5 results about "Enterprise integration" patented technology

The invention provides a LangGraph-based multi-agent commercial plan intelligent analysis system, which comprises a retrieval agent module for realizing dynamic retrieval by adopting a ReAct mode; the analysis agent module supports multi-dimensional parallel analysis; the knowledge graph enhancement module is used for constructing a four-dimensional domain knowledge graph; the expert collaboration module is used for realizing real and virtual expert mixed analysis; the streaming output module is used for providing real-time interaction based on an SSE protocol; and the enterprise integration module is seamlessly connected with the existing system. According to the method, the problems of low information retrieval efficiency, serious expert dependence and non-uniform analysis standards in traditional business plan (BP) analysis are solved, and the business plan (BP) processing efficiency and analysis quality of an industrial research institute are remarkably improved.

Systems and methods for constructing an access-controlled idea graph from heterogeneous inputs. The system ingests user submissions and / or parses enterprise documents to extract candidate ideas, computes similarity between ideas, and links or merges related ideas to form a versioned evolution graph. For any querying user, the system evaluates access privileges against source-document permissions to derive a user-specific permutation of the idea graph. In response to a query, the system synthesizes a comprehensive disclosure from the accessible subgraph and outputs a structured report, optionally using generative AI to produce summaries, problem / solution statements, prior-art indications, and predicted classifications. Portfolio tools permit one-click generation of disclosure forms from selected versions and provide analytics such as uniqueness scores. Training and inference components implement the extraction, linking, merging, and synthesis workflows. The architecture supports enterprise integration, version control across permutations of an idea and role-based visibility (e.g., portfolio manager able to view merged, organization-wide version).

This invention provides a method for predicting corporate credit risk based on integrated graph neural networks. It pioneers a dual-dimensional (transaction-guarantee) integrated isomorphic graph construction system for enterprises, enabling simultaneous modeling of supply chain operational risks and guarantee-related risks. It constructs an enhanced three-layer graph attention network (GAT) model adapted to corporate credit risk prediction, achieving accurate extraction and effective propagation of risk features. This invention proposes a time-series verification mechanism for overlapping core enterprises across years, solving the data leakage problem of static network training and improving the model's generalization ability in real-world business scenarios. It designs an adaptive class-imbalanced weighted loss function and an automatic hyperparameter tuning framework, perfectly adapting to the small sample size and strong imbalance characteristics of corporate credit risk samples. This invention possesses strong versatility and scalability, applicable not only to the new energy vehicle industry chain but also seamlessly adaptable to corporate credit risk prediction scenarios across all industries, including manufacturing, retail, and commodity trading, demonstrating high industrial application value.

The invention relates to the technical field of information systems, and provides a multi-information system navigation proxy method and system based on a large language model, and the method comprises the steps: obtaining a user operation instruction, an API interface document, a page URL and behavior description of each information system, and classifying each information system into a query type and an interactive type through the large language model; if the information system is a query type, reconstructing the information system into an enterprise integration system, and adding metadata information into a metadata knowledge base of the enterprise integration system; if yes, an interaction start page and function description are added to a metadata knowledge base of the enterprise integration system; and in response to the user request, identifying a user intention, performing matching in the metadata knowledge base of the enterprise integration system based on the user intention, and sending the ID of the information system which is successfully matched and has the user permission to the http request generation agent. And smooth and seamless page jump experience under the multi-information system is realized.

Systems and methods for constructing an access-controlled idea graph from heterogeneous inputs. The system ingests user submissions and / or parses enterprise documents to extract candidate ideas, computes similarity between ideas, and links or merges related ideas to form a versioned evolution graph. For any querying user, the system evaluates access privileges against source-document permissions to derive a user-specific permutation of the idea graph. In response to a query, the system synthesizes a comprehensive disclosure from the accessible subgraph and outputs a structured report, optionally using generative AI to produce summaries, problem / solution statements, prior-art indications, and predicted classifications. Portfolio tools permit one-click generation of disclosure forms from selected versions and provide analytics such as uniqueness scores. Training and inference components implement the extraction, linking, merging, and synthesis workflows. The architecture supports enterprise integration, version control across permutations of an idea and role-based visibility (e.g., portfolio manager able to view merged, organization-wide version).