Resume Parser Algorithm Deep Dive

For the technical curious, this section will dive into the ResumeFlow parser algorithm and walks through the 4 steps on how it works. (Note that the algorithm is designed to parse single column resume in English language)

Step 1. Read the text items from a PDF file

A PDF file is a standardized file format defined by the ISO 32000 specification. When you open up a PDF file using a text editor, you'll notice that the raw content looks encoded and is difficult to read. To display it in a readable format, you would need a PDF reader to decode and view the file. Similarly, the resume parser first needs to decode the PDF file in order to extract its text content.

While it is possible to write a custom PDF reader function following the ISO 32000 specification, it is much simpler to leverage an existing library. In this case, the resume parser uses Mozilla's open source pdf.js library to first extract all the text items in the file.

The table below lists 0 text items that are extracted from the resume PDF added. A text item contains the text content and also some metadata about the content, e.g. its x, y positions in the document, whether the font is bolded, or whether it starts a new line. (Note that x,y position is relative to the bottom left corner of the page, which is the origin 0,0)

Step 2. Group text items into lines

The extracted text items aren't ready to use yet and have 2 main issues:

Issue 1: They have some unwanted noises.Some single text items can get broken into multiple ones, as you might observe on the table above, e.g. a phone number "(123) 456-7890" might be broken into 3 text items "(123) 456", "-" and "7890".

Solution: To tackle this issue, the resume parser connects adjacent text items into one text item if their distance is smaller than the average typical character width, where$$\mathrm{Distance}=\mathrm{RightTextItemX₁}-\mathrm{LeftTextItemX₂}$$The average typical character width is calculated by dividing the sum of all text items' widths by the total number characters of the text items (Bolded texts and new line elements are excluded to not skew the results).

Issue 2: They lack contexts and associations.When we read a resume, we scan a resume line by line. Our brains can process each section via visual cues such as texts' boldness and proximity, where we can quickly associate texts closer together to be a related group. The extracted text items however currently don't have those contexts/associations and are just disjointed elements.

Solution: To tackle this issue, the resume parser reconstructs those contexts and associations similar to how our brains do it, through grouping them into lines and sections.

Step 3. Group lines into sections

Now that we have the lines, the next step is to group them into sections. The main idea is to find section titles and group their following lines (content) together. Section titles are detected by checking whether a line is bolded and has short text content (a heuristic). Once section titles are identified, the rest of the lines that follow are grouped as their content until the next section title is reached.

A section is represented by its section title and an array of lines. The table below shows the 0 sections that are detected: The section titles are bolded and the lines associated with the section are highlighted with the same colors.

Step 4. Extract resume from sections

Feature Scoring System

As a demonstration, the table below shows 3 resume attributes in the profile section of the resume PDF added.

Feature Sets

Core Feature Function

Special Case: Subsections

And that is everything about the ResumeFlow parser algorithm :)

Written by Xitang on June 2023