Enterprise technology evolution trend identification method and system based on large model reasoning

Abstract

The invention belongs to the field of artificial intelligence and enterprise strategic technology prediction, and particularly relates to an enterprise technology evolution trend identification method and system based on large model reasoning. The system is composed of six modules: a data acquisition and preprocessing module unifies structured multi-source data; the enterprise technical portrait module constructs a research and development capability and technical trajectory map; the task analysis module utilizes a large model to understand natural language requirements and generates a multi-layer task structure; the knowledge modeling module is fused with the industry ontology to construct a semantic map; the trend identification module predicts a technical evolution path through time sequence modeling and multi-round reasoning; and the feedback optimization module forms a closed-loop learning mechanism based on the user behavior fine tuning model. According to the method, the semantic understanding and reasoning capabilities of a large model are fused, the knowledge graph, task analysis, trend modeling and a feedback mechanism are combined, the problems of insufficient structuralization, prediction lag and the like are effectively solved, the accuracy and foresight of enterprise technology trend recognition are improved, and intelligent support is provided for enterprise innovative deployment.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and enterprise strategic analysis technology, and in particular relates to a method and system for identifying enterprise technology evolution trends that integrates the reasoning ability of Large Language Models (LLMs) and knowledge graph representation learning. It aims to build a closed-loop system for predicting technology trends with cross-context semantic understanding, trend evolution modeling and dynamic feedback optimization capabilities, so as to improve the ability of enterprises to identify future technology directions in complex environments. Background Technology

[0002] In today's increasingly competitive global technology environment, enterprises face challenges such as faster technological substitution, frequent cross-industry integration, and rising strategic uncertainty. Traditional methods for identifying technology trends mainly rely on expert consultation, the Delphi method, technology roadmap creation, and manual information retrieval. While these methods have some value in early-stage technology planning, they still have the following problems.

[0003] First, information processing capabilities are limited. Traditional methods struggle to quickly extract effective information from large-scale unstructured data (such as patent texts, academic papers, technical standards, and market information), and cannot integrate and analyze multi-dimensional information within the industrial and innovation chains. Second, there is a lack of real-time and forward-looking capabilities. Current mainstream trend analysis relies on lagging statistical data, often reacting only after technological trends have already emerged, making it difficult to provide enterprises with strategically significant predictive support. Third, expert subjectivity is strong. Due to reliance on expert experience for trend judgment, it is easily influenced by cognitive biases, making it difficult to form objective and comprehensive technical judgment criteria, especially in the context of cross-domain technology integration. Fourth, the methods are fragmented and lack systematicity. Existing technology forecasting tools are mostly used in isolation, failing to achieve a closed-loop process from enterprise needs analysis, capability profiling, trend forecasting to outcome matching, resulting in fragmented technical services and low collaborative efficiency.

[0004] Meanwhile, generative large language models have demonstrated powerful capabilities in natural language understanding, knowledge integration, and logical reasoning in recent years, and are gradually being applied to scenarios such as intelligent question answering, text generation, automatic summarization, and content recommendation. Large models possess the ability to automatically extract knowledge from massive amounts of text, identify semantic relationships, and perform logical deductions, making them the core of next-generation intelligent information processing engines. However, how to effectively transfer and integrate the capabilities of large models into enterprise strategic analysis and technology trend identification is still in the initial exploratory stage. Especially in key areas such as enterprise technology identification, industry trend prediction, and technology supply and demand matching, a mature, systematic, and intelligent analytical method has not yet been formed, leading to widespread problems such as low scientific research results conversion rates, poor industry-academia matching efficiency, and weak forward-looking technology layout.

[0005] Therefore, there is an urgent need to develop a method and system for identifying enterprise technology trends based on large-scale model reasoning capabilities and combined with advanced technologies such as knowledge graphs, semantic understanding, and multimodal data fusion. This would help enterprises discern certainty amidst uncertainty, enhance their strategic judgment and innovation deployment capabilities, and provide precise navigation for technology research and development and industrial layout. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for identifying enterprise technology evolution trends based on large model reasoning. This method integrates the powerful language understanding and semantic reasoning capabilities of large models, and combines key technologies such as multi-source heterogeneous data processing, enterprise knowledge graph construction, and time series modeling to systematically solve problems such as insufficient structure, prediction lag, and supply-demand mismatch in enterprise technology trend identification. This will comprehensively improve the level of intelligence of enterprises in R&D planning, strategic decision-making, and technology investment.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] The first aspect of this invention provides a method and system architecture for identifying enterprise technology evolution trends based on large-scale model reasoning, which mainly consists of the following six core modules:

[0009] Data acquisition and preprocessing module. Supports structured extraction and quadruple representation of 12 types of data sources (such as patents, policies, papers, market dynamics, etc.) to form standard input: Among them, h i t i For entities, r i For semantic relations, τ i For timestamps.

[0010] Enterprise Technology Profiling Module. Based on collected enterprise-level data, a multi-dimensional profiling system is constructed, including:

[0011] (1) Technical requirements profile (domain coverage, pain point description, key metrics, etc.)

[0012] (2) Profile of R&D capabilities (number of patent applications, output of scientific and technological achievements, and composition of technical personnel over the years)

[0013] (3) Profile of cooperative behavior (records of upstream and downstream collaboration in the industrial chain, project participation, etc.)

[0014] (4) Profile of technological evolution trajectory (major R&D paths, technology transformation nodes, etc.)

[0015] This module constructs a knowledge graph G = (V, E, R) using multidimensional enterprise data and introduces a GNN for embedding learning, learning the representation e. v ∈R d The dynamic update mechanism ensures

[0016] The requirements analysis and task decomposition module. It combines large language models (such as GPT-4 and Gemini) to perform structured translation of natural language requirements. Through prompt embedding and structured templates, it guides the model to output a hierarchical structure of requirements, automatically decomposing broadly expressed comprehensive objectives into:

[0017] (1) Primary task modules (such as "low temperature battery life optimization")

[0018] (2) Secondary technical breakdown points (such as "battery material selection", "thermal management solution improvement" etc.)

[0019] (3) Three-level key performance indicators (such as "low temperature charge and discharge rate", "heating response speed" etc.)

[0020] The breakdown output is expressed in the form of a technology tree, i.e., T = {T1, {T...} 1.1 ,…,T 1.k},{I 1.1 ,…,I 1.k}}, where T is the task and I is the metric, and it supports embedding knowledge graph node alignment, which can be linked with the enterprise knowledge graph.

[0021] Technical Knowledge Modeling Module. This module unifies the modeling of enterprise technical tasks, user profiles, and external knowledge to construct a large-scale enterprise technical knowledge graph capable of inference. It mainly includes:

[0022] (1) Entity relationship modeling: Establish core node relationships such as "enterprise-task-indicator-time";

[0023] (2) Semantic graph construction: Introduce industry ontology and standard classification to achieve heterogeneous information fusion;

[0024] (3) Graph Neural Network Embedding: Vectorize the knowledge structure to provide semantic input for trend recognition and recommendation.

[0025] This module constructs a multi-layered semantic graph that integrates enterprise knowledge and industry ontology. Nodes in the graph are represented as: g = (V, E) with V = {v} ent ,v task ,v kpi ,v time Embedding optimization is performed using the TransE or ComplEx method.

[0026] Trend Identification and Prediction Module. Integrating time series modeling, semantic comparative analysis, and multi-turn dialogue capabilities of large models, this module identifies discrepancies and intersections between a company's current technology strategy and industry development trends. This module includes:

[0027] (1) Hot Technology Trend Identification Submodule: Generates popularity evolution curves from dimensions such as policy hot words, research high-frequency words, and investment keywords;

[0028] (2) Sub-module for predicting the evolution path of alternative technologies: Analyzes the bottlenecks of current technologies and the probability of the emergence of alternative solutions, and outputs possible transition paths;

[0029] (3) Competitive Landscape Deduction Submodule: Monitor the technological dynamics and strategic evolution of domestic and foreign competitors, conduct "dialogue-style" analysis of their intentions based on a large model, and assist enterprises in formulating strategies.

[0030] (4) Technology Transformation Path Suggestion Sub-module: Recommends collaborative R&D, commissioned development, patent licensing, etc., and supports personalized generation of "Enterprise-Technology Achievement Matching Report" and "Trend Impact Assessment Report".

[0031] Technically, Transformer is introduced for time series modeling to identify potential trend transitions: And combine semantic consistency verification to calculate the score function:

[0032] Feedback and Iterative Optimization Module. Based on user behavior and system output results, a feedback loop is formed to optimize model performance and continuously improve the system's intelligence level. This mainly includes:

[0033] (1) Feedback collection mechanism: collect users' adoption and correction behavior of prediction results;

[0034] (2) Model adjustment and fine-tuning: Optimize prompt word strategy and model parameters based on feedback;

[0035] (3) Knowledge structure update: Feed new trends and relationships into the knowledge graph to improve predictive adaptability.

[0036] This involves collecting user feedback and correcting behavior to drive the optimization of prompt words and semantic representation updates in the large model, thus achieving closed-loop learning: L total =L match +αL feedback +βL graph_update .

[0037] The second aspect of this invention provides an operational logic and data flow method for an enterprise technology evolution trend identification system based on large-scale model reasoning. The system starts with the strategic technology needs proposed by the enterprise and proceeds through the following five stages:

[0038] (1) Natural Language Parsing Stage: The input terminal supports various forms of technical requirement statements, such as enterprise interview records, project proposals, strategic planning documents, etc. The system calls the large model to generate a structured technical requirement representation and constructs a task tree.

[0039] (2) Semantic Representation and Graph Embedding Stage: The decomposed technical elements are converted into vector representations and aligned with entities in historical enterprise knowledge graphs and industry technology graphs to construct semantic connections between knowledge. The multidimensional relationship modeling of enterprises can be formalized as a graph G = (V, E, R). Here, V is the set of entities, such as enterprises, technical tasks, performance indicators, etc.; E is the set of edges, representing the semantic relationships between entities; and R is the set of relationship types. The embedding learning objective is to minimize the objective function: Among them, e h ,e t It is the head and tail entity vector, W r It is a relational weight matrix.

[0040] (3) Time Series Modeling and Dynamic Prediction Stage: Based on the evolution of time series models using embedded vectors, this stage introduces the Transformer structure and attention mechanism to enhance the modeling capabilities for trend abrupt changes and delayed feedback effects, thereby enabling future trend prediction. Technical Indicator Trend x l :T∈R d×T Its evolution path can be modeled using multi-layered Transformers: The final prediction result is

[0041] (4) Multi-round reasoning and result verification stage: The predicted results are input into the context of the large model. Chain-of-Thought Prompting is used to simulate expert judgment, verify the rationality of the trend, and provide multi-angle reiteration to ensure logical coherence and scenario suitability. The weight learning part of multi-source data fusion defines the importance weights w of data from different sources. i The final fusion is represented as: x fused =∑i=1 n w i ·x i ,where∑w i =1, and the weights can be adaptively learned through attention mechanisms or gradient descent. The semantic consistency score for trend rationality verification utilizes a set of multiple candidate trend explanations {r} obtained through chained suggestion reasoning. i}, through the language model scoring function f θ (r i )accomplish: Where C represents the original context semantics, and τ represents the consistency threshold.

[0042] (5) Feedback and Model Iteration Stage: The user's judgment on the trend will be fed back into the system to fine-tune the prediction model parameters and optimize the recommendation logic, forming a closed-loop learning and optimization mechanism.

[0043] The third aspect of this invention provides a core technical support mechanism for enterprise technology trend identification based on a large model, including but not limited to: key technical modules such as prompt word guidance design strategy, semantic consistency backtracking mechanism, cross-domain knowledge enhancement method, graph-text joint representation learning, and result visualization template construction. These technical solutions, while ensuring the system's prediction accuracy and interpretability, improve the model's generalization ability to multi-source heterogeneous data, further enhancing the system's practicality and robustness in real-world enterprise scenarios.

[0044] (1) Leveraging the strong semantic reasoning capabilities of large language models, technology trees, roadmaps, and trend graphs can be automatically generated from text. Simultaneously, it supports a federated learning architecture, ensuring that enterprise data can be modeled locally and deployed at the edge, enhancing data privacy and security.

[0045] (2) Construct dynamic technology maps and enterprise capability maps, and identify the "blank zones" and "fit points" of embedded differences. The system also supports the ability to quickly adapt to few-shot / zero-shot technology trends in new fields and new demand scenarios.

[0046] (3) Introduce a "semantic consistency backtracking" mechanism to verify the rationality of trend prediction results in the context of the large model and improve prediction credibility. Introduce locally interpretable methods such as SHAP / LIME and link them with trend path graph visualization to improve the transparency and credibility of prediction logic.

[0047] (4) Supports automatic fine-tuning of LLM architecture (AutoML+LoRA) and multi-model collaboration (Ensemble Reasoning), through the ability to adapt to the evolution of large model structure.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] (1) This invention combines guided prompts with a large model to achieve structured translation of unstructured text requirements, thus significantly improving the efficiency of structured processing. In addition, this invention also has the ability to adaptively adjust, avoiding the problem of traditional rule engines requiring a lot of manual maintenance.

[0050] (2) This invention makes enterprise technology trend prediction more forward-looking. By combining the ability to dynamically update multi-source data with the contextual semantic reasoning ability of large models, it can predict technology trends that have not yet been widely manifested in advance, providing decision support for enterprises in terms of patent layout and technology reserves.

[0051] (3) This invention significantly improves the intelligence level of result matching. By matching multi-dimensional enterprise profiles with deep semantic embedding of technological achievements, the supply and demand matching is upgraded from "keyword hit" to "semantic understanding", and the matching accuracy and hit rate are significantly higher than those of traditional database retrieval methods.

[0052] (4) The present invention has strong scalability and is not only applicable to industries such as manufacturing, new energy, and life sciences, but can also be extended to multiple application scenarios such as scientific research management, regional innovation analysis, and government industrial guidance, and has good versatility and deployment flexibility.

[0053] (5) The present invention has excellent technical explanatoryness and visualization. The output results include complete trend path diagrams, hot spot heat maps, semantic relationship networks and other graphic content, which effectively help decision-makers to clearly understand the basis for prediction and evolution logic, and improve the acceptability and practicality of the system.

[0054] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0055] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0056] Figure 1 This is a system architecture diagram of the enterprise technology evolution trend identification method based on large model reasoning according to an embodiment of the present invention;

[0057] Figure 2 This is a flowchart of the task decomposition module in an embodiment of the present invention;

[0058] Figure 3 This is a flowchart illustrating the process of constructing an enterprise technology profile according to an embodiment of the present invention;

[0059] Figure 4 This is a flowchart of the technical knowledge modeling module in an embodiment of the present invention;

[0060] Figure 5 This is a flowchart of the technology trend identification and prediction module according to an embodiment of the present invention;

[0061] Figure 6 This is a flowchart of the feedback and iterative optimization module in an embodiment of the present invention. Detailed Implementation

[0062] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0063] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0064] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0065] Example 1: Technology Trend Planning Based on Regional Strategy

[0066] This embodiment is applied to the management committee of a national high-tech zone, providing intelligent support for the key R&D directions of its regional industries over the next three years. The system deployment process is as follows:

[0067] (1) The management committee inputs regional technology development target documents into the system, such as the "Three-Year Action Plan for Technological Innovation in the High-tech Zone".

[0068] (2) The system extracts core task objectives, such as “construction of hydrogen energy vehicle platform” and “domestic substitution of high-density power battery”, through data acquisition and semantic parsing modules.

[0069] (3) Call the large model to decompose the target and form a hierarchical task structure: first-level tasks such as "hydrogen energy system optimization", second-level tasks such as "hydrogen cylinder storage density improvement" and "fuel cell temperature control", and third-level indicators such as "hydrogen storage ratio > 5.8wt%" and "low temperature start-up time < 15s".

[0070] (4) The system accesses profile information such as the historical technology trajectory, patent distribution, and R&D capabilities of key enterprises in the region to complete the construction of the regional technology map.

[0071] (5) The trend recognition module initiates multiple rounds of reasoning and outputs intelligent analysis reports such as “current technical pain points in the region”, “potential alternative paths”, and “areas with active competition”.

[0072] (6) The Management Committee, in conjunction with the system results, convened an expert review meeting to formulate recommendations for the regional key technology layout.

[0073] (7) Through the feedback module, experts adopt or correct the system's reasoning conclusions, and the system completes fine-tuning and enters a new round of trend identification.

[0074] This embodiment demonstrates that the system can achieve knowledge modeling and trend alignment between complex policy objectives and enterprise technological capabilities at the regional level, providing support for science and technology decision-making.

[0075] Example 2: Forward-looking R&D navigation for emerging enterprises

[0076] This example is applicable to a startup intelligent robot company that is unable to accurately judge trends and hot spots when faced with multiple technology options.

[0077] (1) The company inputs its product development vision: "Build a multimodal humanoid robot with self-learning capabilities".

[0078] (2) The system decomposes the target into multiple technical sub-fields, such as "reinforcement learning + ontology perception fusion algorithm", "modular high-strength servo system", and "adjustable skin sensor design".

[0079] (3) The existing technology accumulation of enterprises and the decomposition task are coupled into a graph through graph neural network.

[0080] (4) Use large models to analyze the evolution path of technology trends and predict that “electromyography-voltage coupling materials” will become the focus of international research in the next 18 months. It is recommended to prioritize the application of “active electromyography responsive skin” patents.

[0081] (5) The system outputs trend maps, technology hotspot timelines, international competitor comparison tables, and recommends potential partner organizations.

[0082] (6) Corporate executives will include the suggestions in the monthly strategy meeting for review, and the system will collect adoption marks to trigger a new round of semantic optimization.

[0083] This application demonstrates that the system can provide multi-path, multi-round predictive support for new enterprises without a strong foundation, reducing the risk of R&D blind spots.

[0084] Example 3: Identifying Differences in the Evolution Path of Smart Manufacturing Enterprises and Correcting Strategic Mistakes

[0085] A large industrial automation company faces two competitive paths in its "smart workshop upgrade" strategy: one is to take the route of integrating high-end control systems, and the other is to develop lightweight edge intelligent sensing equipment.

[0086] (1) Enterprises input market data, main R&D directions and potential M&A targets for the past three years.

[0087] (2) The system identifies the enterprise’s R&D strengths through the enterprise profiling module, which are concentrated in the PLC core module, and its weaknesses are in the self-developed chips and sensing components.

[0088] (3) The task decomposition module subdivides the goal of “intelligent workshop upgrade” into 10 sub-tasks, involving subsystems such as communication delay optimization, equipment collaboration, and fault prediction.

[0089] (4) The knowledge modeling module constructs a semantic graph of internal and external patents and competitors' patents to identify differentiated technology paths.

[0090] (5) The trend recognition module calls the big model to compare the evolution trends of the "edge sensing integration solution" and the "centralized control upgrade solution" in terms of technology development rate, patent distribution, policy support, etc., and judges that the latter is in a period of growth bottleneck.

[0091] (6) The system recommends that companies start joint venture projects in the field of sensor chips as soon as possible and provides a list of mergers and acquisitions for reference.

[0092] Example 4: Cross-industry technology trend scanning and opportunity identification by the investment department of a listed company

[0093] A listed technology company plans to establish a strategic investment fund and needs to determine which of the 20 emerging technology directions, such as "brain-like computing," "flexible electronics," "liquid metal," and "bionic materials," will become core technological variables that could potentially leverage its existing industrial chain within the next three years.

[0094] (1) Input investment preferences into the system (power equipment, medical devices, industrial AI, etc.).

[0095] (2) The system calls the large model to quickly analyze the research dynamics, application scenarios and policy trends of each type of technology concept, and constructs a "technology maturity radar chart" by combining large-scale literature and patent data.

[0096] (3) By comparing the existing capabilities of enterprises, the system judges that “bionic materials” will first form an impact window in the field of “minimally invasive instruments - highly adaptable materials”.

[0097] (4) The system recommends continuously tracking the evolution roadmap of flexible neural interface materials and outputting trend navigation reports including semantic evolution path, technology congestion analysis, investment hotspot curve, etc.

[0098] (5) The investment department regularly compares and marks the expert judgments with the system judgments and provides feedback. The system optimizes and improves the interpretability of the suggestions through prompt words.

[0099] Example 5: Consistency Analysis of Technology Evolution in Subsidiaries of Group Enterprises

[0100] A large equipment manufacturing group has eight subsidiaries, belonging to different business units such as steel, energy, and construction machinery. The group's technology management department wants to identify whether there are overlaps, gaps, or opportunities for synergy in the evolution paths of its various subsidiaries along three main technology lines: "high reliability, intelligent diagnostics, and remote control".

[0101] (1) The Group inputs the technology roadmaps, project plans and patent literature databases of its subsidiaries over the past 5 years into the system.

[0102] (2) The system identifies the task priorities of different subsidiaries. For example, the steel subsidiary focuses on "heat treatment automation", while the construction machinery subsidiary focuses on "vibration status perception".

[0103] (3) After task decomposition and graph fusion, the large model is called to analyze the semantic proximity and path similarity of task nodes.

[0104] (4) The system outputs a technology evolution coupling map and similarity matrix to identify that “high temperature environment sensor” is deployed in all three subsidiaries but there is no cross-cooperation.

[0105] (5) The management department, referring to the system recommendations, promoted the establishment of a joint cross-technology research project and established a "Group Internal Trend Alignment Monitoring" module as an institutional tool.

[0106] As can be seen from the above embodiments, the system of the present invention relies on the enterprise's technical semantic understanding capabilities through a large language model, and can achieve: automatic trend decomposition and graph generation for future needs; comparative analysis and semantic evolution prediction across technology paths; path gap identification and strategy suggestions based on enterprise profiles; trend coupling analysis across subsidiaries and industries; and model self-evolution guided by multi-round feedback and expert correction.

[0107] This system can be embedded into various existing platforms such as enterprise PLM (Product Lifecycle Management), knowledge management platforms, R&D management systems, and scientific research evaluation systems through private deployment or API calls. It can also be integrated with patent search tools, scientific literature analysis tools, or intelligent BI tools to expand its intelligent decision-making capabilities. The system supports compatibility with mainstream large-scale model platforms and can be configured with industry terminology databases, knowledge templates, and multilingual semantic parsers to achieve cross-language and cross-context trend recognition services. In the future, this system can be further expanded to application scenarios such as government think tanks, industry associations, and research institutions, supporting their medium- and long-term technology strategy forecasting and industrial planning.

[0108] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying enterprise technology evolution trends based on large model reasoning, characterized in that, Includes the following steps: S1. Multi-source data collection and standardized preprocessing: Collect multi-source heterogeneous data including patent documents, scientific papers, policy documents, and market dynamics, and use semantic parsing and information extraction methods to transform the original text into standardized technical knowledge units. S2. Enterprise Technology Profile Construction: Based on the enterprise's historical R&D data, construct a multi-dimensional enterprise technology profile covering technology needs, R&D capabilities, collaborative relationships, and evolution paths; S3. Natural Language Requirements Analysis and Task Structuring Based on Large Language Model: Input the strategic technology goals expressed in natural language by the enterprise into the large language model, and generate a structured technology task tree through prompt word engineering to form a hierarchical R&D task system. S4. Technical Knowledge Graph Modeling: Integrate enterprise technical tasks with external industry ontology knowledge, construct a graph structure based on entity-relation semantic network, and perform vectorized learning through graph neural network to achieve unified modeling of enterprise knowledge and trend information; S5. Trend Identification and Path Prediction Reasoning: By leveraging the multi-round reasoning capabilities of a large language model and combining time series modeling and graph semantic input, it identifies the differences between the company's existing technologies and industry development trends, and predicts potential technology leap paths. S6. Feedback and Optimization Mechanism: Based on the user's adoption and correction behavior of the system output results, update the prompt word strategy and knowledge structure, optimize the reasoning logic, and realize the closed-loop learning and dynamic adjustment of the model.

2. The method according to claim 1, characterized in that, The natural language requirements parsing in S3 uses a large language model with reasoning capabilities to generate a multi-layered technical task tree that includes primary tasks, technical decomposition points, and key technical indicators, guided by prompt words.

3. The method according to claim 1, characterized in that, The trend identification described in S5 includes the following sub-modules: S101, Hotspot Trend Identification Submodule: Constructing a technical heatmap based on the evolution of industry hot words; S102, Alternative Path Prediction Submodule: Identifies technical bottlenecks and infers possible alternatives; S103, Competitive Landscape Deduction Submodule: Based on the large model, conduct contextual analysis and reasoning to judge the technological layout and strategic intentions of competitive entities; S104, Technology Transformation Suggestion Submodule: Outputs personalized trend navigation suggestions, including results matching paths and impact analysis.

4. The method according to any one of claims 1 to 3, characterized in that, The reasoning process employs chain-of-thought prompting to explicitly control the reasoning path of trend judgment, thereby improving the interpretability and adaptability of trend prediction.

5. A system for identifying enterprise technology evolution trends based on the method described in claims 1 to 4, characterized in that, The system includes: a data acquisition and preprocessing module, an enterprise technology profiling module, a requirements analysis and task decomposition module, a technical knowledge modeling module, a trend identification and prediction module, and a feedback and iterative optimization module.

6. The system according to claim 5, characterized in that: The task parsing module is based on a large language model with language understanding and logical reasoning capabilities, and automatically outputs a technology tree structure by combining prompt word guidance strategies; the knowledge graph module is built based on industry ontology and supports semantic consistency backtracking; the system supports cross-model adaptation and can call large language models with different architectures to enhance the robustness of trend recognition.

7. The system according to claim 5, characterized in that, This system can achieve the following functions: automatically generate structured R&D tasks from unstructured strategic goals input by enterprises; make differential reasoning judgments on the current technological status and industry trends; and output trend predictions, differential coupling, and recommended routes for technology layout support.

8. A computer program product for performing the method according to any one of claims 1 to 7, characterized in that, When executed by the processor, the program enables the system to: invoke a large language model to perform semantic parsing and task decomposition of technical objectives; link the enterprise knowledge graph with external industry corpora; and complete the enterprise technology trend identification and path prediction output.

9. A computer-readable storage medium storing the computer program product as claimed in claim 8, wherein the storage medium may be a local server, a cloud platform, or a middleware environment that interfaces with a large model.

Images