<p>，/PRNewswire/ -- 随着企业数字化转型的深入和精确，能够高效管理和利用海量数据集已成为核心竞争壁垒。面对多源异构数据整合的重大挑战以及在复杂商业场景中对高保真决策的强烈需求，用友网络于 2 月 24 日正式发布了大型本体模型（LOM）。该发布为企业提供了一个真正的 “数字大脑”，能够深入理解其运营并执行复杂的逻辑推理。</p>
<p><strong>一、技术突破：LOM 作为企业智能的新基础</strong></p>
<p>基于用友 BIP 企业人工智能本体代理，LOM 代表了从传统二维表格模型到基于知识图谱架构的根本范式转变。通过使用节点和边来捕捉实体及其关系，LOM 将孤立的企业数据转化为可计算、具备推理能力的 “实时连接”。它将企业知识从静态存储升级为动态、可执行的智能资产。</p>
<p>利用统一的构建 - 对齐 - 推理框架，LOM 无缝连接底层业务系统和数据与顶层本体应用。它实现了企业知识的全生命周期管理，从原始提取到高价值输出。在构建阶段，LOM 自动化多源本体构建，打破结构化数据库与非结构化文本、显性与隐性知识、动态实时与静态历史数据之间的障碍。通过强大的知识构建引擎，它执行实体 - 关系提取、链接预测、知识蒸馏和推理补全，最终形成全面的企业级数据架构。同时，以 BIP 标准本体作为核心锚点，构建模型节点的骨架结构。然后，通过高效的动态文本 - 本体对齐，确保持续的数据流在语义上保持一致，并与本体结构深度融合。</p>
<p>在复杂逻辑推理方面，LOM 表现尤为出色。它在异构企业数据上执行可靠的多跳推理。在涵盖 19 个不同图推理任务的严格基准测试中，我们的 4B 参数 LOM 取得了领先的表现，以 89.47% 的整体准确率排名第一，并在几个核心任务上达到了近乎完美的准确率（100%）。这无疑验证了我们架构的前沿有效性。</p>
<p><strong>二、现实世界影响：LOM 在各个维度上助力敏捷和精益企业管理</strong></p>
<p>任何技术的最终考验是其在现实世界中的应用。凭借强大的推理能力，LOM 深度优化了核心企业工作流程——采购、生产、销售和财务。它将原始计算能力转化为切实的商业动能，从智能决策到复杂系统分析，提供全栈赋能。</p>
<p>以下是大型本体模型如何在这四个关键领域赋能智能决策和复杂系统分析：</p>
<p><strong>采购：构建抗脆弱的供应链</strong></p>
<p>当核心供应商面临中断时，LOM 不仅仅看到局部故障；它立即在全球图中定位该节点，执行深度遍历 Tier-2 和 Tier-3 依赖关系，以广播预测性预警。在采购订单发出之前，系统自主验证前提条件——凭证、质量保证日志、预算限制——消除因信息缺失而导致的合规风险。对于关键材料，LOM 进行中心性和拓扑分析，以识别战略瓶颈供应商——那些 “单点故障”，使企业能够提前构建冗余和备份计划。</p>
<p><strong>生产：全栈可追溯性和动态优化</strong></p>
<p>当产品缺陷发生时，LOM 沿着物料清单（BOM）执行前驱节点搜索，从成品逆向追溯到确切的原材料批次，以精确定位故障点。在智能工厂中，它充当路由引擎，为自动化物料搬运设备计算最短路径，显著减少操作之间的死时间。通过实时监控工作站的任务积压，它识别产出瓶颈，使管理者能够动态分配资源，平滑生产流程。</p>
<p><strong>销售与市场：高精度客户洞察和资源分配</strong></p>
<p>利用历史互动日志和社交关系图，LOM 运行类似 PageRank 的中心性算法，以识别高影响力客户节点，将有限的销售带宽集中在产生最高投资回报率的地方。在自有渠道生态系统运营中，它使用从现有用户出发的共同邻居分析，快速映射和捕捉直接连接的潜在客户群体，从根本上降低客户获取成本。对于高价值账户，它构建完整的行为漏斗，揭示流失背后的隐性关系模式，以推动高度针对性的留存策略。</p>
<p><strong>财务与风险管理：深入监督和自主合规</strong></p>
<p>资金一旦投入，LOM 会识别接收账户的直接交易对手。如果连接的组件包含高风险实体，它会触发自主的硬性停止。对于复杂的企业股权结构，LOM 深入多层所有权，揭示最终受益所有人，为并购和投资决策提供清晰的风险拓扑。在三方匹配中，它从付款请求开始，向上遍历以验证所有前置文件——采购订单、收据、发票——确保通过自动对齐实现财务结算的完全透明。</p>
<p>从裸金属架构到实际部署，LOM 严格基于实际企业需求。其高效、轻量的 4B 参数模型设计破解了复杂企业级推理的难题，大幅降低了部署企业 AI 的门槛和成本。更重要的是，它在结构化数据库和非结构化文本知识之间建立了明确的桥梁，创造了一个不断演变的智能框架，使企业数据资产自我生成和自我优化。</p>
<p>展望未来，用友的 LOM 将不断推动技术前沿。我们将升级我们的强化学习策略，建立开放的评估基准，并应对复杂挑战。我们正在扩大 LOM 在复杂企业环境中的推理能力，使得每个公司都配备一个深思熟虑的 “脑”。</p>
<p>https://chinaxiv.org/abs/202601.00187</p>
<p>来源：用友</p>
<p><img src="https://imageproxy.pbkrs.com/https://rt.prnewswire.com/rt.gif/query-RGF0ZUlkPTIwMjYwMzAxJk5ld3NJdGVtSWQ9Q045ODQ1OSZUcmFuc21pc3Npb25fSWQ9MjAyNjAzMDEyMDA4UFJfTkVXU19VU1BSX19fX19DTjk4NDU5?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://rt.prnewswire.com/rt.gif/query-RGF0ZUlkPTIwMjYwMzAxJk5ld3NJdGVtSWQ9Q045ODQ1OSZUcmFuc21pc3Npb25fSWQ9MjAyNjAzMDEyMDA4UFJfTkVXU19VU1BSX19fX19DTjk4NDU5"/></p>

用友网络

<p>用友网络推出了大型本体模型（LOM），这是一种变革性的企业数字化转型工具。该模型从传统的数据管理转向基于知识图谱的架构，使得复杂数据集的处理更加高效。LOM 通过提供深度洞察和自动化流程，增强了在采购、生产、销售和财务等领域的决策能力。它在逻辑推理任务中表现出色，准确率达到 89.47%。LOM 使企业能够优化工作流程、管理风险并改善客户洞察，最终推动业务发展和运营效率</p>

<p>, /PRNewswire/ -- As enterprise digital transformation advances with increasing depth and precision, the ability to efficiently manage and harness massive datasets has become the core competitive barrier. Facing the significant challenges of integrating multi-source heterogeneous data and the intense demand for high-fidelity decision-making in complex business scenarios, Yonyou officially released the Large Ontology Model (LOM) on February 24th. This release provides enterprises with a true &#34;digital brain&#34; capable of deeply understanding their operations and executing complex logical reasoning.</p>
<div class="lb-trans"><p>，/PRNewswire/ -- 随着企业数字化转型的深入和精确，能够高效管理和利用海量数据集已成为核心竞争壁垒。面对多源异构数据整合的重大挑战以及在复杂商业场景中对高保真决策的强烈需求，用友网络于 2 月 24 日正式发布了大型本体模型（LOM）。该发布为企业提供了一个真正的 “数字大脑”，能够深入理解其运营并执行复杂的逻辑推理。</p>
</div><p><strong>I. Technological Breakthrough: LOM as the New Foundation for Enterprise Intelligence</strong></p>
<div class="lb-trans"><p><strong>一、技术突破：LOM 作为企业智能的新基础</strong></p>
</div><p>Grounded in the Yonyou BIP Enterprise AI Ontology Agent, the LOM represents a fundamental paradigm shift from conventional two-dimensional tabular models to knowledge graph-based architectures. By using nodes and edges to capture entities and their relationships, LOM transforms siloed enterprise data into computable, reason-capable &#34;live connections.&#34; It upgrades enterprise knowledge from static storage into dynamic, executable intelligent assets.</p>
<div class="lb-trans"><p>基于用友 BIP 企业人工智能本体代理，LOM 代表了从传统二维表格模型到基于知识图谱架构的根本范式转变。通过使用节点和边来捕捉实体及其关系，LOM 将孤立的企业数据转化为可计算、具备推理能力的 “实时连接”。它将企业知识从静态存储升级为动态、可执行的智能资产。</p>
</div><p>Leveraging a unified construct-align-reason framework, LOM seamlessly connects bottom-layer business systems and data with top-layer ontology applications. It enables the full-lifecycle management of enterprise knowledge, from raw extraction to high-value output. During the construction phase, LOM automates multi-source ontology construction, bridging the barriers between structured databases and unstructured text, explicit and implicit knowledge, and dynamic real-time versus static historical data. Driven by a robust knowledge construction engine, it performs entity-relation extraction, link prediction, knowledge distillation, and reasoning completion, culminating in a comprehensive enterprise-scale data architecture. Simultaneously, using the BIP standard ontology as the core anchor, it builds the skeletal structure of the model's nodes. Then, through highly efficient dynamic text-ontology alignment, it ensures continuous data streams remain semantically consistent and deeply fused with the ontology structure.</p>
<div class="lb-trans"><p>利用统一的构建 - 对齐 - 推理框架，LOM 无缝连接底层业务系统和数据与顶层本体应用。它实现了企业知识的全生命周期管理，从原始提取到高价值输出。在构建阶段，LOM 自动化多源本体构建，打破结构化数据库与非结构化文本、显性与隐性知识、动态实时与静态历史数据之间的障碍。通过强大的知识构建引擎，它执行实体 - 关系提取、链接预测、知识蒸馏和推理补全，最终形成全面的企业级数据架构。同时，以 BIP 标准本体作为核心锚点，构建模型节点的骨架结构。然后，通过高效的动态文本 - 本体对齐，确保持续的数据流在语义上保持一致，并与本体结构深度融合。</p>
</div><p>When it comes to complex logical reasoning, the LOM is exceptionally capable. It executes reliable, multi-hop reasoning over heterogeneous enterprise data. In rigorous benchmark testing covering 19 diverse graph reasoning tasks, our 4B-parameter LOM achieved state-of-the-art performance, taking the top rank with an overall accuracy of 89.47%, while hitting near-perfect accuracy (100%) on several core tasks. This unequivocally validates the cutting-edge effectiveness of our architecture.</p>
<div class="lb-trans"><p>在复杂逻辑推理方面，LOM 表现尤为出色。它在异构企业数据上执行可靠的多跳推理。在涵盖 19 个不同图推理任务的严格基准测试中，我们的 4B 参数 LOM 取得了领先的表现，以 89.47% 的整体准确率排名第一，并在几个核心任务上达到了近乎完美的准确率（100%）。这无疑验证了我们架构的前沿有效性。</p>
</div><p><strong>II. Real-World Impact: LOM Supercharges Agile and Lean Enterprise Management Across All Dimensions</strong></p>
<div class="lb-trans"><p><strong>二、现实世界影响：LOM 在各个维度上助力敏捷和精益企业管理</strong></p>
</div><p>The ultimate test of any technology is its real-world application. Armed with formidable reasoning capabilities, LOM is deeply optimized for core enterprise workflows—procurement, production, sales, and finance. It translates raw computational power into tangible business momentum, delivering full-stack empowerment from intelligent decision-making to complex system analysis.</p>
<div class="lb-trans"><p>任何技术的最终考验是其在现实世界中的应用。凭借强大的推理能力，LOM 深度优化了核心企业工作流程——采购、生产、销售和财务。它将原始计算能力转化为切实的商业动能，从智能决策到复杂系统分析，提供全栈赋能。</p>
</div><p>Here is how the Large Ontology Model empowers intelligent decision-making and complex system analysis across these four critical domains:</p>
<div class="lb-trans"><p>以下是大型本体模型如何在这四个关键领域赋能智能决策和复杂系统分析：</p>
</div><p><strong>Procurement: Engineering an Antifragile Supply Chain</strong></p>
<div class="lb-trans"><p><strong>采购：构建抗脆弱的供应链</strong></p>
</div><p>When a core supplier faces disruption, the LOM doesn't just see a localized failure; it instantly pinpoints that node within the global graph, executing a deep traversal across Tier-2 and Tier-3 dependencies to broadcast predictive early warnings. Before a purchase order is even issued, the system autonomously validates prerequisites—credentials, QA logs, budget constraints—eliminating compliance risks caused by missing information. For mission-critical materials, the LOM runs centrality and topology analysis to identify strategic bottleneck suppliers—those &#34;single points of failure&#34;—allowing enterprises to engineer redundancy and backup plans ahead of time.</p>
<div class="lb-trans"><p>当核心供应商面临中断时，LOM 不仅仅看到局部故障；它立即在全球图中定位该节点，执行深度遍历 Tier-2 和 Tier-3 依赖关系，以广播预测性预警。在采购订单发出之前，系统自主验证前提条件——凭证、质量保证日志、预算限制——消除因信息缺失而导致的合规风险。对于关键材料，LOM 进行中心性和拓扑分析，以识别战略瓶颈供应商——那些 “单点故障”，使企业能够提前构建冗余和备份计划。</p>
</div><p><strong>Production: Full-Stack Traceability and Dynamic Optimization</strong></p>
<div class="lb-trans"><p><strong>生产：全栈可追溯性和动态优化</strong></p>
</div><p>When a product defect occurs, the LOM performs predecessor node searches along the Bill of Materials (BOM), reverse-tracing from the finished product back to the exact raw material batch to precisely locate the point of failure. In the smart factory, it acts as a routing engine, computing the shortest path for automated material handling equipment, significantly reducing dead time between operations. By monitoring task backlogs across workstations in real-time, it identifies throughput bottlenecks, enabling managers to dynamically allocate resources and smooth out the production flow.</p>
<div class="lb-trans"><p>当产品缺陷发生时，LOM 沿着物料清单（BOM）执行前驱节点搜索，从成品逆向追溯到确切的原材料批次，以精确定位故障点。在智能工厂中，它充当路由引擎，为自动化物料搬运设备计算最短路径，显著减少操作之间的死时间。通过实时监控工作站的任务积压，它识别产出瓶颈，使管理者能够动态分配资源，平滑生产流程。</p>
</div><p><strong>Sales &amp; Marketing: High-Precision Customer Insight and Resource Allocation</strong></p>
<div class="lb-trans"><p><strong>销售与市场：高精度客户洞察和资源分配</strong></p>
</div><p>Leveraging historical interaction logs and social relation graphs, the LOM runs PageRank-style centrality algorithms to identify high-influence customer nodes, focusing your finite sales bandwidth where it generates the highest ROI. In owned channel ecosystems operations, it uses common neighbor analysis starting from existing users to rapidly map and capture directly connected prospect clusters, fundamentally driving down customer acquisition costs. For high-value accounts, it constructs a full-journey behavioral funnel, uncovering the hidden relational patterns behind churn to power highly targeted retention strategies.</p>
<div class="lb-trans"><p>利用历史互动日志和社交关系图，LOM 运行类似 PageRank 的中心性算法，以识别高影响力客户节点，将有限的销售带宽集中在产生最高投资回报率的地方。在自有渠道生态系统运营中，它使用从现有用户出发的共同邻居分析，快速映射和捕捉直接连接的潜在客户群体，从根本上降低客户获取成本。对于高价值账户，它构建完整的行为漏斗，揭示流失背后的隐性关系模式，以推动高度针对性的留存策略。</p>
</div><p><strong>Finance &amp; Risk Management: Penetrative Oversight and Autonomous Compliance</strong></p>
<div class="lb-trans"><p><strong>财务与风险管理：深入监督和自主合规</strong></p>
</div><p>The moment funds are deployed, the LOM identifies direct transactional counterparties of the receiving account. If the connected component contains high-risk entities, it triggers an autonomous hard stop. For labyrinthine corporate equity structures, the LOM penetrates multi-hop ownership layers to expose the ultimate beneficial owner, delivering a crystal-clear risk topology for M&amp;A and investment decisions. In three-way matching, it starts at the payment request and traverses upward to validate all predecessor documents —POs, receipts, invoices—ensuring complete transparency in financial settlements through automated alignment.</p>
<div class="lb-trans"><p>资金一旦投入，LOM 会识别接收账户的直接交易对手。如果连接的组件包含高风险实体，它会触发自主的硬性停止。对于复杂的企业股权结构，LOM 深入多层所有权，揭示最终受益所有人，为并购和投资决策提供清晰的风险拓扑。在三方匹配中，它从付款请求开始，向上遍历以验证所有前置文件——采购订单、收据、发票——确保通过自动对齐实现财务结算的完全透明。</p>
</div><p>From bare-metal architecture to real-world deployment, the LOM is rigorously grounded in actual enterprise needs. Its highly efficient, lightweight 4B-parameter model design cracks the code on complex enterprise-level reasoning, massively lowering the barrier and cost of deploying enterprise AI. More than that, it builds the definitive bridge between structured databases and unstructured textual knowledge, creating a continuously evolving intelligence framework where enterprise data assets self-generate and self-optimize.</p>
<div class="lb-trans"><p>从裸金属架构到实际部署，LOM 严格基于实际企业需求。其高效、轻量的 4B 参数模型设计破解了复杂企业级推理的难题，大幅降低了部署企业 AI 的门槛和成本。更重要的是，它在结构化数据库和非结构化文本知识之间建立了明确的桥梁，创造了一个不断演变的智能框架，使企业数据资产自我生成和自我优化。</p>
</div><p>Looking ahead, Yonyou's LOM will relentlessly push the technical frontier. We will upgrade our reinforcement learning strategies, build open evaluation benchmarks, and tackle complex challenges. We are scaling LOM's inference capabilities in complex enterprise environments so that <em>every</em> company is equipped with a deep-thinking &#34;brain.&#34;</p>
<div class="lb-trans"><p>展望未来，用友的 LOM 将不断推动技术前沿。我们将升级我们的强化学习策略，建立开放的评估基准，并应对复杂挑战。我们正在扩大 LOM 在复杂企业环境中的推理能力，使得每个公司都配备一个深思熟虑的 “脑”。</p>
</div><p>https://chinaxiv.org/abs/202601.00187</p>
<div class="lb-trans"><p>https://chinaxiv.org/abs/202601.00187</p>
</div><p>SOURCE Yonyou</p>
<div class="lb-trans"><p>来源：用友</p>
</div><p><img src="https://imageproxy.pbkrs.com/https://rt.prnewswire.com/rt.gif/query-RGF0ZUlkPTIwMjYwMzAxJk5ld3NJdGVtSWQ9Q045ODQ1OSZUcmFuc21pc3Npb25fSWQ9MjAyNjAzMDEyMDA4UFJfTkVXU19VU1BSX19fX19DTjk4NDU5" alt="" original-src="https://rt.prnewswire.com/rt.gif?NewsItemId=CN98459&amp;Transmission_Id=202603012008PR_NEWS_USPR_____CN98459&amp;DateId=20260301"/></p>

20:08 ET 用友网络发布大型本体模型（LOM）：为企业构建深度思考的数字核心