Launching Concepts for LinkedIn Engagement Insights

9th March 2017 0 Comments

Turning the vast amount of unstructured data available on LinkedIn into privacy-first actionable insights for marketers can be a daunting task. That’s why today we’re launching Concepts for PYLON for LinkedIn Engagement Insights to help to surface what’s driving engagement and what’s grasping the attention of their audiences on LinkedIn’s platform, and get to the valuable insights faster.

PYLON for LinkedIn Engagement Insights processes vast amounts of LinkedIn data. As it sifts through all of the articles shared on the platform, Concepts performs a range of operations designed to add further value to the data. For example, articles are dissected and scanned for entities using Named Entity Recognition. In this process Concepts identifies organizations, locations, products, countries, cities and people mentioned in the content and makes this data available for filtering and analysis.

Each entity found – a concept –  is stored with its name and type as meta-data alongside the demographics of the author of the article as well as anyone who engages with it. The result is a rich data object that can be used to uncover insights about what content drives engagement across audiences. Below is an example of the entities found in a text. Entities are marked in bold.

Artificial intelligence, which was mostly Hollywood fiction until a decade ago with Star Wars, Iron Man and Terminator, is now turning into reality at amazing and sometimes scary pace. With the advent of Artificial Intelligence, Internet of Things, Deep Learning and Robotics, we are witnessing the world transform around us. Everything is being inter-connected into one large neural network. Intelligent machines that can learn and mimic reasoning and human thought processes are now real.

 

Concepts for LinkedIn Engagement Insights recognizes a range of entities and applies the top 20 most significant ones to the data object. Each entity has name and a type: the types are useful if you want to focus on a specific category of entities. We currently support these types:

City Company Country Education Entertainer Government Agency Location
Misc Organization Person Political Party Politician Product Sportsperson

 

A picture says more than a thousand words, so here’s a small example showing how Concepts can be used.

Consider a Media Planner trying to tune content to specific audiences. For example, a Media Planner for a supplier of cognitive computing might want to know what it is about ‘artificial intelligence’ or ‘machine learning’ that resonates with people in different audiences. We can look at which Concepts over-index for, say, people in finance or people in engineering inside the broad topic of ‘artificial intelligence’.

To get started, we search through article titles and summaries to find entities related to the keywords: ‘artificial intelligence’.

Artificial Intelligence
Machine Learning
Automation
Top 3 entities for ‘artificial intelligence’

By extracting the top 3 entities for articles discussing ‘artificial intelligence’, we learned that ‘machine learning’ and ‘automation’ are related terms, so we will include these terms in our following queries.

First we’ll extract the top entities and list them by name if they match content containing our top 3 entities. As evident by the word cloud below there are many and they appear varied.

Cloud1

We can use types to focus our analysis on a subset of entities. Here we are showing two word clouds containing entities of type ‘person’ and ‘company’.

Cloud2Cloud3

Next we look at demographic data in LinkedIn Engagement Insights to find audiences engaging with the terms around our top 3 entities on ‘artificial intelligence’. Let’s break down the result by employer industry and function to hone in on an audience definition. LinkedIn Engagement Insights lets us do this, since the demographic information about people engaging with content matching the entities are automatically included and accessible to us.

We’re using color coding to highlight cross sections between industry and function that are uncommonly engaged with ‘artificial intelligence’.

 

industry by function

We see that a number of functions are relevant judging from the size of the squares in the plot. Let’s take a closer look at ‘business development’ and ‘engineering’ to see how they differ across entities. This can be useful as an input to content creation.

FUNCTION
Business Development Engineering
ENTITIES Forrester Research Open-source model
Forbes Deep learning
Supercomputer Artificial neural network
Natural language processing Mathematics
Predictive analytics Regression analysis
Computer science Data science
Cambridge Sports science
Medicine Application programming interface
Artificial neural network Speech recognition
Big data Global Positioning System
Selected top entities over-indexing on ‘artificial intelligence’ by Function

From the selection above we see that they differ (perhaps unsurprisingly) between people in the two functions. This is just a small example, but it highlights the value of being able to dial in on interests across audiences and use this insight to fine-tune messages.

LinkedIn Engagement Insights contains a host of demographic dimensions and you can pivot the data on all of these. The analysis shown here ran over a 30 days historical data sample that is ready available through the API.

If you would like to find out more about Concepts for PYLON for LinkedIn Engagement Insights, get in touch.

Share This