Introducing Highlights

Since launching in 2013, Slack has helped millions of users across hundreds of thousands of teams communicate more efficiently, effectively, and transparently. But as Slack lowers the barriers to communicating internally, the volume of communication that results can be overwhelming. McKinsey estimates that knowledge workers spend 28% of their time managing digital information; accordingly, proactively…

Slack logo
Slack
•
Jerry Talton
•8 min read•intermediate•
--
•View Original
ChefJenkins

Overview

The article introduces Highlights, a new feature in Slack designed to help users manage information overload by surfacing important messages based on personalized engagement. It discusses the technical challenges and solutions related to predicting user engagement through a work graph and machine learning models.

What You'll Learn

1

How to leverage machine learning for personalized message engagement in Slack

2

Why understanding the work graph is crucial for improving communication efficiency

3

How to implement logistic regression for predicting user engagement

Prerequisites & Requirements

  • Basic understanding of machine learning concepts
  • Familiarity with regression analysis

Key Questions Answered

How does Slack predict which messages are important for users?
Slack uses a personalized engagement model that leverages a work graph to predict which messages users are likely to engage with. This model analyzes user interactions, message content, and channel activity to surface important messages tailored to individual users.
What challenges does Slack face in predicting user engagement?
One major challenge is class imbalance, as most users engage with only a small fraction of messages. Slack addresses this by employing stratified sampling and evaluating model performance using precision-recall metrics to ensure effective predictions.
What is the role of the work graph in Slack's Highlights feature?
The work graph is a structured network of communication within Slack that captures relationships between users, channels, and content. It is central to predicting message importance and enhancing user engagement by understanding how users interact with each other and their content.
How does Slack ensure the personalization of engagement predictions?
Slack personalizes engagement predictions by grouping user behaviors and training aggregate models per team, incorporating user-specific features like affinity for message authors and channel priorities to enhance prediction accuracy.

Key Statistics & Figures

Time spent managing digital information
28%
According to McKinsey, knowledge workers spend this percentage of their time managing digital information, highlighting the need for tools like Highlights to streamline communication.
Improvement in prediction accuracy
20–50x
Online testing has shown that Slack's learned regression model is significantly better at predicting engagements compared to naive models.

Technologies & Tools

Backend
Machine Learning
Used to predict user engagement with messages based on historical interaction data.
Backend
Logistic Regression
Employed to model the probability of user engagement with messages.

Key Actionable Insights

1
Utilize the Highlights feature to prioritize your message reading based on personalized engagement predictions.
This feature is designed to help users focus on the most relevant messages, reducing information overload and improving communication efficiency.
2
Incorporate user feedback mechanisms to refine machine learning models in your applications.
Feedback helps improve the accuracy of predictions, ensuring that the system adapts to user preferences over time.
3
Leverage logistic regression for engagement prediction models in your projects.
This approach allows for clear interpretability of model decisions, which is crucial for building user trust in automated systems.

Common Pitfalls

1
Failing to address class imbalance can lead to misleading model performance metrics.
When training models, if the dataset contains significantly more negative examples than positive ones, the model may achieve high precision by predicting 'no' for all inputs, which does not reflect true performance.

Related Concepts

Machine Learning
User Engagement Prediction
Work Graph Analysis