Overview
The article discusses how LinkedIn leverages Natural Language Processing (NLP) to enhance user support by predicting the best answers to help requests. It details the evolution of their help search system, from the initial Care Search to the current NLP-powered solution, which significantly improves user experience for over 610 million members.
What You'll Learn
1
How to implement a Natural Language Processing workflow to improve user support systems
2
Why traditional keyword-based search methods may fail in user support contexts
3
How to measure the effectiveness of a search algorithm using click-through rates and session metrics
Prerequisites & Requirements
- Understanding of Natural Language Processing concepts
- Familiarity with deep learning frameworks like Convolutional Neural Networks(optional)
Key Questions Answered
How does LinkedIn use NLP to enhance user support?
LinkedIn employs NLP to analyze user queries and predict the best help articles, improving the support experience for its members. The system processes over one thousand tickets daily, addressing a wide range of inquiries effectively.
What were the limitations of the initial Care Search system?
The Care Search system struggled with varied user expressions and terminology mismatches, often returning irrelevant results. For example, searching for 'deactivate my account' would yield articles about 'create account' instead.
What improvements were observed after implementing the new NLP system?
The new NLP system led to a significant increase in click-through rates from 39% to 69%, and the 'happy path' session rate improved from 16.2% to 29%, indicating a better user experience.
What metrics are used to evaluate the performance of the NLP solutions?
Performance is measured through click-through rates, 'happy path' session rates, and 'undesired' session rates, which track user engagement and satisfaction with search results.
Key Statistics & Figures
Click-through rate (CTR)
69%
Improved from 39% after implementing the NLP system.
'Happy path' session rate
29%
Increased from 16.2% as a result of the new search algorithm.
'Undesired' session rate
2.8%
Decreased from 6.3%, indicating fewer users are creating cases directly after searches.
Technologies & Tools
AI/ML
Natural Language Processing
Used to analyze user queries and improve the accuracy of help article recommendations.
AI/ML
Convolutional Neural Network
Employed for intent classification of user queries.
Search Infrastructure
Galene
Used for keyword search in the help system.
Key Actionable Insights
1Implementing a Natural Language Processing system can significantly improve user support by accurately predicting user intent.This is particularly useful in large-scale applications like LinkedIn, where user queries can be diverse and complex, ensuring that users receive relevant assistance quickly.
2Utilizing metrics such as click-through rates and session paths can provide insights into the effectiveness of search algorithms.These metrics help identify areas for improvement and gauge user satisfaction, allowing for continuous enhancement of the support experience.
3Adopting a deep learning model for intent classification can enhance the understanding of user queries.This approach allows for better handling of long-tail queries that traditional keyword matching may not address effectively.
Common Pitfalls
1
Relying solely on keyword-based search methods can lead to irrelevant results for user queries.
This occurs because users often phrase their questions differently than the terminology used in help articles, necessitating a more sophisticated approach like NLP.
2
Neglecting to measure user engagement metrics can hinder the ability to assess the effectiveness of search improvements.
Without tracking metrics like click-through rates and session paths, it becomes challenging to understand user satisfaction and identify areas for further enhancement.
Related Concepts
Natural Language Processing In User Support
Deep Learning For Intent Classification
Search Algorithm Optimization