The Stream Blog

Personalized Job Feeds and Machine Learning

Product managers for job sites face two fundamental problems. First, the top candidates are not actively looking for a job, making them difficult to seek out and find. Second, the top jobs are quickly filled and typically attract in-network candidates. So, while you have top candidates and great jobs on your site, it can be hard to connect the two at the right time. This blog will unpack the 4 steps to unlock personalization technology, as pioneered by social networks, to improve the connection between job seekers and jobs.

Why stop at that though? We’re also offering up the opportunity to talk to one of our experts who will provide a review of your existing application and provide useful suggestions to improve engagement and to ultimately match the right candidate to the right job. We’re so sure that the insights will be so useful we’re throwing in an entry to win a drone for everyone who receives this custom evaluation. Learn more.

Step 1 – Track explicit user preferences

As a first step you’ll want to store all the explicit user preferences. This includes things like location, industry, seniority level, pay range, and job titles. For instance, let’s say that a user searches for programming jobs in Boulder with a salary above $100k. After the user has entered this search parameter you can give them the ability to follow or save the search. While this is a good starting point for creating personalized job feeds, setting up filters is a lot of work and most of your userbase won’t take the time to configure them. This is why tracking implicit user intent (or step 2) is so important.

Step 2 – Track implicit user intent

There are many ‘events’ on your job site which provides good insights into a user’s preferences. Common examples include:

  • Clicking on a job listing
  • Saving a job
  • Opening an employer profile
  • Applying for a job
  • Clicking on a job in an email

Each of these events signal intent and provide clues about what a given user is interested in. Over time you will learn that a given user often looks at open jobs at Apple, or maybe he/she is looking for jobs focused on a certain programming language. These small events build up and allow you to create a profile about a given user’s interests.

Step 3 – Leverage the network

Another way to deliver the most relevant jobs is to leverage the rest of the user base to improve job recommendations. Find users with similar criteria and see what jobs they’re viewing and applying to. There are a lot of correlations that can be used based on user job history (titles, companies, locations) as well as user behavior.

Step 4 – The feedback loop

Combining these 3 techniques will create a strong feedback loop. Every time a user visits your app you learn more about their interests, thus your job recommendations improve. Even passive job seekers continuously get better recommendations if they even occasionally open your digest email. These tailored recommendations encourage people to come back to your site, further improving the quality of your recommendations. As you can imagine, this feedback loop greatly improves user engagement.

feedback loop for job search

Increasing User Engagement using personalized job recommendations and recommender systems

Presenting the job recommendations to job seekers can be done in several ways. The obvious is a single feed shown on your app’s homepage. An alternative approach is a weekly email digest with recommended jobs.

Most product owners are continually looking for ways to increase user engagement, time spent in the app and methods for re-engagement. An email with interesting and relevant jobs not only reminds job seekers to return to the app, but can also keep passive seekers interested. Engagement (opens, clicks, etc.) with the email is also tracked and fed into the engine, refining the results even further.

The power of machine learning and feeds

This engine I spoke of leverages machine learning, which enables you to build, train and tune a model that is designed to predict individual user interests. This type of technology was pioneered by social networks such as Youtube, Instagram and Facebook. Two great examples of machine learning in action are Instagram’s discovery feed and Quora’s digest email. Every time you engage with Quora or Instagram the recommendations become a little bit more targeted.

Imagine a list of jobs presented to users that continue to get more and more relevant – piecing together things that piqued their interest without them even knowing it. Suddenly the perfect job bubbles to the top, combining everything they’re looking for, giving them room to grow but drawing on the best parts of their experience to make them the ideal candidate. Personalization gives job sites a great opportunity to build up a competitive advantage by better understanding a user’s interests.

Building these personalized feeds in-house can be a daunting challenge. It requires expertise with large scale feed storage, analytics tracking and machine learning. This is prohibitively expensive for many companies, so Stream provides an affordable alternative by offering personalized feed technology as a service. Our API is used by over 500 companies and 200 million end users.

Complimentary Application Evaluation (and drone giveaway)

Whether you have an existing feed or haven’t thought about feeds until now, we’re offering a complimentary evaluation and review of your application. Give us just 20 minutes and we believe you’ll walk away with actionable suggestions to drive engagement and improve your candidate-to-job hit rate. Plus to sweeten the deal, and because we’re cool like that, we’re entering all evaluation participants into a drawing to win a drone.

Fill out the below form to get the process started. We promise you won’t regret it…or flying your new drone.


Also published on Medium.