Match job with your skill: A simple job recommendation system
Data scientists are in great demand:
In 2012, Harvard Business Review declared that the s data scientist was the "sexiest job of 21st century." The demand for good data scientists has risen in large industries because big data has become mainstream. Businesses are aggressively looking for ways to use the massive amounts of data they are collecting and storing to gain new insights. McKinsey estimated that by 2018, the U.S. economy will have a shortage of 140,000 to 190,000 people with analytical expertise.
Data scientist skillsets can be very different:
As the job title 'Data Scientist' is used fairly loosely, few people knows what it really means. As a result, there could be quite a gap in skillset requirement between two job listing . A 'Data Scientist' could be someone who only knows crunching data with Python, but it also could be someone like Andrew Ng who leads the Baidu AI team in developing great AI products. While one has some data science skills, the other has to possess not only solid theoretical/programming knowledge but also keen business acumen and strong leadership ability. The difference in the skillsets will obviously leads to a salary gap. To gain a deeper insight of this gap, I decided to scrape the information on data scientist jobs info from Glassdoor.com.
Job hunting websites like, Glassdoor, Indeed and Monster are all good resources for acquiring job info. The reason I am choosing Glassdoor is because of its standardized web-page format. Different from Glassdoor, Indeed and Monster redirect job postings to original recruiters' websites. From the perspective of web-scraping, it is hard to guarantee the dataset quality by scraping web pages with different structures.
Why these features:
It is always interesting to find correlations between two variables that appear completely unrelated. Before diving into the website, I brainstormed some questions:
- Which programming knowledge is a must / plus to be a data scientist?
- How much is the salary margin for learning certain skills?
- Which industry offer the highest salary?
With these questions in mind, I picked the following features:
- Job Title
- Job description
- Salary range: This range is estimated/calculated based on Glassdoor collected data
- Location: City / State
- Company name
- Year founded
- Headquarters location
- Revenue: Company annual revenue indicating business size.
- industry: IT/Business service/Health, etc.
- Company type: Private/Public
The Scraping process:
The scraping process can be divided into the following 2 steps:
- Scrape job posting links and salary range: Given certain job name and location, Glassdoor will present 30 related job postings on each page with the corresponding estimated salary range. The salary info will be scraped at this step and merged with the main dataset using link as primary key. Due to the anti-robot function, no more job posting will be provided after 33 pages.
- Feed links to selenium web-driver: After acquiring all job link, the selenium web-driver divided each link and scraped all relevant info. A lesson learned here is that sleep time should be set long enough (4 or 5 sec), otherwise your dataset will be full of missing values.
EDA / Quick facts:
- Data science is big deal across many industries, from IT to government to healthcare. The 12707 job posting covered 25 industries with 59% concentrated in IT (Apple, Oracle, Microsoft, etc.), Business Services (Monitor Group, Mckinsey, Accenture, etc.) and Finance (Chase, JP.Morgan, Capital One, etc.).
- Larger company love data scientist: Looking across all industries, about 35% Data scientist jobs are posted by relatively large employers with more than 10,000 employees. A 10K+ employee sized company has on average 7.931 data scientist, which is more than double the average number of DS (3.852) in 5k-10K employee sized company. The reasons why larger companies have more data scientists are as follows:
- Larger company usually generates massive amount of data in a relatively short period of time. To collect, visualize, analyze and model real time streaming data, more DS are needed.
- Affordability: Data scientists can be expensive. A larger company typically has more financial resources than a smaller one.
3. CA is the state with largest demand of Data scientist, as IT industry are clustered in CA.
- Skill ranking varies by industries: While SQL, Python and R are the top 3 skills in the overall market, skillset ranking are very different across industries. Take IT industry as an example; Hadoop is the No.3 important skill, followed by Java. Spark, as a rising demanded skill, placed sixth. However in business service, the No.3 important skill is Excel, followed by R and python. The skillset ranking largely depends on industry characteristics.
5. The median salary varies a great deal across different industries: In real estate the median salary ($116.94K) is almost double the median salary ($62.42K) for the not-for-profit industry. The following figure delivered a comprehensive image of how data scientist are positioned in each industry. Considering both the facts of number of positions and median salary, the IT industry outperformed all the others.
Within the IT industry, most hiring positions came from Walmart eCommerce. Netflix provides the higher median salary, around $180K.
Project Scope and Deliverable:
Key UI Features and Design Philosophy:
Match your background and skill with job postings.
Users can upload their CV and input skill-set to find job postings requiring similar skill-set. To facilitate user input, the skill-set input bar is categorized into different buckets, including education, major, programming skills, business intelligence and big data skills. When they click the search button, the shiny app will call help.py script to calculate the Jaccard similarity score between user skill-set and job requirements.
Jaccard similarity: The Jaccard index is a statistic used for comparing the similarity and diversity of sample sets. The Jaccard coefficient measures similarity between finite sample sets, and is defined as the size of the intersection divided by the size of the union of the sample sets.
In this scenario, A represents user skill-sets and B represents job requirements.
The similarity between the user and the job is 0.1428 (1/7).
The delivered product:
- NYC Data Science: Shu Yan
- Inspiration from the Landing my dream job by scraping Glassdoor.com by Diego De Lazzari job by scraping Glassdoor.com by Diego De Lazzari