Jobs by the Ocean: Solving Complex Geography Problems in Job Search

Introduction:

Job search is supposed to be simple. Enter the job title, then a location, click a button and get your search results. However, unfortunately as evidenced by the appalling search experience for job seekers on most job sites, this is simply not the case.

At DirectlyApply we have spent the last 5 years building job search technology to make job search faster, more accurate and more enjoyable for job seekers. To date, we have released more search filters than any other job site, allowing job seekers to search key features such as benefits, salary, education, experience requirements, job type and so on, all built on top of our proprietary job feature extraction and indexing capabilities. We know these abilities help with positive search outcomes (job seeker applying to job) and down funnel (job seeker completing application for job) and based on these metrics we demonstrate the highest performance across the industry.

However, one of the most challenging filters to develop was location as historically this has been achieved by job sites in 1 of two ways:

1. Keyword matching - EG: The user types in “New York” and the job site looks for location fields containing those keywords.

2. Radius from a point - EG:  Enter a known location and find all jobs within a fixed radius distance of this point.

Keyword matching is thankfully mostly phased out in favor of radius from a point - however the implementation of radius from a point by job sites still leaves many issues for job seekers and the overall effectiveness of their job search.

• Geocoding individual job listings to accurate locations is expensive, so shortcuts are often taken which minimizes accuracy

• Multiple locations within the state and/or country with the same name

• If a job is 1 foot outside of the radius it wont show

• It doesn't take into consideration what commuting routes are available across a city

• It doesn’t take into consideration geographic entities such as the ocean, rivers, mountains etc.

• It assumes all cities are circular in design, if a city is long and skinny in shape for example, radius search cuts off the top and bottom of the city.

This post aims to demonstrate what we have done to date to solve the challenging problem of location search and also where we intend to focus efforts to create an even better solution in the future. 

It is important to note, that this is the culmination of 4 years work and expertise, continuously building on top of all our previous learnings.

Going Back to the Beginning & Laying the Foundations

One of the first big projects we undertook when starting DirectlyApply in 2018 was working on a way to geocode (generate coordinates) for every single job and provide a precise location, not only just at the city level, but neighborhood, business location (shop, office) for the individual job post. Despite not knowing at the time how this system would evolve, we knew that it was important to lay the foundations years in advance.

Our in-house geocoder processes millions of job postings a day, extracting and parsing location information and building our internal knowledge-base of job location information. Want a job at McDonalds in New York? Well we know the location of the 27 McDonalds that are hiring across Manhattan, which ones are most convenient for you, and can display all your options accurately to you in less than 1/10th of a second. The result is a level of precision in simple location searching that remains unrivalled by other job sites.

Even in its primitive form, job seekers on DirectlyApply were able to accurately narrow down their job search to specific neighborhoods simply by using a tighter radius search with a high degree of confidence that they aren't missing out on seeing what could be their perfect job.

Continuously building, learning and improving over four years of data and experience in advanced geocoding of job postings gave us the best possible foundation for creating huge strides in location search in 2023.

Part One: Make search visual - Zone Search

In June 2023 we released Zone Search - a new way to search for jobs by location, in a completely new but very intuitive way. Unsurprisingly, most people can’t accurately visualize what a “15 mile radius from Chicago, IL” looks like.

Our initial initial improvement to the 'radius' user experience was therefore the release of zones, a point and radius search, translated into “zones” utilizing Uber’s open-source Hexagonal Hierarchical Spatial Index ( H3 ). This gave a visual reference to job seekers as to what their search query was doing behind the scenes and for the first time on a job site, allowed for easy modifications.

This first version saw a 15% increase in finding a job from job seekers who used this feature - a resounding success.

Part Two: Jobs in the Ocean - Detect & Move

According to the US National Office for Coastal Management (NOAA), 40% of the US population (128 million people) live within coastal counties, this is despite only 10% of the USA's land mass being on the coast.

When we started seeing visually what job seekers' searches looked like, we noticed a lot of their search radius was spanning out into the ocean.

This exposed the problem that, for coastal populations, we were in some cases halving their searchable area. Coastal cities aren't any smaller than inland cities, but we knew we had to detect where the search was going beyond the bounds of land and then bring the search back towards it, unlocking a larger on land search, where the jobs are actually located.

We achieved this by creating a geojson multi-polygon of the whole of the US and then checking each zone (h3 cell) selected in a search to see if it intersects. If it does then we know the selected zone is at least partially over land, and if there is no intersection we know it is out to sea.

We then took this now redundant zone and placed it back on land, closest to the origin as possible. This solved the issue of not trying to search the ocean (or other large bodies of water such as lakes) for jobs and gave more opportunity to discover jobs on land. However, despite the improvement, when reviewing the resulting search areas we noticed that they didn’t reflect the actual shame or makeup of individual cities.

Part Three: Guided by the Jobs - Mapping Posting Density

The final piece of the puzzle therefore was to ensure that the zones selected on land best reflect where the job postings actually are. We did this by calculating the job posting density of each zone and ranking them in order to best pick which zones to choose.

The below graphic shows a search near Miami and what the search would look like without job density prioritisation and then what the final search area looks like after job density has been taken into account.

Results

The impact of these changes on how we search locations on DirectlyApply has been immense. Our key metric for measuring the impact in the first instance is "job visibility" - simply did the number of jobs we were able to show for a given search increase or decrease and by how much.

Across all searches in the US, job visibility has increased by 35%, depending on the location this ranges from 7% increased job visibility to a huge 364% improvement. A sample of searches from real job seekers and the job visibility improvement is below.

Where Next?

The biggest impacts from these updates are on coastal towns, which does impact a large percentage of our job seekers. However, we have also seen a positive uptick in job visibility on inland searches where the job density algorithm has readjusted the default zones to better reflect the makeup of the city being searched.

Our next update to location will therefore include realtime feedback to job seekers as to the number of jobs in each zone as they are being selected. Furthermore, we will also advise where their search could be expanded to an area where their general job search parameters are ideal, but the location might be slightly out of their regular search area.

If you are interested in our engineering updates follow us on Linkedin and check out our releases page .