Let's Talk Job Descriptions: Specifically Using Data to Determine What Language to Use
Are you a fullstack ninja that is excited about wildly-fast-growing startups with fun offices? Or are you a thoughtful developer looking for a supportive workplace with potential for advancement?
Either style of language could be used to describe the same position, and choices like this can be critical for attracting kind of candidates that a company needs. Intuitively choosing a voice that matches a workplace seems like a natural strategy. But when you want to cast a wide net, what is the best way forward?
We're interested in how things like text content correlates with other metrics -- like "apply to job" clicks. One method for measuring and comparing characteristics of text documents (among many) is sentiment analysis. Broadly, methods of sentiment analysis often measure how "positive" or "negative" a text document is by counting key words and terms that are associated with these two opposites.
To get a quick feel for how sentiment might affect job apply clicks, we used a pre-trained sentiment analyzer inside a tool called textblob. We used this to analyze the text of all jobs that ever went live on The Muse. This plot below shows that, according to this off-the-shelf tool, most job posts use mildly positive language.
With each job assigned a sentiment score, we put all job posts into 6 equally-sized groups, from most-negative to most-positive sentiment. The sentiment distributions of each group can be compared in the plot below:
This is a kind of data visualization is called a box plot and helps to summarize how our 6 groups differ. For example, the line in the middle of each rectangle marks the median sentiment score for each group; typical sentiment scores for jobs in a group are near this line. The full rectangle encloses the 50% of the data that are nearest to this line (i.e., the most typical). This kind of summary (which features some raw data overlaid) helps us understand that jobs that feature more positive words, when looking across all job categories, have historically gotten more apply clicks.
There are much more sophisticated ways to look at these qualities, and the plots above only scratch the surface of what data can help us understand. Also different companies have different goals for their job posting -- quality or specificity of job applicants may be more important quantities, for example.
At The Muse, we're using data to understand these and other problems to help job seekers find their dream job, and help companies hire dream employees. If you're a developer interested in working on problems like this, helping people find their dream job, please get in touch.
Chris Ryan is a data scientist at The Muse. He loves writing code to discover stories told by patterns in data, and spent many years doing this for problems in biophysics before taking on his current role. When he's not doing that, he's often planning his next kitchen experiments, going to see bands play in remote parts of NYC, or trying to hold crow pose just a little longer.More from this Author