How to be Data-Driven and not Data-Distracted around LinkedIn
Author: Richard Hall
LinkedIn
EarthTunnelling.com
This tutorial was inspired by a LinkedIn article I wrote titled:
"When Data-Driven Becomes Data-Distracted"
In the article, I explored what I believe to be a troubling shift in how being data-driven is now often reduced to chasing shallow, gamified metrics, especially on platforms like LinkedIn. Likes, follows, and viral engagement are increasingly treated as signals of expertise and credibility. In reality, they are optimised for emotional clicks, not insight.
That article came about because I noticed a particular creator producing a remarkable volume of content. Several daily posts, often long, image-heavy, and written in highly emotive and performative language. It did not sit right. The pattern felt more like an engineered engagement loop than authentic thought leadership. As I looked deeper, I began to see this behaviour in many other places. But it only really became clear once I started looking at the data.
That was when I realised. If we are going to deal with this problem of influence overriding experience, we need to analyse the numbers behind it. It is easy to suggest people “look at the data”, but most people do not know how to access it. The data is there. Every time we browse LinkedIn, it is loaded onto our machines. But it is buried inside hundreds of megabytes of noise. Even if you know where to look, how do you extract it in a usable form?
Fortunately, this is my world. I have worked with JSON and data analysis for years. Once I had built something that worked for my own investigation, it occurred to me that others might find it useful. So I cleaned it up, added a step-by-step tutorial, and made it available. Now anyone can use it to analyse LinkedIn behaviour for themselves.
This is a simple Google Colab-based notebook designed to help you extract, structure, and explore LinkedIn post data from HAR files (saved from Chrome).
The notebook processes the raw data and produces:
✅ Post content (text)
✅ Post type (original post, share with comment, share with no comment)
✅ Author details
✅ Timestamp of each post (converted from epoch time gathered from image metadata and using interpolation of epoch data)
✅ Engagement data (likes, comments, shares, views where available)
✅ Links between posts (mapping which posts were shared by whom)
It also generates an Excel file containing:
✅ Structured post-level data
✅ Summary statistics:
• Posting frequency over time
• Average word and character counts
• Distribution of post types
Why?
Well, I’ve already explained the background. But once I’d built the code to acquire this data, it struck me that turning it into a tutorial might be useful for others. It would also serve as an example of the kind of “outside of the box” tutorial content I like to create.
Being “Open To Work” leaves you with time on your hands. I figured this was a productive way to use some of it.
I actually discovered Colab while working on training an LLM. I didn’t have access to a decent GPU and wasn’t keen on buying one. Colab was recommended to me because it offers GPU access for a relatively small cost, and the ability to mount Google Drive made life so much easier for managing training data.
That got me thinking, most people already have a Google account. By building this as a Colab notebook, I could make it really simple to use with there being no need to install Python or set up anything locally. Just open the notebook and run it in the cloud.
1️⃣ Open this notebook in Google Colab by clicking on the link:
(Note: GitHub limitation, you'll need to right click on this and select "Open Link in New Tab" to keep this page open)
2️⃣ Follow the instructions in the notebook:
- How to mount your Google Drive
- Set up your required variables
- How to export a LinkedIn HAR file
- How to upload it to Colab
- How to run the analysis
This project is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.
I’ve also recorded a short video demo that walks through the process:
All notes and common questions will be added here.
When browsing LinkedIn’s infinite scrolling areas (such as the feed or profile pages), Chrome continuously loads more content through network requests. If you record a HAR file during this, the file will grow as you scroll.
The issue is not that Chrome drops older data, but that very large HAR files (above 500MB) can cause problems when saving. Once a HAR file grows too large, Chrome may fail to export it properly. In some cases, you may end up with a zero-byte HAR file.
To avoid this, it is best to create multiple smaller HAR files during a capture session. You should experiment with how much data you can scroll before a HAR file reaches an unmanageable size. This depends on your machine and browser version, but keeping each HAR file well below 500MB is generally safe.
- Open Chrome DevTools and go to the Network tab.
- Enable Preserve log.
- Start recording.
- Begin scrolling through LinkedIn content.
- After a reasonable amount of scrolling, before the HAR file grows too large:
- Export the HAR file (for example:
linkedin_1.har). - Clear the Network log.
- Export the HAR file (for example:
- Continue scrolling and repeat the process:
- Scroll → Save HAR → Clear → Scroll → Save HAR → Clear.
- At the end you will have multiple HAR files:
linkedin_1.harlinkedin_2.harlinkedin_3.har- etc.
- It prevents Chrome from failing to save large HAR files.
- It keeps HAR file sizes manageable and reliable.
- It allows you to safely capture large datasets across multiple files.
- The notebook is designed to handle and merge multiple HAR files automatically.
- Do not refresh the LinkedIn page during capture. You can switch tabs or windows, but do not reload the LinkedIn feed.
- The amount of scrolling needed per HAR file depends on your system. Test a few times to find a safe range of data per file.