Driven by the trend of natural language search, LinkedIn announced that its brand-new AI-powered job search feature is now available to all users. This innovative function leverages a refined and fine-tuned model trained on LinkedIn’s professional social media platform knowledge base, aiming to enable job seekers to describe their job preferences using more natural and conversational language, thereby receiving more precise and tailored job recommendations.

Erran Berger, Vice President of Product Development at LinkedIn, said in an interview with VentureBeat: "This new search experience allows members to describe their goals in their own words and receive results that truly reflect their needs. It's the first step in a broader journey to make the job search process more intuitive, inclusive, and empowering for everyone."

A previous survey conducted by LinkedIn found that users tend to overly rely on precise keyword queries when searching for jobs on the platform. This often leads to suboptimal search results, such as finding both media journalists and court reporters when searching for "journalist," which have entirely different skill requirements. Zhang Wenjing, Vice President of Engineering at LinkedIn, emphasized that this improvement aims to better understand user needs and fundamentally change the way people find the most suitable jobs.

AI robot interviews and negotiations

"In the past, when we used keywords, we basically looked at the keywords and tried to find exact matches. Sometimes a job description might say 'journalist,' but the candidate may not actually be one; we would still retrieve such information, which isn't ideal for the candidate," explained Zhang Wenjing.

Now, LinkedIn has improved its ability to understand user queries, allowing users to search with richer keywords. For example, users no longer need to simply search for "software engineer"; they can input "find recent software engineering positions released in Silicon Valley," and the system will better understand and match relevant content.

To achieve this goal, LinkedIn has thoroughly reformed its search functionality's understanding capability. Zhang Wenjing stated that the entire process involves three stages: first, understanding the user query; second, retrieving the correct information from the vast job database; third, precise ranking to present the most relevant jobs at the forefront.

Previously, LinkedIn relied on fixed, taxonomy-based methods, ranking models, and some older LLMs lacking deep semantic understanding capabilities. Now, the company has shifted to a more modern, finely tuned large language model (LLM) to enhance its platform's natural language processing (NLP) capabilities.

Considering the high computational cost of LLMs, LinkedIn adopted a data distillation approach to reduce costs. They divided the application of LLMs into two steps: one for data and information retrieval, and another for sorting results. By using a teacher model to rank queries and jobs, LinkedIn successfully coordinated the retrieval model and ranking model and streamlined its job search system stages from the previous nine to a more concise process.

In addition, LinkedIn also developed a query engine to generate customized job suggestions for users.

LinkedIn is not the only company recognizing the potential of enterprise search based on LLMs. Google previously predicted that 2025 would be a year when enterprise search becomes even stronger thanks to advanced models. Models like Cohere's Rerank3.5 help break down language silos within enterprises. OpenAI, Google, and Anthropic's various "deep research" products indicate growing demand among organizations for agents to access and analyze internal data sources.

In the past year, LinkedIn has successively introduced several AI-powered features. In October this year, the company launched an AI assistant aimed at helping recruiters more effectively find the best candidates, further solidifying its leadership position in the professional networking field.