A Slack Thread That Sparked Change

During a recent Slack thread, one of our recruiters expressed frustration with the number of applicants whose resumes didn’t match the job descriptions we had crafted. They pointed out that many candidates were using the same generic terms, which made it difficult to assess their true fit for roles like software engineer or frontend developer. This led us to question how effectively we were parsing and matching CV keywords for developers.

The Stakes of Accurate Matching

As we delved deeper, it became evident that the accuracy of our keyword matching directly impacted our hiring efficiency. With numerous vacancies to fill, time wasted sorting through resumes filled with irrelevant terms meant lost opportunities for both candidates and our hiring team. The mismatch not only frustrated recruiters but also left promising candidates overlooked, ultimately affecting our ability to build strong teams. We realized that resolving this issue was critical to streamline our hiring processes and improve candidate experiences on our platform.

Identifying the Core Problem

The primary issue we faced was that our existing parsing algorithm struggled with synonyms and variations of terms. For example, a candidate might list “JS” instead of “JavaScript,” and our algorithm often failed to recognize that these referred to the same skill. This resulted in potential matches being discarded. One particular instance involved a qualified candidate whose resume emphasized their experience with “React” but lacked the exact phrasing we had in our job postings, leading to their application being overlooked entirely.

Initial Attempts at Resolution

Our first attempt to tackle this issue involved simply expanding our keyword list to include common variations. However, this led to a case of keyword stuffing. We noticed that candidates began to overload their resumes with numerous variations of the same skill in an attempt to game the system. This not only diluted the quality of the resumes but also made it harder for our parsing system to discern genuine expertise from mere padding. Clearly, this approach wasn't sustainable.

The Technical Solution: Enhanced Skill Extraction

After reevaluating our strategy, we decided to implement a more sophisticated skill extraction algorithm based on natural language processing (NLP) techniques. We introduced a synonym recognition feature that could identify various terms for the same skill, ensuring that candidates using “JS” would not be penalized for not using “JavaScript.”

Here's a simplified version of the code that illustrates our approach:

import nltk
from nltk.corpus import wordnet

# Function to extract skills from resume
def extract_skills(resume_text):
    skills = set()
    words = nltk.word_tokenize(resume_text)
    for word in words:
        synonyms = wordnet.synsets(word)
        for synonym in synonyms:
            skills.add(synonym.name())
    return skills

This implementation allows us to recognize synonyms, broadening our matching capabilities without relying on keyword stuffing.

Observable Changes in the Product

After deploying the new algorithm, we observed a notable improvement in the quality of matches. Recruiters reported a higher number of relevant applications, and candidates felt more confident that their skills were recognized accurately. The feedback loop improved significantly, evidenced by increased engagement on our /jobs page and a drop in the time taken to shortlist candidates. We also saw a positive shift in candidate satisfaction, with more developers feeling their expertise was adequately represented.

Key Lessons Learned

Through this process, we gleaned several important insights:

• Avoiding Keyword Stuffing: Candidates should focus on genuine skills rather than trying to game the system.
• Importance of Synonyms: Recognizing variations in terminology can drastically improve matching accuracy.
• User-Centric Design: Enhancements should prioritize the candidate experience to build trust in the application process.
• Iterative Improvements: Regularly revisiting and updating our algorithms is essential to keep pace with industry changes.

Insights for Candidates

For candidates, this means it's crucial to tailor your resume to highlight relevant skills without overloading it with variations. Focus on clarity and context. Ensure your skills section reflects genuine expertise, as modern parsing systems are designed to recognize proficiency without needing every possible synonym. This way, your resume has a better chance of aligning with the job descriptions that matter to you.

Insights for Recruiters

From a recruiter’s perspective, implementing a more intelligent parsing system can significantly improve candidate quality. By understanding that candidates may use different terminology for the same skills, you can fine-tune job descriptions and ensure that promising candidates are not overlooked. This approach fosters a more inclusive hiring process, enabling you to tap into a broader talent pool.

Looking Ahead

While we've made substantial progress, there are still areas for improvement. We are monitoring how well the new skill extraction performs across various domains and remain open to refining our approach based on user feedback. If we had to undo any part of this process, it would likely be the initial reliance on keyword expansion without considering context. Moving forward, we’re committed to continually enhancing our systems to ensure that both candidates and recruiters find the best matches possible.