():{30}}- {ent.text}')NAME - Alice Clark
LOCATION - Delhi
DESIGNATION - Stream Analytics
DESIGNATION - Software Engineer
COMPANIES WORKED AT - Microsoft –
DEGREE - Indian Institute of Technology – Mumbai
SKILLS - Machine Learning, Natural Language Processing, and Big Data Handling ADDITIONAL INFORMATION Professional Skills • Excellent analytical, problem solving, communication, knowledge transfer and interpersonalThe model correctly extracts Alice's name, location, two job titles, her employer, her degree, and her skills — all without seeing this resume during training. The SKILLS entity bleeds slightly into the ADDITIONAL INFORMATION section, which is a common boundary error in NER models trained on limited data. Training on more diverse resume samples will sharpen these boundaries.
Conclusion
In this tutorial you built a complete resume parsing pipeline using spaCy. You loaded 200 annotated resumes, trained a custom NER model with 10 iterations and 0.2 dropout, saved the model to disk, and tested it on a brand-new PDF — extracting name, designation, employer, skills, and education fields automatically.
Key takeaways:
- Custom NER labels let you define domain-specific entity types (like
Companies worked atorGraduation Year) that generic spaCy models do not provide out of the box. - Shuffling training data at each iteration prevents the model from learning spurious ordering patterns and leads to better generalization.
- A dropout rate of
0.2is a practical starting point — it reduces overfitting without slowing convergence too much. - PDF text extraction with
PyMuPDFis straightforward but may introduce noisy formatting; normalizing whitespace before inference improves entity boundary detection. - More training samples directly improve extraction accuracy — especially for entities like
SKILLSwhose text boundaries are inconsistent across resumes.
Next steps:
- Apply your NER skills to opinion text with Sentiment Analysis using Scikit-learn to see how text classification complements information extraction.
- Explore Multi-label Text Classification on Stack Overflow Tag Prediction to handle documents that belong to more than one category simultaneously.
- Learn how Logistic Regression can serve as a lightweight baseline classifier when you want to score resumes after extracting their entities.
