Enterprise portal content intelligent analysis method, device and medium based on large model

Abstract

The present application discloses a method, device and medium for intelligent parsing of enterprise portal content based on a large model, which relates to the field of artificial intelligence technology. The method includes: performing multimodal content parsing on the metadata of published content to obtain structured semantic information; defining graph attributes through entity fusion to obtain a knowledge graph with initial edge weights; performing multi-factor edge weight optimization construction on the knowledge graph with initial edge weights to determine the enterprise portal content knowledge graph; constructing an enterprise portal content intelligent body, and collaboratively associating the enterprise portal content intelligent body with the enterprise portal content knowledge graph; determining the user semantic intelligent body through semantic jump path planning based on user interaction parameters; and dynamically detecting the graph structure of the enterprise portal content knowledge graph to obtain collaboratively updated enterprise portal content intelligent body update items. The present application solves the technical problems of poor semantic relevance of content display and insufficient interactive parsing capabilities of large models through the above methods.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device, and medium for intelligent analysis of enterprise portal content based on a large model. Background Art

[0002] As the core platform for publishing and sharing internal information, enterprise portal systems integrate critical information previously scattered across departments, systems, and filing cabinets into a unified, easily accessible platform. This facilitates cross-departmental collaboration and knowledge sharing by instantly distributing important notifications, updating process guides, and providing self-service resources. Knowledge graphs, as structured semantic networks, are widely used in various information systems and are suitable for modeling entity relationships between content. Large models, with their powerful semantic understanding and generation capabilities, are attracting significant attention and are increasingly being used.

[0003] On the one hand, existing enterprise portal content display methods are typically based on chronological order, column classification, manual tag management, or manual sorting. These methods lack effective semantic or business logic-based association mechanisms between displayed content. Due to the lack of intelligent semantic recommendations, existing enterprise portal content display methods are unable to make intelligent recommendations based on user behavior and content value. On the other hand, existing large-scale model interaction methods primarily focus on processing a single modality and lack the ability to uniformly analyze multimodal content within enterprise portals. While they can calculate node similarity, edge weights are often calculated using a single factor, lacking comprehensive consideration of multi-dimensional information such as user behavior, temporal factors, and business logic. Summary of the Invention

[0004] The embodiments of the present application provide a method, device and medium for intelligent parsing of enterprise portal content based on a large model, which solves the technical problems of poor semantic relevance of existing enterprise portal content display and insufficient interactive capabilities of large model parsing.

[0005] In the first aspect, an embodiment of the present application provides a method for intelligent parsing of enterprise portal content based on a large model, characterized in that the method includes: obtaining the published content metadata of the enterprise portal, and performing multimodal content parsing on the published content metadata to obtain structured semantic information; based on the structured semantic information, defining the graph attributes of entity fusion to obtain a knowledge graph with initial edge weights; performing multi-factor edge weight optimization construction on the knowledge graph with initial edge weights to determine the enterprise portal content knowledge graph; constructing an enterprise portal content intelligent body, and collaboratively associating the enterprise portal content intelligent body with the enterprise portal content knowledge graph to obtain a visual interactive display interface; obtaining user interaction parameters, and determining the user semantic intelligent body through semantic jump path planning based on the user interaction parameters; performing dynamic graph structure detection on the enterprise portal content knowledge graph to obtain collaboratively updated enterprise portal content intelligent body update items.

[0006] In one implementation of the present application, multimodal content parsing is performed on the published content metadata to obtain structured semantic information, specifically including: preliminary entity recognition is performed on the published content metadata to obtain content entity units; cross-modal semantic fusion is performed on the content entity units to obtain unified semantic content; based on the unified semantic content, candidate entities are determined through text entity recognition based on semantic classification; based on the candidate entities, event triples are extracted to obtain event summary fragments, and based on the event summary fragments, structured semantic information is obtained through semantic vector enhancement.

[0007] In one implementation of the present application, based on structured semantic information, a graph attribute definition of entity fusion is used to obtain a knowledge graph with initial edge weights, specifically including: mapping structured semantic information into nodes and edges of the knowledge graph; configuring node attributes of the nodes to obtain node parameters; wherein the node parameters include: node type, attribute fields; configuring edge attributes of the edges to determine the initial edge weights; based on the node parameters and the initial edge weights, a knowledge graph with initial edge weights is obtained through context fusion of semantic similarity.

[0008] In one implementation of the present application, a multi-factor edge weight optimization construction is performed on the knowledge graph of the initial edge weight to determine the enterprise portal content knowledge graph, specifically including: classifying the knowledge graph of the initial edge weight by edge weight factors to obtain edge weight association factors; based on the edge weight association factors, the main weights of the edge weight association factors are determined through weight coefficient constraints; according to the main weights, the enterprise portal content knowledge graph to be optimized is obtained through multi-factor weight fusion; the weight of the enterprise portal content knowledge graph to be optimized is dynamically optimized to determine the enterprise portal content knowledge graph.

[0009] In one implementation of the present application, the enterprise portal content agent is collaboratively associated with the enterprise portal content knowledge graph to obtain a visual interactive display interface, specifically including: performing user request logic configuration on the enterprise portal content agent to determine the agent processing logic; based on the agent processing logic, obtaining a visual graph structure relationship through front-end graph rendering integrated with the antv G6 component; according to the visual graph structure relationship and the enterprise portal content knowledge graph, obtaining a visual interactive display interface through graph node jump navigation.

[0010] In one implementation of the present application, a user semantic agent is determined based on user interaction parameters through semantic jump path planning, specifically including: constructing an initial user semantic agent, and based on user interaction parameters, determining a recommended jump path through graph path association; visualizing the recommended jump path in a chain structure to obtain a visualized recommended path node; and determining the user semantic agent based on the visualized recommended path node.

[0011] In one implementation of the present application, a dynamic graph structure detection is performed on the enterprise portal content knowledge graph to obtain collaboratively updated enterprise portal content intelligent body update items, specifically including: performing graph structure change detection on the enterprise portal content knowledge graph to determine the graph structure change status; based on the graph structure change status, obtaining an updated identification type by identifying the change type; wherein the updated identification type includes: newly added nodes, edge type changes, path evolution, and edge weight changes; according to the updated identification type, collaborative updates are performed through intelligent bodies to obtain collaboratively updated enterprise portal content intelligent body update items.

[0012] In one implementation of the present application, after dynamically detecting the graph structure of the enterprise portal content knowledge graph to obtain collaboratively updated enterprise portal content intelligent body update items, the method also includes: updating the corresponding intelligent body through the collaboratively updated enterprise portal content intelligent body update items, and based on the updated intelligent body, determining a traceable user question and answer mechanism through graph path tracing; according to the user question and answer mechanism, obtaining a main path driven summary card through the main path extraction algorithm.

[0013] In the second aspect, the embodiment of the present application also provides an enterprise portal content intelligent parsing device based on a large model, characterized in that the device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can: obtain the published content metadata of the enterprise portal, and perform multimodal content parsing on the published content metadata to obtain structured semantic information; based on the structured semantic information, define the graph attributes of the entity fusion to obtain a knowledge graph with initial edge weights; perform multi-factor edge weight optimization construction on the knowledge graph with initial edge weights to determine the enterprise portal content knowledge graph; construct an enterprise portal content intelligent body, and collaboratively associate the enterprise portal content intelligent body with the enterprise portal content knowledge graph to obtain a visual interactive display interface; obtain user interaction parameters, and determine the user semantic intelligent body through semantic jump path planning based on the user interaction parameters; perform graph structure dynamic detection on the enterprise portal content knowledge graph to obtain collaboratively updated enterprise portal content intelligent body update items.

[0014] On the third aspect, the embodiment of the present application also provides a non-volatile computer storage medium for intelligent parsing of enterprise portal content based on a large model, which stores computer executable instructions, and is characterized in that the computer executable instructions are set to: obtain the published content metadata of the enterprise portal, and perform multimodal content parsing on the published content metadata to obtain structured semantic information; based on the structured semantic information, define the graph attributes of the entity fusion to obtain a knowledge graph with initial edge weights; perform multi-factor edge weight optimization construction on the knowledge graph with initial edge weights to determine the enterprise portal content knowledge graph; construct an enterprise portal content intelligent body, and collaboratively associate the enterprise portal content intelligent body with the enterprise portal content knowledge graph to obtain a visual interactive display interface; obtain user interaction parameters, and determine the user semantic intelligent body through semantic jump path planning based on the user interaction parameters; perform dynamic graph structure detection on the enterprise portal content knowledge graph to obtain collaboratively updated enterprise portal content intelligent body update items.

[0015] The embodiments of the present application provide a method, device and medium for intelligent parsing of enterprise portal content based on a large model. Through multimodal content semantic analysis, multi-factor edge weight fusion and optimization of knowledge graphs, and visualization and dynamic update of dual-end intelligent agents and graph structures, the technical problems of poor semantic relevance of existing enterprise portal content display and insufficient interactive capabilities of large model parsing are solved, and structured associations, personalized semantic recommendations and instant intelligent question-and-answer capabilities between enterprise portal content are achieved, thereby improving the semantic relevance of content display and the interactive capabilities of large model parsing. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0017] Figure 1 A flow chart of a method for intelligent parsing of enterprise portal content based on a large model provided in an embodiment of the present application;

[0018] Figure 2 An architecture diagram of intelligent content parsing for an enterprise portal based on a large model provided in an embodiment of the present application;

[0019] Figure 3 An embodiment of the present application provides a schematic diagram of the internal structure of a large-model-based enterprise portal content intelligent parsing device. DETAILED DESCRIPTION

[0020] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0021] The embodiments of the present application provide a method, device and medium for intelligent parsing of enterprise portal content based on a large model. Through multimodal content semantic analysis, multi-factor edge weight fusion and optimization of knowledge graphs, and visualization and dynamic update of dual-end intelligent agents and graph structures, the technical problems of poor semantic relevance of existing enterprise portal content display and insufficient interactive capabilities of large model parsing are solved, and structured associations, personalized semantic recommendations and instant intelligent question-and-answer capabilities between enterprise portal content are achieved, thereby improving the semantic relevance of content display and the interactive capabilities of large model parsing.

[0022] The technical solutions proposed in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0023] Figure 1 This is a flow chart of a method for intelligent analysis of enterprise portal content based on a large model provided in an embodiment of the present application. Figure 1 As shown, the embodiment of the present application provides a method for intelligent parsing of enterprise portal content based on a large model, which specifically includes the following steps:

[0024] Step 101: Obtain the published content metadata of the enterprise portal, and perform multimodal content analysis on the published content metadata to obtain structured semantic information.

[0025] For example, in order to perform semantic analysis and knowledge unit extraction on multimodal content, this application performs multimodal content analysis on the metadata of the published content to obtain structured semantic information, realizes automatic extraction and pre-analysis of the metadata structure, and provides a data foundation for the construction of the knowledge graph.

[0026] Specifically, multimodal content parsing is performed on the metadata of the published content to obtain structured semantic information, including: preliminary entity recognition of the published content metadata to obtain content entity units; cross-modal semantic fusion of the content entity units to obtain unified semantic content; based on the unified semantic content, candidate entities are determined through text entity recognition through semantic classification; based on the candidate entities, event triples are extracted to obtain event summary fragments, and based on the event summary fragments, structured semantic information is obtained through semantic vector enhancement.

[0027] Figure 2 An embodiment of the present application provides an architecture diagram of intelligent content parsing for an enterprise portal based on a large model.

[0028] In one embodiment, when a content administrator uploads or publishes content in an enterprise portal system, the system automatically collects and structures its metadata. Preliminary entity recognition of the content metadata yields content entity units, including: basic content attributes (author name, department, publication time, update time, content ID, etc.); access control dimensions (user ID, organizational structure, position information (e.g., department head), and user group); content classification tags (automatically assigning primary tags based on the content's assigned category (e.g., rules and regulations, corporate culture); content structural information (e.g., number of paragraphs in the body, title outline, presence of attachments, and the type and number of attachments); and multimodal feature identification (automatically identifying multimodal elements contained in the content and their distribution). A metadata field automatic parsing module performs preliminary entity recognition and generates basic "content entity units," preparing the structure for subsequent deep semantic analysis.

[0029] Next, cross-modal semantic fusion is performed on the content entities to obtain unified semantic content. The target is then parsed through joint analysis of text and attachments. Text and review types primarily include images, PDF / PPT, audio and video, and text. By establishing semantic correspondences between different modalities, a unified semantic representation is generated to obtain unified semantic content.

[0030] Finally, target parsing requires semantic classification of the content. Based on the existing columns, the large model further categorizes them into categories such as "notifications, decisions, activities, summaries, and interpretations." Entity recognition identifies key entities within the text and attachments, building a list of candidate entities. Next, keyword and topic extraction is performed, extracting multi-granular keywords and subject terms. These are combined with TF-IDF and the large model summary results to form the core content markers. "Subject-action-object" structures, such as "Human Resources Department - Release - Department KPI Assessment System," "Department - Organization - Training Activities," and summary snippets are identified. All results are structured into a content knowledge unit table, providing raw data for graph construction.

[0031] Furthermore, the multimodal features are combined to generate a comprehensive semantic vector representation of the content for subsequent similarity calculation.

[0032] Step 102: Based on the structured semantic information, the graph attribute definition of entity fusion is used to obtain the knowledge graph with initial edge weights.

[0033] For example, in order to enable intelligent analysis of enterprise portal content, this application implements knowledge graph construction and entity fusion through the definition of graph attributes of entity fusion.

[0034] Specifically, based on structured semantic information, the graph attribute definition of entity fusion is used to obtain a knowledge graph with initial edge weights, including: mapping the structured semantic information into nodes and edges of the knowledge graph; configuring node attributes of the nodes to obtain node parameters; wherein the node parameters include: node type, attribute fields; configuring edge attributes of the edges to determine the initial edge weights; based on the node parameters and the initial edge weights, the knowledge graph with initial edge weights is obtained through context fusion of semantic similarity.

[0035] In one embodiment, all extracted structured semantic information (entities, relationships, categories, keywords, etc.) is mapped into nodes and edges in the knowledge graph.

[0036] Among them, content nodes include: news, notices, announcements, system documents, meeting minutes, etc.; entity nodes include: "R&D Department", "Zhang Mou", "AI Chip Project", "Construction", etc.; concept nodes include: abstract concepts such as themes, categories, and tags; multimodal nodes include: multimedia objects such as images, audio, and video; relationship edges include: "belong to", "mention", "issue", "participate", "cite", "influence", "support", etc. Each node initially contains basic attributes and semantic vector information obtained from the large model.

[0037] In actual deployment, Tugraph can be used to build an enterprise content knowledge graph, which includes the following modeling forms.

[0038] First, node type and attribute fields:

[0039] {

[0040] "id": "content_20250601_001",

[0041] "type": "Notification",

[0042] "properties": {

[0043] "title": "Notice on the evaluation criteria for a certain project",

[0044] "label": ["project", "policy", "R&D"],

[0045] "department": "Planning Department",

[0046] "timestamp": "2025-06-01T10:00:00"

[0047] },

[0048] "embeddings": [0.231, -0.872, ...]

[0049] }.

[0050] Second, entity node example:

[0051] {

[0052] "id": "entity_project_ai_chip",

[0053] "type": "project",

[0054] "properties": {

[0055] "name": "AI chip project",

[0056] "leader": "Zhang",

[0057] "status": "In progress"

[0058] }

[0059] }.

[0060] Then, edge attributes and structure are defined, supporting weight, confidence, source, time range (start_time, end_time), and directionality.

[0061] To avoid the appearance of a large number of redundant nodes or similar but unfused entities in the knowledge graph, a context-based disambiguation logic is set up. Through Embedding semantic similarity calculation, the semantic vectors of the current entity and the existing entities are compared, and then context path matching is performed; if the matching threshold is reached but there is still doubt, the merge will be postponed and only the "possibly consistent" edges will be opened, and the merge will be confirmed after subsequent user interaction reinforcement.

[0062] Finally, based on node parameters and initial edge weights, a knowledge graph with initial edge weights is generated through contextual fusion of semantic similarity. Each time new content is added, the system adds new nodes, updates existing entity attributes, or strengthens the weights of certain edges. Content nodes are automatically associated with their meta-attributes, and automatic graph versioning is supported.

[0063] Step 103: Perform multi-factor edge weight optimization construction on the knowledge graph of the initial edge weight to determine the enterprise portal content knowledge graph.

[0064] For example, in order to be able to adaptively optimize the edge weights of the variable weights of the enterprise portal content knowledge graph, the present application constructs a multi-factor edge weight optimization of the knowledge graph of the initial edge weights to determine the enterprise portal content knowledge graph, thereby achieving the evaluation and optimization of all edges of the enterprise portal content knowledge graph and improving the relevance of the content display semantics.

[0065] Specifically, a multi-factor edge weight optimization construction is performed on the knowledge graph of the initial edge weight to determine the enterprise portal content knowledge graph, including: edge weight factor classification of the knowledge graph of the initial edge weight to obtain edge weight association factors; based on the edge weight association factors, the main weights of the edge weight association factors are determined through weight coefficient constraints; according to the main weights, the enterprise portal content knowledge graph to be optimized is obtained through multi-factor weight fusion; the weight of the enterprise portal content knowledge graph to be optimized is dynamically optimized to determine the enterprise portal content knowledge graph.

[0066] In one embodiment, first, seven core factors that affect entity relationship weights are identified, and a factor classification system is established:

[0067] Semantic relevance factor ( ): Measures the semantic similarity and conceptual association strength between entities;

[0068] Co-occurrence statistic factor ( ): Analyze the frequency and pattern of common occurrence of entities in different contents;

[0069] User behavior factor ( ): Statistics on users’ interaction behaviors and attention to entity relationships;

[0070] Business logic factor ( ): Integrate into the enterprise-specific organizational structure and business rule weights;

[0071] Time decay factor ( ): Consider the impact of content publishing time on relationship importance;

[0072] Cross-modal correlation factor ( ): Processing the semantic association strength between different modal contents;

[0073] Contextual structure factor ( ): Analyze the structured context information of entities in the knowledge graph.

[0074] The analysis of sovereign weight is explained by the following formula.

[0075] (1)

[0076] in, is the semantic correlation factor, with a value range of [0, 1]. The larger the value, the more semantically related the two entities are. It is calculated through multi-level semantic analysis.

[0077] It is a co-occurrence statistical factor with a value range of [0, 1], reflecting the frequency of two entities appearing together in different contents and needs to be standardized;

[0078] The user behavior factor has a value range of [0, 1] and is a comprehensive score based on user clicks, queries, favorites, and other behaviors, including timeliness decay.

[0079] The business logic factor has a value range of [0, 1] and is a weight value calculated based on the enterprise organizational structure and predefined business rules. The time decay factor has a value range of [0, 1] and decreases as the content is published. A segmented decay model is used.

[0080] is the time decay factor, with a value range of [0, 1]. It decreases as the content is released longer, and adopts a segmented decay model.

[0081] is the cross-modal association factor, with a value range of [0, 1], which measures the strength of semantic association between different modal contents;

[0082] It is the context structure factor with a value range of [0, 1] and is calculated based on the structural position and neighbor relationship of the entity in the knowledge graph.

[0083] For the weight coefficient, is the semantic weight coefficient, which controls the influence of semantic similarity in the final weight. It usually takes a value in the range of [0.15, 0.35] and is suitable for scenarios that focus on content relevance.

[0084] is the co-occurrence weight coefficient, which controls the contribution of co-occurrence frequency in the final weight. It usually takes a value in the range of [0.10, 0.25] and is suitable for statistically driven association analysis;

[0085] The behavior weight coefficient controls the influence of user behavior in the weight calculation. It usually takes a value in the range of [0.20, 0.40]. The value is larger in scenarios with a high degree of personalization.

[0086] The business weight coefficient controls the role of business logic in weight distribution. It usually ranges from [0.15 to 0.30]. This value is higher in enterprise internal application scenarios.

[0087] The decay weight coefficient controls the influence of time decay. It is usually in the range of [0.05, 0.15]. In scenarios that focus on timeliness, this value is larger.

[0088] is the cross-modal weight coefficient, which controls the weight contribution of cross-modal association. It usually takes a value in the range of [0.05, 0.20]. This value is higher when the multimedia content is rich.

[0089] It is the context weight coefficient, which controls the influence of graph structure information. It usually takes a value in the range of [0.10, 0.25]. This value is larger in scenarios that focus on relationship discovery.

[0090] The constraints on the weight coefficients are explained by the following formula.

[0091] (2)

[0092] Among them, the time decay factor The coefficient before A negative value indicates a decreasing weight effect over time.

[0093] The weight normalization formula is standardized weight = (original weight - minimum value) / (maximum value - minimum value). In the standardization parameters, the original weight is the initial weight value calculated by the multi-factor fusion formula; the minimum value is the minimum value of all edge weights in the current graph, which is used to determine the normalization lower bound; the maximum value is the maximum value of all edge weights in the current graph, which is used to determine the normalization upper bound; the standardized weight is the edge weight value finally used in the knowledge graph, which is ensured to be within the interval [0, 1].

[0094] Furthermore, the calculation method of each factor is explained by the following formula.

[0095] Semantic relevance factor (also known as similarity factor) , multi-level semantic similarity fusion calculation is adopted, the formula is:

[0096] (3)

[0097] in, is the vocabulary similarity, based on the edit distance and stemming It is explained by the following formula.

[0098] (4)

[0099] in, is the importance weight of the content unit;

[0100] For embedding similarity, semantic similarity based on pre-trained word vectors;

[0101] Contextual similarity is based on the semantic representation of an entity in a specific context.

[0102] The co-occurrence statistical factor considers the co-occurrence pattern of entities in content of different granularities and is explained by the following formula.

[0103] (5)

[0104] in, For the entity The number of co-occurrences in a content unit;

[0105] For the Total number of entities in content units;

[0106] The importance weight of the content unit (based on reading volume, citation number, etc.).

[0107] The user behavior factor is obtained by integrating the weighted scores of multiple user behavior patterns and is explained by the following formula.

[0108] (6)

[0109] in, is the total number of behavior types;

[0110] For the The weight of each behavior type (click = 0.3, Q&A = 0.4, favorite = 0.2, share = 0.1);

[0111] For the The number of times the behavior occurs;

[0112] is the time attenuation coefficient, usually 0.01-0.05;

[0113] For the The time difference between this behavior and the current time (in days).

[0114] The business logic factor is based on the weight distribution of the enterprise organizational structure and business rules, and is obtained by multiplying the organizational hierarchy weight, business rule weight and authority level weight.

[0115] Organizational hierarchy weight , , The hierarchical distance of the entity in the organizational structure.

[0116] The business rule weight is obtained by looking up the predefined business rule table.

[0117] The permission level weight is obtained by dividing the minimum level by the maximum level among different levels.

[0118] The time decay factor uses a segmented time decay model to distinguish the life cycles of different types of content. The specific expression is as follows:

[0119] f_decay = {

[0120] 1.0, if age ≤ fresh_period

[0121] exp(-decay_rate × (age - fresh_period)), if fresh_period <age ≤valid_period

[0122] min_weight, if age>valid_period

[0123] }.

[0124] The time parameters of different content types are shown in Table 1.

[0125] Table 1 Time parameter table

[0126]

[0127] The cross-modal correlation factor is processed as follows for different modalities:

[0128] Text-image: Use CLIP model to calculate cross-modal embedding similarity;

[0129] Text-audio: based on semantic similarity after speech-to-text conversion;

[0130] Image-video: visual similarity based on keyframe extraction;

[0131] Document-Attachment: Semantic Similarity Based on Content Summary.

[0132] The context structure factor considers the structured context information of the entity in the knowledge graph and is explained by the following formula.

[0133] (7)

[0134] in, is the path weight, which is set to the inverse of the shortest path length;

[0135] is the structural similarity, which is related to the number of common neighbors;

[0136] is the neighbor overlap, which is calculated by Jaccard similarity.

[0137] Step 104: Build an enterprise portal content intelligent body, and collaboratively associate the enterprise portal content intelligent body with the enterprise portal content knowledge graph to obtain a visual interactive display interface.

[0138] For example, in order to improve user interactivity, the present application realizes visualization of the graph structure relationship between the current content and related entities by collaboratively associating the enterprise portal content intelligent body with the enterprise portal content knowledge graph.

[0139] Specifically, the enterprise portal content agent is collaboratively associated with the enterprise portal content knowledge graph to obtain a visual interactive display interface, including: configuring the user request logic of the enterprise portal content agent to determine the agent processing logic; based on the agent processing logic, obtaining a visual graph structure relationship through the front-end graph rendering integrated with the antv G6 component; according to the visual graph structure relationship and the enterprise portal content knowledge graph, obtaining a visual interactive display interface through graph node jump navigation.

[0140] In one embodiment, the antv G6 component is used to integrate front-end graph rendering to display the graph structure relationship between the current content and related entities. The agent actively triggers prompt cards based on the current content theme and user portrait.

[0141] After the user clicks the "recommended item" or "Q&A button", a natural language query is initiated. The system will construct a corresponding prompt and return intelligent results.

[0142] Step 105: Obtain user interaction parameters, and determine the user semantic agent through semantic jump path planning based on the user interaction parameters.

[0143] For example, in order to achieve personalized recommendations and semantic navigation, this application determines the user semantic agent through semantic jump path planning based on user interaction parameters, thereby improving the large model's ability to parse and interact.

[0144] Specifically, according to user interaction parameters, the user semantic agent is determined through semantic jump path planning, including: constructing an initial user semantic agent, and based on the user interaction parameters, determining the recommended jump path through graph path association; visualizing the chain structure of the recommended jump path to obtain visualized recommended path nodes; and determining the user semantic agent based on the visualized recommended path nodes.

[0145] In one embodiment, the user can click on any node in the graph to trigger a jump to the corresponding content details; path navigation in the graph is supported, that is, the recommended path for the user to jump from the current content to the target content is displayed.

[0146] To achieve personalized recommendations and semantic navigation, the system builds a user semantic agent, which includes the following structure:

[0147] Interaction path records: Track the user's jump path and click records in the graph;

[0148] Reading preference modeling: Count the distribution of keywords involved in historical access content;

[0149] Question content analysis: Analyze the semantic types of users’ historical Q&A questions;

[0150] The logical class for constructing user portraits is expressed as follows:

[0151] {

[0152] "role": "Technical R&D position",

[0153] "interests": ["chip", "trust innovation", "R&D tasks"],

[0154] "knowledge_level": "Intermediate",

[0155] "interaction_history": ["ContentA", "ContentB"],

[0156] "recent_focus": ["Standard Development", "Project Progress"]

[0157] }.

[0158] Based on the content the user is currently reading, the associated graph path, and the interest profile, the agent dynamically constructs a "recommended jump path". The recommended path is visualized in the form of a chain structure diagram, supporting user click jumps and feedback; the size of the node in the diagram represents relevance or popularity, and the path is accompanied by a recommendation.

[0159] Step 106: Dynamically detect the graph structure of the enterprise portal content knowledge graph to obtain collaboratively updated enterprise portal content agent update items.

[0160] For example, in order to ensure the continuous availability and timeliness of the system knowledge graph, this application performs dynamic detection of the graph structure of the enterprise portal content knowledge graph to obtain collaboratively updated enterprise portal content intelligent body update items, thereby realizing the dynamic update of the knowledge graph, solving the technical problems of poor semantic relevance of existing enterprise portal content display and insufficient interactive parsing capabilities of large models, and improving the robustness of the system.

[0161] Specifically, the graph structure of the enterprise portal content knowledge graph is dynamically detected to obtain collaboratively updated enterprise portal content intelligent body update items, including: performing graph structure change detection on the enterprise portal content knowledge graph to determine the graph structure change status; based on the graph structure change status, obtaining the update identification type by identifying the change type; wherein the update identification type includes: newly added nodes, edge type changes, path evolution, edge weight changes; according to the update identification type, obtaining the collaboratively updated enterprise portal content intelligent body update items through intelligent body collaborative update.

[0162] Furthermore, after dynamically detecting the graph structure of the enterprise portal content knowledge graph to obtain collaboratively updated enterprise portal content intelligent body update items, the method also includes: updating the corresponding intelligent body through the collaboratively updated enterprise portal content intelligent body update items, and based on the updated intelligent body, determining a traceable user question and answer mechanism through graph path tracing; according to the user question and answer mechanism, obtaining a main path driven summary card through the main path extraction algorithm.

[0163] In one embodiment, the intelligent update behavior of the content agent is triggered through regular graph structure analysis and change detection processes. The update identification types include: new nodes, edge type changes, path evolution, and edge weight changes.

[0164] Once a change in the graph structure is detected, the system will automatically trigger the content agent's update process. The update items include: summary reconstruction, recommendation path reconstruction, response strategy optimization, and interaction prompt adjustment.

[0165] Summary reconstruction: Based on newly added entities or relationships, the large model is called to generate an updated intelligent summary;

[0166] Recommended path reconstruction: re-planning the semantic jump chain based on the updated graph path;

[0167] Response strategy optimization: by adding policy background paths, the corresponding capabilities of intelligent agents are enhanced;

[0168] Interactive prompts are adjusted by automatically updating relevant recommendation cards and tag sets.

[0169] Each user's semantic agent records the content and entities they follow. When the followed object changes, the system automatically pushes an updated summary.

[0170] To ensure the credibility and explainability of intelligent question-answering, a knowledge graph reference path mechanism is introduced into the large-scale model's answering process. When constructing a prompt, the graph path of the current content is explicitly embedded. When the model generates an answer, it is guided by which nodes, edges, and relationships the answer references. Finally, the information source is automatically annotated in the answer content.

[0171] Furthermore, the returned results include a graph path JSON structure for front-end display. Based on PageRank and user behavior popularity, the most important content entities are selected, the most appropriate link structure is automatically extracted, and summary cards driven by the main path are output.

[0172] Example summary card structure:

[0173] {

[0174] "title": "About the content summary of 'Innovation Adaptation Progress'",

[0175] "Main Path": "Innovation notification to the Ministry of Planning to promote the plan in 2019 to AI chip adaptation",

[0176] "Key Node": ["Notice", "Department", "Policy", "Meeting"],

[0177] Recommended Reading: ["Xxxx Policy Guidance Points", "XXX Meeting Minutes"]

[0178] }.

[0179] The front-end of this application is based on Angular and integrates antv's G6 component to implement portal content display and graph components. The back-end is based on Spring Boot to build content services and graph services. Tugraph is used as the graph database, and combined with the use of a large model, through steps such as content parsing, graph construction, agent modeling, graph collaborative display and intelligent path guidance, to form structured associations between content, personalized semantic recommendations and intelligent question-answering capabilities.

[0180] The above is an embodiment of the method proposed in this application. Based on the same inventive concept, this application embodiment also provides an enterprise portal content intelligent analysis device based on a large model, whose structure is as follows: Figure 3 shown.

[0181] Figure 3 This is a schematic diagram of the internal structure of a large-scale model-based enterprise portal content intelligent analysis device provided in the embodiment of this application. Figure 3 As shown, the equipment includes:

[0182] at least one processor 301;

[0183] and, a memory 302 in communication with the at least one processor;

[0184] The memory 302 stores instructions that can be executed by at least one processor, and the instructions are executed by the at least one processor 301 to enable the at least one processor 301 to:

[0185] Obtain the published content metadata of the enterprise portal and perform multimodal content analysis on the published content metadata to obtain structured semantic information; based on the structured semantic information, define the graph attributes of entity fusion to obtain the knowledge graph with initial edge weights; perform multi-factor edge weight optimization construction on the knowledge graph with initial edge weights to determine the enterprise portal content knowledge graph; construct the enterprise portal content intelligent body and collaboratively associate the enterprise portal content intelligent body with the enterprise portal content knowledge graph to obtain a visual interactive display interface; obtain user interaction parameters and determine the user semantic intelligent body through semantic jump path planning based on the user interaction parameters; perform dynamic graph structure detection on the enterprise portal content knowledge graph to obtain collaboratively updated enterprise portal content intelligent body update items.

[0186] Some embodiments of the present application provide corresponding Figure 1 A non-volatile computer storage medium for intelligent analysis of enterprise portal content based on a large model stores computer executable instructions, wherein the computer executable instructions are set as follows:

[0187] Obtain the published content metadata of the enterprise portal and perform multimodal content analysis on the published content metadata to obtain structured semantic information; based on the structured semantic information, define the graph attributes of entity fusion to obtain the knowledge graph with initial edge weights; perform multi-factor edge weight optimization construction on the knowledge graph with initial edge weights to determine the enterprise portal content knowledge graph; construct the enterprise portal content intelligent body and collaboratively associate the enterprise portal content intelligent body with the enterprise portal content knowledge graph to obtain a visual interactive display interface; obtain user interaction parameters and determine the user semantic intelligent body through semantic jump path planning based on the user interaction parameters; perform dynamic graph structure detection on the enterprise portal content knowledge graph to obtain collaboratively updated enterprise portal content intelligent body update items.

Claims

1. A method for intelligent analysis of enterprise portal content based on a large model, characterized in that: The method comprises: Obtaining metadata of published content from the enterprise portal, and performing multimodal content analysis on the metadata of the published content to obtain structured semantic information; Based on the structured semantic information, a knowledge graph with initial edge weights is obtained through the definition of graph attributes of entity fusion; Performing multi-factor edge weight optimization construction on the knowledge graph of the initial edge weight to determine the enterprise portal content knowledge graph; Constructing an enterprise portal content agent, and collaboratively associating the enterprise portal content agent with the enterprise portal content knowledge graph to obtain a visual interactive display interface; Obtaining user interaction parameters, and determining the user semantic agent through semantic jump path planning based on the user interaction parameters; Dynamically detect the graph structure of the enterprise portal content knowledge graph to obtain collaboratively updated enterprise portal content agent update items; The knowledge graph of the initial edge weight is optimized and constructed with multiple factors to determine the knowledge graph of the enterprise portal content, specifically including: Performing edge weight factor classification on the knowledge graph of the initial edge weight to obtain an edge weight association factor; Based on the edge weight association factor, determining the main weight of the edge weight association factor through weight coefficient constraints; According to the master weight, a knowledge graph of the enterprise portal content to be optimized is obtained by integrating multiple factor weights; Dynamically optimizing the weight of the enterprise portal content knowledge graph to be optimized to determine the enterprise portal content knowledge graph; The enterprise portal content agent is collaboratively associated with the enterprise portal content knowledge graph to obtain a visual interactive display interface, specifically including: Performing user request logic configuration on the enterprise portal content agent to determine agent processing logic; Based on the agent processing logic, the front-end graph rendering integrated with antv G6 components is used to obtain a visual graph structure relationship. According to the visual graph structure relationship and the enterprise portal content knowledge graph, the visual interactive display interface is obtained by navigating through the graph nodes; According to the user interaction parameters, the user semantic agent is determined through semantic jump path planning, specifically including: Constructing an initial user semantic agent, and determining a recommended jump path based on the user interaction parameters through graph path association; Performing chain structure visualization processing on the recommended jump path to obtain a visualized recommended path node; The user semantic agent is determined according to the visualized recommendation path node.

2. The method for intelligent analysis of enterprise portal content based on a large model according to claim 1, characterized in that: Perform multimodal content parsing on the published content metadata to obtain structured semantic information, specifically including: Performing preliminary entity recognition on the published content metadata to obtain content entity units; Performing cross-modal semantic fusion on the content entity units to obtain unified semantic content; Based on the unified semantic content, determining candidate entities through text entity recognition based on semantic classification; According to the candidate entities, event triples are extracted to obtain event summary fragments, and according to the event summary fragments, semantic vector enhancement is performed to obtain the structured semantic information.

3. The method for intelligent analysis of enterprise portal content based on a large model according to claim 1, characterized in that: Based on the structured semantic information, the graph attributes of entity fusion are defined to obtain the knowledge graph with initial edge weights, which specifically includes: Mapping the structured semantic information into nodes and edges of the knowledge graph; Performing node attribute configuration on the node to obtain node parameters; wherein the node parameters include: node type and attribute fields; Performing edge attribute configuration on the edge to determine an initial edge weight; Based on the node parameters and the initial edge weights, a knowledge graph of the initial edge weights is obtained through context fusion of semantic similarity.

4. The method for intelligent analysis of enterprise portal content based on a large model according to claim 1, characterized in that: Dynamically detect the graph structure of the enterprise portal content knowledge graph to obtain collaboratively updated enterprise portal content agent update items, including: Performing graph structure change detection on the enterprise portal content knowledge graph to determine the graph structure change status; Based on the graph structure change state, the updated identification type is obtained by identifying the change type; wherein the updated identification type includes: newly added nodes, edge type changes, path evolution, and edge weight changes; According to the update identification type, the collaborative update is performed through the agent to obtain the collaboratively updated enterprise portal content agent update item.

5. The method for intelligent analysis of enterprise portal content based on a large model according to claim 1, characterized in that: After dynamically detecting the graph structure of the enterprise portal content knowledge graph to obtain collaboratively updated enterprise portal content agent update items, the method further includes: Update the corresponding agent through the collaboratively updated enterprise portal content agent update item, and based on the updated agent, determine the traceable user question and answer mechanism through graph path tracing; According to the user question-and-answer mechanism, a main path-driven summary card is obtained through a main path extraction algorithm.

6. An intelligent analysis device for enterprise portal content based on a large model, characterized in that: The device comprises: at least one processor; and, a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: Obtaining metadata of published content from the enterprise portal, and performing multimodal content analysis on the metadata of the published content to obtain structured semantic information; Based on the structured semantic information, a knowledge graph with initial edge weights is obtained through the definition of graph attributes of entity fusion; Performing multi-factor edge weight optimization construction on the knowledge graph of the initial edge weight to determine the enterprise portal content knowledge graph; Constructing an enterprise portal content agent, and collaboratively associating the enterprise portal content agent with the enterprise portal content knowledge graph to obtain a visual interactive display interface; Obtaining user interaction parameters, and determining the user semantic agent through semantic jump path planning based on the user interaction parameters; Dynamically detect the graph structure of the enterprise portal content knowledge graph to obtain collaboratively updated enterprise portal content agent update items; The knowledge graph of the initial edge weight is optimized and constructed with multiple factors to determine the knowledge graph of the enterprise portal content, specifically including: Performing edge weight factor classification on the knowledge graph of the initial edge weight to obtain an edge weight association factor; Based on the edge weight association factor, determining the main weight of the edge weight association factor through weight coefficient constraints; According to the master weight, a knowledge graph of the enterprise portal content to be optimized is obtained by integrating multiple factor weights; Dynamically optimizing the weight of the enterprise portal content knowledge graph to be optimized to determine the enterprise portal content knowledge graph; The enterprise portal content agent is collaboratively associated with the enterprise portal content knowledge graph to obtain a visual interactive display interface, specifically including: Performing user request logic configuration on the enterprise portal content agent to determine agent processing logic; Based on the agent processing logic, the front-end graph rendering integrated with antv G6 components is used to obtain a visual graph structure relationship. According to the visual graph structure relationship and the enterprise portal content knowledge graph, the visual interactive display interface is obtained by navigating through the graph nodes; According to the user interaction parameters, the user semantic agent is determined through semantic jump path planning, specifically including: Constructing an initial user semantic agent, and determining a recommended jump path based on the user interaction parameters through graph path association; Performing chain structure visualization processing on the recommended jump path to obtain a visualized recommended path node; The user semantic agent is determined according to the visualized recommendation path node.

7. A non-volatile computer storage medium for intelligent analysis of enterprise portal content based on a large model, storing computer-executable instructions, characterized in that: The computer executable instructions are configured to: Obtaining metadata of published content from the enterprise portal, and performing multimodal content analysis on the metadata of the published content to obtain structured semantic information; Based on the structured semantic information, a knowledge graph with initial edge weights is obtained through the definition of graph attributes of entity fusion; Performing multi-factor edge weight optimization construction on the knowledge graph of the initial edge weight to determine the enterprise portal content knowledge graph; Constructing an enterprise portal content agent, and collaboratively associating the enterprise portal content agent with the enterprise portal content knowledge graph to obtain a visual interactive display interface; Obtaining user interaction parameters, and determining the user semantic agent through semantic jump path planning based on the user interaction parameters; Dynamically detect the graph structure of the enterprise portal content knowledge graph to obtain collaboratively updated enterprise portal content agent update items; The knowledge graph of the initial edge weight is optimized and constructed with multiple factors to determine the knowledge graph of the enterprise portal content, specifically including: Performing edge weight factor classification on the knowledge graph of the initial edge weight to obtain an edge weight association factor; Based on the edge weight association factor, determining the main weight of the edge weight association factor through weight coefficient constraints; According to the master weight, a knowledge graph of the enterprise portal content to be optimized is obtained by integrating multiple factor weights; Dynamically optimizing the weight of the enterprise portal content knowledge graph to be optimized to determine the enterprise portal content knowledge graph; The enterprise portal content agent is collaboratively associated with the enterprise portal content knowledge graph to obtain a visual interactive display interface, specifically including: Performing user request logic configuration on the enterprise portal content agent to determine agent processing logic; Based on the agent processing logic, the front-end graph rendering integrated with antv G6 components is used to obtain a visual graph structure relationship. According to the visual graph structure relationship and the enterprise portal content knowledge graph, the visual interactive display interface is obtained by navigating through the graph nodes; According to the user interaction parameters, the user semantic agent is determined through semantic jump path planning, specifically including: Constructing an initial user semantic agent, and determining a recommended jump path based on the user interaction parameters through graph path association; Performing chain structure visualization processing on the recommended jump path to obtain a visualized recommended path node; The user semantic agent is determined according to the visualized recommendation path node.

Images