Hallucination Governance and Personal IP Construction: A Study on the Collaborative Application of AI Agents and Knowledge Bases in Media Content Production

Xiao Yong¹, Yue Qi², Wang Jiaqi¹
(1. China Business News, Chengdu, Sichuan 610017; 2. Chengdu Meijing New Vision Technology Co., Ltd., Chengdu, Sichuan 610017)

Abstract

[Purpose] Large language models have demonstrated significant application value in the news domain, yet they exhibit "hallucination" phenomena—such as factual errors and logical inconsistencies—during text generation, posing threats to journalistic authenticity and media credibility. [Method] This paper analyzes typical cases and root causes of large model hallucinations, explaining how the collaborative mechanism between AI agents and knowledge bases can mitigate these issues. Using the case of China Business News' construction of over 20 "super individual" IPs, it elaborates on the application value of agents and knowledge bases in media contexts. [Results] Through knowledge integration, context management, and multi-model collaboration, agents can effectively supervise and correct large model outputs, enhancing the accuracy, rationality, and timeliness of generated content. [Conclusion] Practice demonstrates that the agent-knowledge base collaborative mechanism not only strengthens the reliability of large model outputs but also creates a new paradigm for personal IP development, offering fresh insights for media intelligence transformation.

Keywords: Large Language Models; Hallucination; AI Agent; Knowledge Base; Media Application

1. Manifestations and Harms of Large Model Hallucinations in Media

Large language models have gradually permeated the news media sector, demonstrating tremendous potential in news summarization, interview outline generation, script creation, and massive data processing. However, alongside these benefits lurks the hallucination problem—where models generate plausible yet false or erroneous information. This phenomenon presents a severe challenge for news organizations.

1.1 Typical Manifestations of Hallucination

Hallucinations in news production take various forms and are often highly deceptive. Based on three "Large Model Evaluation Reports" published by China Business News in June, September, and November 2024, combined with practical newsroom observations, we identify several typical manifestations.

1.1.1 Fabrication of Facts or Data

Large models generate text based on probabilistic statistics without direct perception or verification capabilities regarding the real world, leading them to fabricate non-existent news events, invent interview content, or falsify and manipulate data. For instance, in the November 2024 China Business News Annual Large Model Evaluation Report, the evaluation team found that multiple models added false information when writing news reports based on given materials, resulting in factual errors regarding time, location, and other details. In one case involving a Starbucks announcement about leadership restructuring in China—where "Liu Wenjuan was appointed CEO of Starbucks China effective September 30"—some models conflated "today" with "September 30," generating the erroneous statement: "Starting September 30, Starbucks China announced its leadership restructuring." In another interview report, a model inexplicably attributed the news event "BYD and other automakers requested suppliers to cut prices by 10%" to the interviewee, thereby fabricating interview content.

1.1.2 Logical Contradictions or Inconsistencies

When confronted with complex news events requiring logical reasoning, large models frequently exhibit contradictions and broken logical chains. If applied directly to news reporting, this can result in conflicting arguments within the same article or severe factual errors due to lack of rigorous logic in analysis and inference. For example, on December 18, 2024, Shenzhen Nanyue Property Appraisal Co., Ltd. was administratively penalized for providing certified data beyond its qualification scope. However, some models erroneously linked this incident to a July traffic accident involving a AITO M7 vehicle in Guangzhou, generating the false logical statement that "this matter is related to the AITO M7 traffic accident."

1.1.3 Violations of Common Sense or Ethics

Large model outputs often reflect inherent correlations in training data rather than the dynamic, rich, and comprehensive real world. The selection criteria for datasets and their core concept definitions affect generation results and may suffer from reliability issues. Additionally, while large models can mirror real-world complexities and explain intricate logical relationships, they simultaneously reflect and amplify ethical issues such as discrimination and bias present in reality, potentially impacting human dignity, freedom, and fairness. Discriminatory, biased, or ethically problematic content generated by large models, if published by media without verification, can trigger strong negative social impacts or conflicts. Most large AI models dominating the international market are developed by Western companies that may embed historical biases and cultural superiority into their training data based on Western cultural values.

1.2 Root Causes of Hallucination

The hallucination problem in large models is not accidental but stems from multiple underlying factors.

1.2.1 Data-Level Factors

Training large models requires massive corpora that inevitably contain biases, noise, or erroneous information. Imbalanced data distribution, as well as excessive or insufficient topic representation, gets "inherited" by models during training, leading to false or inaccurate outputs.

1.2.2 Model-Level Factors

Current Transformer-based generative models rely heavily on training corpora and lack interpretability. This makes theoretical and methodological research on controlling model safety critically important. Existing methods cannot guarantee the absence of unsafe content generation, nor does any theory definitively explain what methods can strictly ensure model safety and control. When models incorrectly apply knowledge or use wrong knowledge to produce fluent responses, the lack of corresponding verification mechanisms means that stronger language capabilities make it increasingly difficult for users to identify information accuracy risks.

1.2.3 Lack of Knowledge and Common Sense

Large models primarily rely on statistical patterns for text generation and struggle to effectively invoke external knowledge or common sense for fact-checking and judgment. Consequences include major factual errors in specialized domains such as finance, healthcare, and law, as well as outputs violating basic human ethics and scientific common sense.

1.3 Harms of Hallucination to News Communication

The proliferation of large model hallucinations in news media inevitably causes serious consequences. First, it damages journalistic authenticity and objectivity—the foundation of news—eroding media professionalism through false content and logical errors. Second, it undermines media credibility and authority, which are built through rigorous fact-checking and professional ethics; frequent hallucinations erode public trust. Third, it misleads public opinion and social cognition. As primary information channels, media outlets publishing erroneous content can easily guide public misunderstanding and even trigger social panic or conflict. Fourth, it amplifies disinformation and rumor propagation. The high generation capacity and deceptive nature of hallucinations enable rapid and widespread dissemination of false information online, exacerbating information environment chaos and threatening social stability and cybersecurity. AI-generated misinformation is more persuasive than human-created false content, and generative AI systems can recycle conspiracy theories and other misinformation found on the open web. According to NewsGuard, GPT-4 is more likely than its predecessor GPT-3.5 to spread misinformation, excelling at presenting false narratives more persuasively across various formats including news articles, Twitter feeds, health rumors, and well-known conspiracy theories.

2. AI Agents and Knowledge Bases: Effective Approaches to Mitigating Hallucination

In recent years, the technical approach based on AI agents and knowledge bases has been recognized as effective for mitigating large model hallucinations by providing external knowledge constraints and fact-checking mechanisms. This section explores their functional mechanisms, construction methods, and collaborative working models.

2.1 The Role of AI Agents in Mitigating Hallucination

An AI agent is a system or program that can autonomously perform tasks on behalf of users or other systems by designing workflows and utilizing available tools. Beyond natural language processing, agents possess broad capabilities including decision-making, problem-solving, environmental interaction, and action execution. They can be deployed across various applications to solve complex tasks ranging from software design and IT automation to code generation. The core of an agent is a large language model, but while traditional models generate responses constrained by their training data and knowledge limitations, agents can access up-to-date information, optimize workflows, and autonomously create subtasks to achieve complex objectives. This tool invocation can be realized without human intervention, expanding the possibilities for real-world AI applications.

2.2 The Role of Knowledge Bases in Mitigating Hallucination

In AI, a knowledge base is a centralized system for storing, organizing, and managing information using AI technologies to enhance user interaction. Traditional knowledge bases contain only content, whereas AI-driven systems go further by employing machine learning, natural language processing, and other AI technologies to understand user queries, retrieve relevant information, and improve their knowledge provision. AI knowledge bases not only store static articles or documents but also learn from user behavior and feedback, analyzing patterns in data, predicting what users might seek, and even suggesting new content to fill existing gaps, creating a more dynamic and personalized experience.

Different types of knowledge bases can be constructed based on knowledge categories:

• Factual Knowledge Bases: Store objective facts and data such as entity relationships and attribute values for fact-checking, including verifying whether entity relationships in news reports are correct.
• Common Sense Knowledge Bases: Store common-sense knowledge (e.g., "birds can fly," "water flows downhill") for detecting whether generated content violates common sense.
• Vertical Domain Knowledge Bases: Store specialized domain knowledge such as finance, medicine, and law. In financial media applications, vertical industry knowledge bases can be built containing news events, person relationships, background knowledge, etc., for hallucination detection in relevant fields.
• Multimodal Knowledge Bases: As information in various formats (images, video, audio) increases and multimodal large models develop, building multimodal knowledge bases becomes increasingly important for cross-modal hallucination detection.

2.3 Synergistic Mechanism Between Agents and Knowledge Bases

AI agents and knowledge bases have an inseparable relationship. By definition, possessing knowledge and decision-making capability are two fundamental characteristics of AI agents. Agents can only learn from their knowledge bases, with higher success rates for more standardized knowledge. Recent research shows that by learning a diverse knowledge repository, AI agents can query and learn more flexibly, and integrating external knowledge bases and databases can mitigate AI hallucination problems.

This collaboration manifests in several ways. First, agents provide authoritative data support for large language model outputs through knowledge base integration and verification mechanisms, fundamentally addressing hallucinations caused by outdated knowledge and unreliable data sources. In news communication, where information accuracy is paramount, traditional large models relying on static training data cannot meet real-time update requirements. By accessing dynamic knowledge bases, agents can integrate policy documents, industry data, and authoritative reports as the basis for content generation. For example, in financial news, agents can extract the latest economic data or policy details from databases in real-time and cross-verify them with model-generated content to ensure accuracy and consistency. Agents also support multi-source cross-validation, identifying potential errors through multi-dimensional comparison and marking them for editor verification, thereby reducing misinformation.

Second, agents' context management and multi-task chaining capabilities effectively mitigate hallucinations caused by logical errors and content incoherence when large language models handle complex news tasks. In feature reporting, journalists must integrate multiple perspectives and generate content covering background analysis, data interpretation, and trend forecasting, demanding high logical consistency. Agents deconstruct complex news issues into subtasks—such as generating event backgrounds, core viewpoints, and supporting data analysis sequentially—ensuring logical coherence throughout.

Third, agents' dynamic feedback and real-time update functions significantly improve timeliness, crucial for breaking news coverage. Traditional models suffer from training data lag, while agents can access real-time data interfaces, scrape latest information from news sources, government databases, or authoritative industry reports, and rapidly integrate it into generated content. For instance, when policy adjustments occur in the new energy vehicle industry, agents can generate draft reports containing the latest policy interpretations and market impact analyses within minutes, with clear markings for dynamically updated information to remind journalists for secondary verification.

Finally, agents' multi-model collaboration mechanism further enhances content generation accuracy and reliability, particularly in specialized domains like finance, healthcare, or technology. A single large language model may have limited knowledge coverage or insufficient reasoning ability, but agents can simultaneously invoke multiple models, compare outputs, and synthesize optimal results. In finance, agents can combine market data interpretations from different models with real-time economic indicators or industry analysis tools for cross-validation, generating more precise reports. This multi-model collaboration reduces hallucination risks from single-model knowledge limitations while providing comprehensive solutions for complex news tasks.

4. Empowering Practice: China Business News' Personal IP Development

AI agents and knowledge bases can significantly improve efficiency, optimize decision-making, and drive industrial transformation, becoming a "new gateway" for enterprises and individuals to harness generative AI capabilities. In 2024, major Chinese tech companies launched their agent platforms, intensifying competition in the AI agent space. ByteDance launched "Coze," a one-stop AI Bot development platform, on February 1, 2024, enabling no-code AI Bot generation with multiple plugins covering news reading and travel planning. Baidu introduced its "Wenxin Agent Platform" and "Wenxin Yiyan" products on April 6, aiming to lower development barriers. Zhipu AI launched AutoGLM in October 2024, an agent capable of understanding user intentions through voice commands and automatically completing tasks like ordering food or booking flights. Zhipu also offers robust knowledge base functionality, allowing users to upload up to 1,000 files (100 million characters) via its Qingyan platform, while its Qingliu platform supports unlimited storage and RAG retrieval enhancement. Alibaba launched its agent platform "Zhishi Bing" (later renamed "Baibaoxiang") on September 25, along with frameworks like AgentScope and Mobile-Agent. Tencent released its "AI Think Tank" tool for creating agents and managing knowledge bases. These platforms are accelerating the popularization of generative AI technology and driving intelligent development across vertical industries.

Amidst this digital media boom, personal IP value has become increasingly prominent as a new focal point of media competition. However, traditional content creation and operation models rely heavily on manual labor, suffering from high costs, low efficiency, and difficulty in scaling and refined operation, making them ill-adapted to the new media era. In 2024, China Business News transformed its personal IP development paradigm by building large model-based agents and knowledge bases.

On October 8, 2024, CBN launched over 20 financial media "super individuals," marking another significant step toward full-staff IP and media intelligence transformation. These personal IPs, specializing in more than 20 financial verticals including international politics, real estate, corporate affairs, investment, and photovoltaics, leverage agents, knowledge bases, and the "Swift AI Video" automatic short-video generation platform to complete the entire workflow from script creation to video production and multi-platform distribution, demonstrating the powerful efficacy of AI products built on large models.

4.1 Architectural Innovation: Construction and Operation Mechanism of CBN's Personal IP Agents

CBN's personal IP agent construction is based on large models' powerful language understanding and generation capabilities, supplemented by system-level prompts, knowledge bases, external tools, memory modules, and workflow coordination. This architecture's core lies in deconstructing and reconstructing content production and operation links, achieving process automation and intelligence through AI technology.

4.1.1 Multi-Module Collaborative Agent Architecture

The CBN personal IP agent architecture comprises several key modules:

• System-Level Prompts: Acting as the agent's "soul," these prompts define AI behavior patterns, role positioning, and task objectives. Through refined prompt engineering, they guide large models to simulate professional thinking logic and expression styles in specific domains and scenarios, ensuring output professionalism and consistency.
• Knowledge Base: Serving as the agent's "brain," the knowledge base stores structured and unstructured knowledge across financial verticals. Through deep learning on massive data, it provides solid knowledge support for content generation. In operation, agents parse creators' requests, precisely calling relevant background information, data, and documents from the knowledge base to ensure generated content's accuracy and depth.
• External Tools: To overcome limitations of mainstream large models, CBN's personal IP agents integrate rich external tools such as search engines, database APIs, and data analysis tools. Flexible tool invocation enables agents to obtain real-time information, process data, and execute specific operations, expanding capability boundaries and improving task execution efficiency and accuracy.
• Memory Module: This module endows agents with "memory," enabling storage and retrieval of historical interaction information including user preferences, past tasks, and feedback results. Based on this, agents can provide more precise and coherent services tailored to individual needs, achieving personalized content recommendations and interactions.
• Workflow: As the agent's "action guide," workflow defines task processes and execution logic across content production stages. Through pre-designed workflows, agents can automatically complete topic planning, material mining, manuscript writing, text review, publishing, and operation, achieving standardization and automation.

4.1.2 Operational Logic of Personal IP Agents

CBN's personal IP agents receive requests or tasks from journalists and editors as input. First, system-level prompts help understand task backgrounds and objectives. Then, knowledge base and external tool modules activate to provide relevant data, documents, and external capability support. The memory module reviews historical information for personalized service reference. Finally, based on preset workflow designs, agents conduct task planning and execution, continuously learning and optimizing through feedback mechanisms to improve performance and adaptability.

4.2 Practical Exploration: Agent Application in CBN's Personal IP Development

CBN extensively applies agent technology across more than 20 vertical domains including international politics, real estate, corporate affairs, investment, and photovoltaics. By building vertical knowledge base-based agents, CBN achieves content production and operation automation, intelligence, and personalization.

[FIGURE:1]

4.2.1 Deep Vertical Cultivation: Building Specialized Knowledge Bases

For each vertical industry, CBN constructs dedicated knowledge bases covering listed companies and enterprise directories, financial data and annual reports, selected research reports, in-depth industry analyses, news media coverage, brand and hot product information, third-party consulting data, and competitor analysis. These knowledge bases provide solid data foundations and professional support for agents' deep content creation in specific domains.

[FIGURE:2]

4.2.2 Refined Agent Capabilities: Multiple AI Assistants Working in Coordination

Based on different usage scenarios, CBN develops multiple AI assistants with specific functions, including persona Q&A assistants, topic-finding assistants, material-digging assistants, manuscript assistants, creative-idea assistants, and learning-optimization assistants. These AI assistants achieve automated execution of specific tasks through optimized prompts, small-scale knowledge bases, and search enhancement plugins. For example, persona Q&A assistants accurately answer user questions based on preset personal IP personas and knowledge bases, maintaining consistency and professionalism; topic-finding assistants recommend topics by combining industry hotspots, user interests, and knowledge base content; material-digging assistants efficiently collect materials from knowledge bases and the internet based on selected topics; manuscript assistants help draft, polish, and edit content, generating different styles for various platforms and audiences; creative-idea assistants provide creative ideas, punchlines, and headline suggestions to enhance content appeal; learning-optimization assistants analyze competitor content to learn strengths and successful experiences.

[FIGURE:3]

4.2.3 From Automation to Intelligence: Workflow Construction and Optimization

Initially, CBN built basic business workflows centered on prompt optimization, small-scale knowledge bases, and search enhancement plugins. Moving forward, CBN plans to enable mutual invocation among AI assistants, integrate large-scale knowledge bases, incorporate web crawler plugins, and connect to selected large models to build more sophisticated and complex business workflows, achieving a leap from automation to intelligence.

[FIGURE:4]

4.3 Future Outlook: Intelligent Middle Platform Empowering "Super Individuals"

CBN's practice demonstrates that the human-agent collaborative model enables every journalist and editor to become a "super individual." By building an intelligent middle platform integrating vertical content capabilities and AI application capabilities, each editorial staff member can be equipped with multiple AI assistants, forming a dedicated agent matrix. In this model, editorial staff primarily focus on strategic work such as account operation, commercial monetization, and matrix dissemination, while tactical tasks like topic planning, material mining, idea generation, text processing, knowledge accumulation, competitor learning, audio-visual broadcasting, vertical media asset management, and distribution are collaboratively completed by the agent matrix. This significantly improves editorial efficiency, enabling journalists to truly become "super individuals" with full-process content production and operation capabilities under the support of the intelligent middle platform.

4.4 Technical Path: Balancing Open-Source and Commercial Platforms

Agent construction requires stable and reliable technical platform support. CBN currently utilizes commercial platforms like Coze for rapid prototyping and testing while actively evaluating open-source platforms for long-term stable and controllable solutions. CBN's future technical path includes building business workflows supporting mutual AI assistant invocation, integrating large-scale knowledge bases, incorporating web crawler plugins, and connecting to domestic leading large models to continuously enhance agent capabilities and application scope.

5. Conclusion

This study focuses on large language model applications in the media domain, particularly the challenge of "hallucination" to news authenticity and credibility, proposing and demonstrating an agent-knowledge base-centered solution. Through case analysis of CBN's construction of an intelligent system for over 20 "super individual" IPs, we draw the following conclusions.

5.1 Building an Anti-Hallucination Intelligent System

The deep integration of agents and knowledge bases provides an effective technical path for addressing large model hallucinations. Knowledge bases inject facts and common sense into large models; external tools and real-time verification mechanisms enable models to perceive reality and correct biases; multi-model collaboration achieves complementary advantages; and dynamic updates ensure knowledge timeliness and continuous agent evolution. Together, these elements construct an "anti-hallucination" barrier through knowledge constraints, real-time verification, multi-model collaboration, and dynamic updates, ensuring generated content's authenticity, accuracy, and reliability.

5.2 Agents Driving Personal IP Paradigm Innovation

Agent technology not only solves the hallucination problem but also revolutionizes personal IP development paradigms. Through the synergy of system-level prompts, knowledge bases, external tools, memory modules, and workflows, agents achieve automation, intelligence, and personalization in content production and operation. A single individual can complete traditional IP full-process work, achieving a full-chain leap in content production efficiency and effectiveness, enabling personal IPs to evolve into "super individuals" under agent empowerment.

5.3 Professional Knowledge Bases Building Vertical Barriers

CBN's dedicated knowledge bases for over 20 vertical industries are critical for agent effectiveness. These knowledge bases cover not only enterprise information, financial data, in-depth reports, and news coverage but also integrate third-party data and competitor analysis, building profound vertical knowledge barriers. This provides solid data foundations and professional support for agents' deep content creation in specific domains, ensuring generated content's accuracy, professionalism, and authority.

5.4 AI Assistant Ecosystem Construction

CBN's development of AI assistant groups for different scenarios—through refined prompt engineering, small-scale knowledge bases, and search enhancement plugins—achieves automation of specific tasks. These human-designed assistants (persona Q&A, topic-finding, material-digging, manuscript, creative-idea, and learning-optimization assistants) not only perform their duties but also collaborate with each other, forming a highly efficient intelligent ecosystem. Future improvements through more sophisticated workflows enabling deep collaboration among AI assistants and large-scale knowledge bases will further enhance overall agent effectiveness.

5.5 Human-Agent Collaboration Driving Deep Media Intelligence Transformation

CBN's practice demonstrates that the human-agent collaborative model represents the inevitable trend of media intelligence transformation. By building an intelligent middle platform integrating vertical content and AI application capabilities, empowering every editorial staff member to become a "super individual" will reshape media production relations and organizational structures. This strategic layout not only enhances media institutions' overall competitiveness and influence but also leads the media industry toward an intelligent future. As AI technology evolves and large model applications deepen, agents and knowledge bases will undoubtedly play increasingly important roles in media and broader industries.

References

[1] Zhang Zheng, Liu Chenxu. Large Model Hallucination: Cognitive Risks and Co-Governance Possibilities in Human-Machine Communication[J]. Journal of Soochow University (Philosophy & Social Science Edition), 2024, 45(5).

[2] Tu Liangchuan. A Historical Materialist Examination of AI's "Life-Like" Technology Narrative—Further Philosophical Inquiry into AI Singularity Theory[J]. Academic Exchange, 2023(12): 5-16.

[3] Teng Yan, Wang Guoyu, Wang Yingchun. Ethics and Governance of General Models: Challenges and Countermeasures[J]. Bulletin of Chinese Academy of Sciences, 2022, 37(09): 1290-1298.

[4] Che Wanxiang, Dou Zhicheng, Feng Yansong, et al. Natural Language Processing in the Era of Large Models: Challenges, Opportunities, and Development[J]. Scientia Sinica (Informationis), 2023, 53(09): 1645-1687.

[5] Ouyang L, Wu J, Jiang X, et al. Training Language Models to Follow Instructions with Human Feedback. In: Proceedings of the Advances in Neural Information Processing Systems[C]. 2022: 27730-27744.

[6] Fang Xingdong, Zhong Xiangming. Rational Judgment and China's Response to the ChatGPT Revolution—How to Discern ChatGPT's Disruptive Transformation Logic and Future Trends[J]. Journal of Northwest Normal University (Social Sciences Edition), 2023, 60(4).

[7] Spitale, Giovanni, Biller-Andorno, Nicola & Germani, Federico. AI Model GPT-3 (Dis)informs Us Better Than Humans[J]. Science Advances, 2023, 9, 26.

[8] Hu Yong. AI-Driven Disinformation: Present and Future[J]. Nanjing Journal of Social Sciences, 2024(01): 96-109.

[9] Anna Gutowska. What are AI agents?[EB/OL]. (2024-07-03)[2025-01-12]. https://www.ibm.com/think/topics/ai-agents.

[10] Edwin Lisowski. AI Knowledge Base: A Comprehensive Guide 2024[EB/OL]. (2024-06-30)[2025-01-11]. https://medium.com/@elisowski/ai-knowledge-base-a-comprehensive-guide-2024-949233ea8e98.

[11] Wynn C. Stirling. Coordinated Intelligent Control via Epistemic Utility Theory[D/OL]. Brigham Young University BYU ScholarsArchive. 1993(1). https://scholarsarchive.byu.edu/cgi/viewcontent.cgi?article=1704&context=facpub.

[12] Martin Ihrig, Alan S. Abrahams. BREAKING NEW In_SIMULATING_KNOWLEDGE_MANAGEMENT_PROCESSES_SIMISPACE2.

[13] Zih-Yun Chiu, Yi-Lin Tuan, William Yang Wang, Michael C. Yip. Flexible Attention-Based Multi-Policy Fusion for Efficient Deep Reinforcement Learning[J/OL]. 2023(1). https://arxiv.org/abs/2210.03729.

[14] Zhiheng Xi, Wenxiang Chen, Xin Guo. https://arxiv.org/pdf/2309.07864v1.

[15] Yu Kaicheng. Next-Generation AI Internet Driven by AI Agent Technology[J]. Zhejiang Economy, 2024(06): 21-23.

[16] Liu Xuedong, Yue Qi. "AI + Personal IP": An Analysis of Innovative Communication Models for Financial Media[J]. News Front, 2024(16): 83-84.

Author Biographies:
Xiao Yong (1977—), male, Master of Economics, Associate Editor, Editorial Board Member of China Business News; Yue Qi (1990—), male, Bachelor of Journalism, Chief Product Officer of Meijing Technology; Wang Jiaqi (1989—), male, Master of English Language and Literature, Assistant Editor, Deputy Director of International News Department at China Business News.

(Responsible Editor: Li Jing)