Transforming news search with RAG
%201.svg)
About
Korea Press Foundation
The Korea Press Foundation (KPF) is a Korean government organization dedicated to advancing journalism and media innovation in South Korea. It supports the press industry by providing research, training, and technological solutions to address evolving challenges in the digital era. Through initiatives like BIG KINDS, its flagship news database and search platform, the Korea Press Foundation continues to pioneer efforts in leveraging technology to empower journalists and improve information dissemination nationwide.
Problem
News platforms face significant challenges when it comes to generating accurate responses to natural language queries. Traditional news search engines rely heavily on keyword-based searches, which are ineffective at handling user queries expressed in everyday language. Additionally, hallucinations in Large Language Models (LLMs) often result in erroneous or irrelevant answers, creating a barrier to providing reliable information. With a constant influx of articles published in real-time, there was a pressing need for a system that could deliver precise, up-to-date answers without compromising on accuracy.
Technical challenges
- User experience: Keyword-based systems fail to meet the expectations of users seeking intuitive and natural interactions in news searches.
- Hallucination and accuracy: Generative AI models often generate plausible but incorrect information, making them unreliable for critical applications like news search.
- Data volume and real-time updates: The vast number of articles published daily necessitates a system that can efficiently process, index, and retrieve information without delays.
Solution
To tackle these challenges, Upstage developed a cutting-edge Retrieval-Augmented Generation (RAG) system in collaboration with the Korea Press Foundation. This system leverages advanced AI capabilities to enhance natural language query handling, providing accurate, context-aware responses based on real-time information. By using approximately 82 million articles provided by the Korea Press Foundation, Upstage built the BIG KINDS AI service, which integrates key functionalities such as real-time data updates, natural language question-based search, and highly stable performance suitable for public institutions.
The BIG KINDS AI system utilizes a two-stage retrieval process to optimize resource use and ensure the highest quality search results. The query engine first processes user queries to discern search intent and extract core queries, followed by the search engine which generates candidate responses and ranking them for relevance. Upstage also employed comprehensive prompt engineering to optimize the system’s performance for specific use cases, ensuring it met the precise needs of the Korea Press Foundation.
Additionally, Upstage leveraged AWS resources, including EC2 for computing power and S3 for secure data storage, to build a scalable and efficient infrastructure.
Results
BIG KINDS AI achieved a quality evaluation score of 86, surpassing the benchmark of 80 points set by the client. The Korea Press Foundation also rated its satisfaction at 92.2 points, reflecting Upstage’s exceptional performance.
Key outcomes include:
- Enhanced user experience with intuitive, natural language-based search.
- Increased article search traffic, validating the effectiveness of the new system.
- Reliable and secure infrastructure capable of handling real-time data processing.
By tailoring its RAG system to meet the unique demands of public institutions, Upstage has demonstrated how advanced AI can revolutionize the news search experience, setting a new standard for accuracy, efficiency, and user satisfaction.
Business impact
Why Upstage
Upstage’s expertise in RAG systems and tailored solutions sets it apart from generic providers. For BIG KINDS AI, Upstage demonstrated its capabilities in:
- Specialized prompt engineering: Customized prompts were designed to improve tone, classify intent, and ensure guardrails for safe responses.
- Custom search: Models were fine-tuned for factors like recency and accuracy to meet client-specific needs.
- Iterative feedback loop: A rigorous evaluation system involving both client and in-house teams ensured continual improvements and reliable deployment.
About
Korea Press Foundation
The Korea Press Foundation (KPF) is a Korean government organization dedicated to advancing journalism and media innovation in South Korea. It supports the press industry by providing research, training, and technological solutions to address evolving challenges in the digital era. Through initiatives like BIG KINDS, its flagship news database and search platform, the Korea Press Foundation continues to pioneer efforts in leveraging technology to empower journalists and improve information dissemination nationwide.
Problem
News platforms face significant challenges when it comes to generating accurate responses to natural language queries. Traditional news search engines rely heavily on keyword-based searches, which are ineffective at handling user queries expressed in everyday language. Additionally, hallucinations in Large Language Models (LLMs) often result in erroneous or irrelevant answers, creating a barrier to providing reliable information. With a constant influx of articles published in real-time, there was a pressing need for a system that could deliver precise, up-to-date answers without compromising on accuracy.
Technical challenges
- User experience: Keyword-based systems fail to meet the expectations of users seeking intuitive and natural interactions in news searches.
- Hallucination and accuracy: Generative AI models often generate plausible but incorrect information, making them unreliable for critical applications like news search.
- Data volume and real-time updates: The vast number of articles published daily necessitates a system that can efficiently process, index, and retrieve information without delays.
Solution
To tackle these challenges, Upstage developed a cutting-edge Retrieval-Augmented Generation (RAG) system in collaboration with the Korea Press Foundation. This system leverages advanced AI capabilities to enhance natural language query handling, providing accurate, context-aware responses based on real-time information. By using approximately 82 million articles provided by the Korea Press Foundation, Upstage built the BIG KINDS AI service, which integrates key functionalities such as real-time data updates, natural language question-based search, and highly stable performance suitable for public institutions.
The BIG KINDS AI system utilizes a two-stage retrieval process to optimize resource use and ensure the highest quality search results. The query engine first processes user queries to discern search intent and extract core queries, followed by the search engine which generates candidate responses and ranking them for relevance. Upstage also employed comprehensive prompt engineering to optimize the system’s performance for specific use cases, ensuring it met the precise needs of the Korea Press Foundation.
Additionally, Upstage leveraged AWS resources, including EC2 for computing power and S3 for secure data storage, to build a scalable and efficient infrastructure.
Results
BIG KINDS AI achieved a quality evaluation score of 86, surpassing the benchmark of 80 points set by the client. The Korea Press Foundation also rated its satisfaction at 92.2 points, reflecting Upstage’s exceptional performance.
Key outcomes include:
- Enhanced user experience with intuitive, natural language-based search.
- Increased article search traffic, validating the effectiveness of the new system.
- Reliable and secure infrastructure capable of handling real-time data processing.
By tailoring its RAG system to meet the unique demands of public institutions, Upstage has demonstrated how advanced AI can revolutionize the news search experience, setting a new standard for accuracy, efficiency, and user satisfaction.
Business impact
Why Upstage
Upstage’s expertise in RAG systems and tailored solutions sets it apart from generic providers. For BIG KINDS AI, Upstage demonstrated its capabilities in:
- Specialized prompt engineering: Customized prompts were designed to improve tone, classify intent, and ensure guardrails for safe responses.
- Custom search: Models were fine-tuned for factors like recency and accuracy to meet client-specific needs.
- Iterative feedback loop: A rigorous evaluation system involving both client and in-house teams ensured continual improvements and reliable deployment.
Explore more success stories
Explore how industry leaders have achieved success and transformed their workflows with Upstage.
Real-time life insurance claims digitization
Tax administration made efficient with Upstage Document Parse
Tax work is shifting toward digital transformation and automation. To keep up with these trends, the CEO of a Korean tax firm created a program powered by Upstage Document Parse to boost efficiency. Advanced technologies like Document AI are now essential for handling complex tasks, such as inheritance tax cases, which involve processing large volumes of documents.
Automating product price comparisons with a purpose-trained LLM
Upstage collaborated with ConnectWave to develop a domain-specific large language model (LLM) tailored for e-commerce tasks, addressing inefficiencies in manual product data extraction and organization. The customized LLM, built using Upstage’s advanced mixture-of-experts architecture, automated the extraction of attribute-value pairs and streamlined product catalog management. By significantly reducing manual workloads and improving data accuracy, the solution enhanced operational efficiency and user experience. Outperforming general-purpose LLMs like GPT-3.5, Upstage’s tailored model demonstrated the transformative potential of private LLMs in revolutionizing e-commerce operations.
Building Tomorrow’s Solutions Today
Talk to AI expert to find the best solution for your business.
