Enterprise Knowledge Search
The Paper wanted to use generative AI in news products, but needed a way to answer readers' questions without hallucinations while preserving the accuracy standards of professional journalism.
Reported impact
The Paper turned its reporting archive into a conversational news product that answers reader questions interactively while preserving trust through source-grounded generation.
The Paper used Baidu AI Cloud's Qianfan platform and ERNIE capabilities to build an AI Q&A assistant grounded in its own newsroom archives. The key design choice was retrieval-augmented generation rather than free-form generation: more than 27 million articles and roughly 35 billion characters of high-quality original reporting were turned into the retrieval corpus that the assistant uses to answer user questions. When readers ask about hot topics, the system analyzes the context of the request, finds related reporting and source material inside The Paper's content ecosystem, and then generates a personalized answer with supporting topic coverage and news provenance. This lets the newsroom extend its editorial assets into an interactive product without accepting the failure modes of an unguided model. Architecturally, Qianfan supplies the model and orchestration layer while The Paper's content base acts as the grounding and evidence layer, producing a reader-facing assistant tuned for trustworthy news explanation rather than generic chat.
Reader questions are analyzed by the AI assistant, which retrieves relevant reporting from The Paper's article archive through a RAG layer and returns grounded answers, related stories and topic context to the user interface.
The Paper used more than 27 million articles and nearly 35 billion characters of original news content to co-create the AI Q&A assistant (original: 基于澎湃新闻十年来沉淀的超2700万篇文章，近350亿字的海量原创优质新闻内容生态，百度智能云千帆平台提供文心大模型能力，共创“AI问答助手”功能).
The assistant can understand user context, answer real-time hotspot questions and provide related reports or source material (original: “AI问答助手”可以针对用户提问进行上下文理解和分析，更好地回答各类用户关切的新闻实时热点问题，并提供相关新闻专题报道或者舆情事件信源).
The main challenge was controlling hallucinations in professional news production (original: 专业新闻报道对内容严谨、准确度有极高要求，需要解决大模型内容生成存在的幻觉问题).
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
Baidu AI Cloud Sample May 15, 2025 CCTV.com CCTV.com needed a better way to surface the value of its huge video archive because title- and keyword-based retrieval could not expose enough scene-level meaning, leaving many useful videos hard to find. Resource retrieval efficiency improved 90%. Multimodal Media Retrieval · Real-Time Speech Recognition Baidu AI Cloud Sample May 15, 2025 Tianmai Juyuan Tianmai Juyuan needed to organize and search a massive broadcast-video archive quickly enough to support timely publishing, short-video creation and intelligent content retrieval. MCA analysis efficiency improved 3-5x over manual tagging. BOS Object Storage · Media Content Analysis (MCA) · Micro-Editing Platform 
Lenovo Sample Oct 13, 2026 Formula E Formula E wanted fan experiences that feel immediate, immersive, personalized, and participatory, but its broadcast and event experience was limited by inconsistent onboard video quality, information spread across… Developed in under 30 days. AI Fan Experience Guide · AI Video Optimization · AR broadcast graphics 