Digital service resource supply and demand matching method, system and device and storage medium

Abstract

The invention relates to the technical field of data processing, and particularly provides a digital service resource supply and demand matching method, system and device and a storage medium, and the method comprises the steps: a demand side interaction module collects multi-modal demand data and carries out the standardization processing; deep semantic analysis of demand information is realized through a large model, and a structured demand portrait is generated and coded into a high-dimensional demand vector; semantic understanding is carried out on the supply side resource library, and resource vectors and deep semantic tags are generated; calculating the semantic similarity of the demand vector and the resource vector, and performing multi-factor weighted decision by combining the scene adaptation degree, the creditworthiness and the commercial feasibility; and outputting a recommendation list through the service management platform. According to the invention, the efficiency and success rate of resource docking in digital transformation ecology are greatly improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of data processing, and particularly relates to a digital service resource supply and demand matching method, system, device and storage medium. BACKGROUND

[0002] In the current wave of industrial digital transformation, the low efficiency of supply and demand matching has become a key bottleneck restricting its development. On the one hand, the transformation needs of small and medium-sized enterprises are urgent, but their demand descriptions are often vague and unstructured, making it difficult to accurately express them. On the other hand, there are a large number of digital transformation resources (such as solutions, software, and services) in the market, which are heterogeneous and described by different professionals. Existing matching platforms mostly rely on keyword-based retrieval or simple classification and screening, which have the following significant defects: they cannot understand the deep semantics of user needs, are difficult to process unstructured information such as documents and pictures, and lack the cognitive and reasoning ability of the industrial field. This results in low matching accuracy and insufficient personalization, causing the general dilemma of "demand side cannot find, supply side cannot accurately push", which seriously hinders the effective circulation and application of high-quality resources. SUMMARY

[0003] In view of the above deficiencies of the prior art, the present application provides a digital service resource supply and demand matching method, system, device and storage medium to solve the above technical problems.

[0004] In a first aspect, the present application provides a digital service resource supply and demand matching method, comprising: obtaining digital service demand information of a demand side through a demand side interaction module, and performing multi-modal data fusion and standardization processing on the demand information; performing deep semantic analysis on the processed demand information through a large model, generating a structured demand portrait, and encoding the structured demand portrait into a high-dimensional demand vector; performing deep semantic understanding on resource information in a supply side resource library through the large model, and generating deep semantic tags, generating a resource vector of the resource based on the deep semantic understanding result and the deep semantic tags; calculating the semantic similarity between the demand vector and the resource vector, combining the semantic similarity and the scene adaptation degree, resource credibility and commercial feasibility calculated based on the demand portrait and a pre-constructed industrial knowledge graph to make a multi-factor weighted decision, and generating a resource recommendation list; displaying the resource recommendation list to the demand side through a service management platform.

[0005] In a second aspect, the present application provides a digital service resource supply and demand matching system, comprising: The demand acquisition module is used to acquire digital service demand information from demanders through the demand-side interaction module, and to perform multimodal data fusion and standardization processing on the demand information. The requirement parsing module is used to perform deep semantic parsing on the processed requirement information through a large model, generate a structured requirement profile, and encode the structured requirement profile into a high-dimensional requirement vector. The supply analysis module is used to perform deep semantic understanding of resource information in the supply-side resource library through the large model, generate deep semantic tags, and generate resource vectors based on the deep semantic understanding results and the deep semantic tags. The similarity calculation module is used to calculate the semantic similarity between the demand vector and the resource vector. It combines the semantic similarity with the scenario adaptability, resource reputation and commercial feasibility calculated based on the demand profile and the pre-built industrial knowledge graph to make a multi-factor weighted decision and generate a resource recommendation list. The list display module is used to display the resource recommendation list to the requester through the service management platform.

[0006] Thirdly, a device is provided, comprising: The memory is used to store the supply and demand matching program for digital service resources; A processor is configured to implement the steps of the digital service resource supply and demand matching method as provided in the first aspect when executing the digital service resource supply and demand matching program.

[0007] Fourthly, a computer-readable storage medium is provided, on which a digital service resource supply and demand matching program is stored, wherein when the digital service resource supply and demand matching program is executed by a processor, the steps of the digital service resource supply and demand matching method provided in the first aspect are implemented.

[0008] The beneficial effects of this invention are as follows: the digital service resource supply and demand matching method, system, device, and storage medium provided by this invention, through the deep semantic understanding and demand deconstruction capabilities of large models, achieve high-precision intent recognition of fuzzy and unstructured demands, fundamentally overcoming the limitations of traditional keyword matching; by utilizing the vectorized representation of supply and demand information and multi-factor intelligent decision-making, it achieves a leap from "surface keyword matching" to "deep value matching," significantly improving the accuracy and scenario adaptability of resource recommendations; at the same time, the system has online learning and evolution capabilities, which can continuously optimize the matching effect, greatly improving the efficiency and success rate of resource docking in the digital transformation ecosystem, and effectively reducing the search and trial-and-error costs of enterprises. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic flowchart of a method according to an embodiment of the present invention.

[0011] Figure 2 This is a schematic block diagram of a system according to an embodiment of the present invention.

[0012] Figure 3 This is a schematic diagram of the structure of a device provided in an embodiment of the present invention. Detailed Implementation

[0013] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0014] Unless otherwise defined, 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. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0015] The digital service resource supply and demand matching method provided in this embodiment of the invention is executed by a computer device, and correspondingly, the digital service resource supply and demand matching system runs on the computer device.

[0016] Figure 1 This is a schematic flowchart illustrating a method according to an embodiment of the present invention. Wherein, Figure 1 The implementing entity can be a digital service resource supply and demand matching system. Depending on different needs, the order of steps in this flowchart can be changed, and some steps can be omitted.

[0017] like Figure 1 As shown, the method includes: S1. Obtain the digital service demand information of the demand side through the demand-side interaction module, and perform multimodal data fusion and standardization processing on the demand information; S2. Perform deep semantic parsing on the processed demand information through a large model to generate a structured demand profile, and encode the structured demand profile into a high-dimensional demand vector; S3. Perform deep semantic understanding on the resource information in the supply-side resource library through the large model, generate deep semantic tags, and generate resource vectors based on the deep semantic understanding results and the deep semantic tags; S4. Calculate the semantic similarity between the demand vector and the resource vector, and combine the semantic similarity with the scenario adaptability, resource credibility and commercial feasibility calculated based on the demand profile and the pre-built industrial knowledge graph to make a multi-factor weighted decision and generate a resource recommendation list. S5. Display the resource recommendation list to the requester through the service management platform.

[0018] In one embodiment of the present invention, based on step S1, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0019] S101. For uploaded document format requirements, use a text parsing library to extract the text content; for image format requirements, integrate optical character recognition service to extract the text information.

[0020] During step S101, the system processes various user-uploaded request information through a multimodal information extraction unit. For document-format requests, the system uses the open-source Apache PDFBox library (version 2.0.x) to parse PDF files and the python-docx library (version 0.8.x) to parse Word documents, thereby accurately extracting the plain text content. For image-format requests (such as images of on-site equipment, flowcharts, etc.), the system integrates commercial optical character recognition (OCR) cloud services (such as Baidu Cloud OCR General Text Recognition API or Alibaba Cloud OCR Document Self-Learning Service) to achieve high-precision text extraction. The system sends image data to the API endpoints of the aforementioned services through a pre-defined HTTP client and parses the returned JSON format results to obtain the structured text in the images. To ensure robustness, this unit has a built-in retry mechanism, automatically retrying up to two times if the first call fails.

[0021] S102. The extracted original text is denoised and segmented, and colloquial expressions and industry terms are mapped to standardized terms in the standard terminology database.

[0022] During step S102, the system processes the raw text extracted in S101 using a text cleaning and standardization unit. First, text cleaning is performed: the system uses predefined regular expression rules (e.g., / [\x00-\x1F\x7F] / g) to remove control characters and invisible characters from the text. Then, precise pattern segmentation is performed using the open-source Chinese word segmentation tool Jieba (version 0.42.x). Next, terminology standardization is performed: the system maintains an "Industrial Digitalization Standard Terminology Database" stored in a MySQL database, which maps common colloquial expressions (e.g., "implementing the system") with standardized terms (e.g., "implementing the MES system"). The segmented sequence is matched against this terminology database. For matched colloquial words or synonyms, the system uniformly replaces them with preset standardized terms, thus forming standardized requirement text and laying the foundation for subsequent semantic understanding.

[0023] S103. Automatically link the enterprise profile data of the demand side, which includes the industry, enterprise size and current level of digitization, to enhance the background information for demand analysis.

[0024] During step S103, the system automatically associates and imports background information from the requester through the context completion unit. When a user submits a request, the system obtains their unique enterprise identifier (EnterpriseID) from their login session information. Subsequently, this unit uses this identifier as a query key to access the backend enterprise profile database (e.g., the enterprise_profiles collection deployed in MongoDB). From this profile, the system automatically extracts three key dimensions: industry (following the GB / T 4754-2017 National Economic Industry Classification Standard, such as "C366-Automotive Parts and Accessories Manufacturing"), enterprise size (based on the Ministry of Industry and Information Technology's Joint Enterprise

[2011] No. 300 standard, such as "medium-sized"), and current level of digitalization (quantified through a pre-set assessment questionnaire score, such as "L2-Single-point Informatization"). This structured data is automatically injected into the current request processing context, forming an enhanced request data object, providing indispensable background information for subsequent scenario-based reasoning in the industrial big data model.

[0025] In one embodiment of the present invention, based on step S2, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0026] First, the system retrieves the industry scenario information associated with the enterprise profile database linked to the demander. Specifically, the system queries the enterprise_profile collection stored in the MongoDB database using the user's unique enterprise identifier to obtain the industry_code field. This field conforms to the National Industrial Classification of Economic Activities (GB / T 4754-2017) standard and can map to a specific industry scenario, such as "C366 - Manufacturing of Automotive Parts and Accessories".

[0027] Subsequently, the system uses a finely tuned large language model (e.g., a model based on LLaMA 2 or ChatGLM architecture, finely tuned with full parameters or LoRA using massive amounts of industrial text, knowledge graphs, and digital transformation scenario instruction data) to perform multi-label intent classification on the preprocessed requirement text. The model is required to output one or more probabilistic intent labels from a predefined list of intents (e.g., "production efficiency improvement", "quality control", "cost control", "energy management"), for example, ["production efficiency improvement": 0.95, "cost control": 0.72].

[0028] Next, the system performs core causal reasoning and requirement deconstruction. The system pre-builds a causal reasoning prompt template with the following text structure: "You are a senior expert in the [industry variable] domain. Please analyze the 3-5 possible root causes behind the client's requirement of '[requirement variable]'. Please list only the root causes, each beginning with '-'." The system inputs industry scenario information (such as "automotive parts and accessories manufacturing") into [industry variables] and preprocessed user demand text (such as "low production efficiency in injection molding workshop") into [demand variables], forming a complete prompt instruction. After inputting this instruction into the aforementioned large language model, it receives the returned text response and uses rule parsing (such as by newline character and '-' prefix) to extract a structured list of root causes, such as: frequent equipment downtime, lack of predictive maintenance; unreasonable production scheduling, high number of line changes; untimely material supply, causing production line waiting.

[0029] Based on this list of root causes, the system again uses a large language model to derive derived technology requirements with instructions (such as "For the above root causes, please list the digital technologies or systems that need to be introduced"), such as ["Equipment data acquisition and monitoring system", "Advanced Planning and Scheduling (APS) software", "Material Management System"].

[0030] In parallel, the system extracts key entities (such as "injection molding machine" and "production line A"), quantitative indicators (such as "OEE target 85%) and business constraints (such as "budget within 500,000") from the requirement text through a named entity recognition model (e.g., a model trained on industrial text based on the BERT-base architecture and using the BIO annotation system). These information together constitute the set of key constraints.

[0031] Finally, the system integrates and serializes all the above information—industry scenarios, core intent tags, root cause lists, derived technical requirements, and key constraints—into a structured JSON object, namely, a requirement profile data object. The text content of this JSON object is then input into the encoder of a large model (e.g., the TransformerEncoder of the BERT model), and the output vector of its [CLS] flag is used as the aggregated output to generate a 768-dimensional high-dimensional requirement vector. This vector comprehensively represents the complete semantic information of the requirement.

[0032] In one embodiment of the present invention, based on step S3, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0033] First, the system integrates multi-source heterogeneous data from the supply-side resource repository. Specifically, it queries structured attribute data of resources from a MySQL relational database via a database connector (such as JDBC), including fields such as product name, supplier, price, and technical parameters. Simultaneously, it obtains unstructured descriptive text of resources from object storage services (such as AWS S3 or Alibaba Cloud OSS) via a file parsing service, including product manuals, technical white papers, and success story documents, and uses text extraction libraries such as Apache Tika to extract raw text content from PDF, Word, and other document formats.

[0034] Next, the system performs value proposition extraction. The system pre-constructs a value proposition extraction prompt template, the text structure of which is as follows: "Please extract the three core business problems (i.e., value propositions) that the following digital services or products address. Description content: [Insert unstructured description text of the resource here]. Please list only the business problems, labeling each problem with a numerical number." After filling the template with unstructured descriptions of resources, a complete prompt instruction is generated and input into a large language model fine-tuned for industrial applications (such as a model based on GPT-3.5 Turbo or LLaMA 2). Upon receiving the text response from the model, a value proposition summary is extracted through rule parsing, for example: "1. Solves the problem of opaque production schedules; 2. Solves the problem of low efficiency in manual recording of quality data; 3. Solves the problem of inability to quantify and analyze equipment utilization." Then, the system performs deep semantic annotation. Based on a predefined industrial domain tagging system (stored in JSON format, containing standardized tag vocabularies for three dimensions: "applicable scenarios," "pain points solved," and "core capabilities"), specific classification and annotation prompts guide the same large language model to assign deep semantic tags to resources. For example, for an MES system resource, the output tags might be: Applicable Scenarios: ["Discrete Manufacturing", "Assembly Workshop"]; Pain Points Solved: ["Information Silos", "Production Black Box"]; Core Capabilities: ["Real-time Monitoring", "APS Scheduling", "ANDON System"].

[0035] Finally, the system performs resource vector generation. The structured attribute data of the resource (in key-value pairs), the value proposition summary, and the deep semantic tags are concatenated into a complete, semantically rich resource description text according to a preset template. This text is then input into a pre-trained large model encoder (e.g., the Encoder module of BERT-base, or the Sentence-BERT model) to obtain the output vector of its [CLS] flag, or to perform mean pooling on the sequence output. This ultimately generates a 768-dimensional high-dimensional resource vector, which is stored in a vector database (such as Milvus or Pinecone) for subsequent efficient similarity retrieval.

[0036] In one embodiment of the present invention, based on step S4, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0037] S401. Multidimensional Factor Calculation The system first calculates scores across four core dimensions: Semantic similarity score: The system uses the cosine_similarity function from the scikit-learn library to calculate the cosine similarity between the demand vector and each resource vector. To improve retrieval efficiency with massive resources, the system integrates the vector database Milvus (version 2.3.x), and quickly obtains the Top-1000 most relevant resource vectors and their initial similarity scores through its approximate nearest neighbor search interface. This score is then normalized to the [0,1] interval using Min-Max and used as the semantic similarity (S_semantic).

[0038] Scenario Adaptability Score: The system is based on a pre-built industrial knowledge graph (stored using the Neo4j graph database, containing entity relationships such as "industry-scenario-resource"). For example, when the industry scenario in the demand profile is "automotive parts manufacturing," the system uses Cypher queries to match applicable scenario paths for resources highly relevant to this industry, and combines this with pre-defined business rules in the Drools rule engine (e.g., IF industry="automotive" AND demand contains "traceability" THEN resource tag contains "IATF 16949") for logical reasoning. A score of 1.0 is awarded if all strong rules are met; otherwise, the score (S_scenario) is calculated based on the weight of the matching path.

[0039] Reputation Score: The system extracts click-through rates and inquiry volumes corresponding to resource vectors over the past 30 days from a behavior log database (such as Elasticsearch), extracts user ratings (1-5 points) from the business database, and uses the SnowNLP library to perform sentiment analysis on text reviews to obtain a sentiment polarity score (-1 to 1). A time decay function, weight = exp(-λ * Δt) (where λ is the decay coefficient and Δt is the number of days since the current date), is used to assign higher weights to recent data. Finally, a weighted average is used to obtain the comprehensive reputation score (S_reputation).

[0040] Business Feasibility Score: The system first applies hard rule filtering, directly excluding resources whose prices exceed the required budget or whose service area does not include the user's location through SQL queries. For resources that pass the hard filtering, a Gaussian function f(x) = exp(-(x - μ)) is used to address soft constraints such as delivery time. 2 / (2 * σ 2 )) Calculate the compliance score, where x is the resource delivery cycle, μ is the user's expected cycle, and σ is the tolerance parameter (e.g., set to μ / 3). The result is used as the business feasibility score (S_business).

[0041] S402. Decision Fusion and List Generation The system calculates the comprehensive recommendation score using a weighted summation formula: Score_total = W_semantic * S_semantic + W_scenario * S_scenario + W_reputation * S_reputation + W_business * S_business The initial values ​​of the weighting factors are configured using the following dynamic strategy.

[0042] After sorting, the system generates a natural language explanation for each recommended resource: using a predefined template (such as "This solution is recommended because of [highly matching functionality | industry-specific design | good market reputation]"), combined with the actual scores of the resource in each factor, the system selects high-scoring items to fill the template, generating the final readable explanation.

[0043] S403. Dynamic Weight Configuration Strategy The system implements a dynamic weight configuration mechanism based on the platform's business lifecycle: Stage Division and Determination: The platform predefines three business stages and their quantitative thresholds: initial stage of system launch (registered users < 10,000), growth stage (10,000 ≤ users < 100,000, and monthly transaction volume < 5 million), and maturity stage (users ≥ 100,000 or monthly transaction volume ≥ 5 million). The system periodically (daily) monitors these operational indicators and automatically determines the current stage.

[0044] Phase weighting strategy: In the initial stage of system launch: the weight configuration is [W_semantic=0.7, W_scenario=0.2, W_reputation=0.05, W_business=0.05], with technical matching as the core.

[0045] Platform growth stage: The weights are adjusted to [W_semantic=0.5, W_scenario=0.2, W_reputation=0.15, W_business=0.15], increasing the weight of business factors.

[0046] During the platform's maturity phase, the system will enable one or two of the following parallel dynamic weight allocation modes: Mode 1: Differentiated weight configuration based on user profile segmentation User profiling and segmentation: Data source: The system integrates users' enterprise profile data (such as industry and size) and their historical behavior data (such as browsing, searching, consulting, and transaction records).

[0047] Feature engineering: Extracting key features from behavioral data, such as technical document reading rate, high-precision solution click preferences, price filtering frequency, and attention to success cases.

[0048] Clustering: Unsupervised learning algorithms, such as K-Means clustering (implemented using the scikit-learn library), are used to divide users into several groups based on the above features. Typical business groups include: technology-preference-oriented (features: frequently clicking on technical parameters and documents), cost-sensitive (features: frequently using price filtering and paying attention to discounts), reputation-oriented (features: highly valuing ratings and reviews), and comprehensive decision-making.

[0049] Group-specific weighting strategy: The system predefines a set of weighting strategies for each identified user group. These strategies are stored in the `weight_strategy` table in the configuration center (such as when using Apollo or Nacos), and their core logic is as follows: Technology preference: [W_semantic=0.65, W_scenario=0.20, W_reputation=0.10, W_business=0.05] Cost-sensitive: [W_semantic=0.40, W_scenario=0.15, W_reputation=0.10, W_business=0.35] Word-of-mouth oriented: [W_semantic=0.35, W_scenario=0.15, W_reputation=0.40, W_business=0.10] When a user makes a request, the system queries the user's group in real time and loads the corresponding weight strategy for calculation.

[0050] Mode 2: Weighting strategy optimization based on A / B testing framework Traffic splitting and strategy deployment: At the API gateway (such as Spring Cloud Gateway) or load balancer (such as Nginx) level, the system splits the total user traffic into multiple mutually exclusive buckets based on the hash value of the user ID, for example, into three groups A, B and C, with each group accounting for 33%.

[0051] A distributed configuration center dynamically assigns a different weight strategy to each bucket. For example: Bucket A: Strategy Base-[0.5, 0.2, 0.15, 0.15] (control group); Bucket B: Strategy TechFocus-[0.6, 0.2, 0.1, 0.1] (experimental group 1); Bucket C: Strategy BizFocus-[0.4, 0.2, 0.1, 0.3] (experimental group 2).

[0052] Effectiveness evaluation and winning strategy decision-making: Data collection: The system marks the source of each recommendation strategy in the data tracking points and continuously collects the core business metrics under each strategy bucket, such as recommendation click-through rate, in-depth consultation conversion rate, and transaction conversion rate.

[0053] Statistical tests: After each evaluation period (e.g., 24 hours), the data analysis service uses the Python statsmodels library to perform independent samples t-tests or chi-square tests on the conversion rate indicators of each experimental group and control group to determine the statistical significance of the difference (usually p-value < 0.05).

[0054] Automatic decision-making: If a strategy in an experimental group (such as the TechFocus strategy) significantly outperforms the control group in key metrics and the advantage remains stable (e.g., for two consecutive periods), the system will promote this winning strategy to the new global default strategy through the configuration center, end the current test, and prepare to start the next round of optimization experiments.

[0055] Online learning optimization: The system collects implicit feedback data such as user clicks, in-depth consultations, and transactions. Using these behaviors as optimization targets, the system uses a linear regression model from scikit-learn every cycle (e.g., weekly) to fit the relationship between the scores of each factor and the positive feedback rate, and then deduce the better initial weight values ​​for each stage in the next cycle, thus achieving the system's self-evolution.

[0056] In one embodiment of the present invention, based on step S5, a possible embodiment will be given below, and its specific implementation will be described in a non-limiting manner.

[0057] Data Interface Encapsulation: The backend recommendation service provides recommendation results via a RESTful API, returning structured JSON data. The data package contains the following core fields: recommendation_list: An array of resources sorted in descending order of overall score; Each resource item includes: resource_id, resource_name, supplier_info, comprehensive_score, and detailed scores for each dimension; Explanation: The natural language recommendation reason automatically generated for this resource.

[0058] The front-end uses visual rendering and a responsive design, displaying recommendation results through the following components: Main Recommendation Area: Displays the Top 10 recommended resources in the form of a card list. Each card clearly displays the resource name, supplier, and overall recommendation rating (star rating). Details panel: After the user clicks "View Details", the panel dynamically expands to display the following structured information: Value Proposition Summary (from resource value penetration results); Matching analysis: The scores for each dimension (semantic fit, scene adaptability, etc.) are displayed as a percentage progress bar. Reason for recommendation: Directly displays the content of the explanation field; Comparison Function: Provides a multi-selection comparison control, allowing users to compare the advantages and disadvantages of multiple solutions in parallel.

[0059] In some embodiments, the user churn prediction system may include multiple functional modules composed of computer program segments. The computer programs for each segment in the digital service resource supply and demand matching system may be stored in the memory of a computer device and executed by at least one processor to perform (see details). Figure 1 (Description) The function of matching supply and demand for digital service resources.

[0060] In this embodiment, the digital service resource supply and demand matching system can be divided into multiple functional modules based on its functions, such as... Figure 2 As shown. The module referred to in this invention is a series of computer program segments that can be executed by at least one processor and perform a fixed function, and is stored in memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0061] The demand acquisition module is used to acquire digital service demand information from demanders through the demand-side interaction module, and to perform multimodal data fusion and standardization processing on the demand information. The requirement parsing module is used to perform deep semantic parsing on the processed requirement information through a large model, generate a structured requirement profile, and encode the structured requirement profile into a high-dimensional requirement vector. The supply analysis module is used to perform deep semantic understanding of resource information in the supply-side resource library through the large model, generate deep semantic tags, and generate resource vectors based on the deep semantic understanding results and the deep semantic tags. The similarity calculation module is used to calculate the semantic similarity between the demand vector and the resource vector. It combines the semantic similarity with the scenario adaptability, resource reputation and commercial feasibility calculated based on the demand profile and the pre-built industrial knowledge graph to make a multi-factor weighted decision and generate a resource recommendation list. The list display module is used to display the resource recommendation list to the requester through the service management platform.

[0062] Figure 3 The digital service resource supply and demand matching method provided in the embodiments of this application can be applied to devices. Those skilled in the art will understand that the device structure involved in the embodiments of this invention does not constitute a limitation on the device. A device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. In the embodiments of this invention, the device includes, but is not limited to, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of this application described and / or claimed herein.

[0063] The device 300 may include a processor 310, a memory 320, and a communication unit 330. These components communicate via one or more buses. Those skilled in the art will understand that the server structure shown in the figure does not constitute a limitation of the present invention. It may be a bus topology or a star topology, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0064] The memory 320 can be used to store execution instructions of the processor 310. The memory 320 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. When the execution instructions in the memory 320 are executed by the processor 310, the device 300 is able to perform some or all of the steps in the above method embodiments.

[0065] The processor 310 serves as the control center of the storage device, connecting various parts of the electronic device via various interfaces and lines. It executes software programs and / or modules stored in the memory 320, and calls data stored in the memory to perform various functions of the electronic device and / or process data. The processor can be composed of integrated circuits (ICs), such as a single packaged IC or multiple packaged ICs with the same or different functions connected together. For example, the processor 310 may consist only of a central processing unit (CPU). In this embodiment of the invention, the CPU may have a single processing core or include multiple processing cores.

[0066] The communication unit 330 is used to establish a communication channel, enabling the storage device to communicate with other devices. It can receive user data sent by other devices or send user data to other devices.

[0067] The present invention also provides a computer storage medium, wherein the computer storage medium may store a program, which, when executed, may include some or all of the steps provided in the embodiments of the present invention. The storage medium may be a magnetic disk, an optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0068] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, or any other medium capable of storing program code. It includes several instructions to cause a computer device (which may be a personal computer, a server, or a second device, network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0069] The same or similar parts between the various embodiments in this specification can be referred to mutually. In particular, the device embodiments are basically similar to the method embodiments, so the description is relatively simple, and the relevant parts can be referred to the description in the method embodiments.

[0070] In the embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or modules may be electrical, mechanical, or other forms.

[0071] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0072] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0073] Although the present invention has been described in detail with reference to the accompanying drawings and preferred embodiments, the present invention is not limited thereto. Various equivalent modifications or substitutions can be made to the embodiments of the present invention by those skilled in the art without departing from the spirit and essence of the invention, and such modifications or substitutions should all be within the scope of the present invention. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should also be covered within the protection scope of the present invention.

Claims

1. A method for matching the supply and demand of digital service resources, characterized in that, include: The demand-side interaction module obtains the digital service demand information from the demand side, and performs multimodal data fusion and standardization processing on the demand information. The processed demand information is subjected to deep semantic analysis through a large model to generate a structured demand profile, and the structured demand profile is encoded into a high-dimensional demand vector. The large model is used to perform deep semantic understanding on the resource information in the supply-side resource library and generate deep semantic tags. Based on the deep semantic understanding results and the deep semantic tags, resource vectors of the resources are generated. Calculate the semantic similarity between the demand vector and the resource vector, and combine the semantic similarity with the scenario adaptability, resource reputation and commercial feasibility calculated based on the demand profile and the pre-built industrial knowledge graph to make a multi-factor weighted decision and generate a resource recommendation list. The resource recommendation list is displayed to the requesting party through the service management platform.

2. The method according to claim 1, characterized in that, The system obtains digital service demand information from demanders through a demand-side interaction module, and performs multimodal data fusion and standardization processing on the demand information, including: For uploaded document format requirements, a text parsing library is used to extract the text content; for image format requirements, an optical character recognition service is integrated to extract the text information; the text content and text information are then merged to form the original text. The extracted raw text is denoised and segmented, and colloquial expressions and industry terms are mapped to standardized terms in a standard terminology library to generate preprocessed requirement text. The system automatically links to the enterprise profile data of the demand side, which includes the industry, enterprise size, and current level of digitization, to enhance the background information for demand analysis.

3. The method according to claim 1, characterized in that, The processed demand information is subjected to deep semantic parsing using a large model to generate a structured demand profile, which is then encoded into a high-dimensional demand vector, including: Obtain industry scenario information from enterprise profile data associated with the demand side; The preprocessed requirement text is used to perform multi-label intent classification using a large language model that has been fine-tuned by instructions, and the probabilistic core intent labels are identified. Combining the industry scenario information, and using a preset causal reasoning prompt template, the large language model is guided to perform root cause analysis of user needs based on specific industry knowledge, and generate a list of inferred root causes. Based on the list of root causes, the derived technological requirements for addressing these root causes are derived through the large language model. By using named entity recognition technology, key entities, quantitative indicators and business constraints are extracted from the preprocessed requirement text, and key constraints are generated based on the key entities, quantitative indicators and business constraints. The industry scenarios, along with the core intent tags, root cause list, derived technical requirements, and key constraints obtained from parsing and reconstruction, are integrated to generate a structured demand profile data object. The text content of the structured demand profile data object is input into the encoder of the large model. By obtaining the aggregated output vector of the encoder, a high-dimensional demand vector representing the comprehensive semantics of the demand is generated.

4. The method according to claim 3, characterized in that, Combining the industry scenario information, and using a pre-set causal reasoning prompt template, the large language model is guided to perform root cause analysis of user needs based on specific industry knowledge, generating a list of inferred root causes, including: Construct a causal reasoning prompt template, which is a text structure containing industry variables and demand variables; The obtained industry scenario information is filled into the industry variable, and the preprocessed user requirement text is filled into the requirement variable to form a complete prompt instruction; Input the complete prompt command into the large model; Receive and parse the text response returned by the large model, and extract a structured list of root causes from it.

5. The method according to claim 1, characterized in that, The large model performs deep semantic understanding on resource information in the supply-side resource repository and generates deep semantic tags. Based on the deep semantic understanding results and the deep semantic tags, resource vectors are generated, including: Integrate structured attribute data and unstructured descriptive text of resources from the supply-side resource pool; By using a preset value proposition extraction prompt template, the large model is guided to extract the core business problem solved by the resource from the unstructured description text of the resource, and generate a value proposition summary based on the core business problem. Based on a predefined industrial domain labeling system, the large model is used to attach deep semantic tags to resources, including applicable scenarios, pain points solved, and core capabilities. The structured attribute data of the resource, the value proposition summary, and the deep semantic tags are integrated into a complete semantic description text of the resource, and the text is encoded into a high-dimensional resource vector by the encoder of the large model.

6. The method according to claim 1, characterized in that, The semantic similarity between the demand vector and the resource vector is calculated. A multi-factor weighted decision is then made by combining the semantic similarity with scenario suitability, resource reputation, and commercial feasibility calculated based on the demand profile and a pre-built industrial knowledge graph, to generate a resource recommendation list, including: Calculate the cosine similarity between the demand vector and each resource vector, and use it as the semantic similarity. Based on a pre-built industrial knowledge graph, the correlation strength between industry scenarios in the demand profile and applicable scenarios corresponding to resource vectors is queried, and reasoning is performed in combination with business rules to generate scenario suitability scores. Based on the historical market data corresponding to the resource vector, and integrating its recent click-through rate, inquiry volume, user rating and comment sentiment analysis results, a comprehensive credibility score is calculated through a time decay function; The key constraints in the demand profile are compared with the business attributes corresponding to the resource vector. First, hard rules of budget and region are applied for filtering. Then, the Gaussian function is used to calculate the compliance of the soft constraints of delivery cycle to generate a business feasibility score. Configurable weights are assigned to semantic similarity, scene adaptability score, resource reputation score and business feasibility score respectively, and the comprehensive recommendation score corresponding to each resource vector is calculated by weighted summation formula; Based on the comprehensive recommendation score, all resource vectors are sorted to generate a final recommendation list, and a natural language explanation describing the reasons for the recommendation is automatically generated for each resource item in the list.

7. The method according to claim 6, characterized in that, Configurable weights are assigned to semantic similarity score, scene fit score, comprehensive credibility score, and business feasibility score, including: The platform business lifecycle is pre-divided into multiple consecutive stages, and a unique set of initial weight factor values ​​is configured for each stage. The platform business lifecycle includes at least the initial stage of system launch, the platform growth stage, and the platform maturity stage. In the initial stage of system launch, the highest weight is assigned to the semantic fit score. In the platform growth stage, the business feasibility score and resource credibility score are assigned higher weights than in the initial stage of system launch. In the platform maturity stage, a dynamic weight allocation mode based on user profile segmentation or real-time A / B testing is adopted. Predefine quantitative operational indicator thresholds corresponding to each stage, including but not limited to the total number of registered enterprise users, average daily resource retrieval volume, and monthly service transaction volume; periodically monitor the real-time data of the quantitative operational indicators; compare the real-time data with the indicator thresholds to determine the current business lifecycle stage and activate the corresponding weight configuration strategy. Collect user feedback behavior data on recommended resources; based on the feedback behavior data, calculate and update the set of initial values ​​of the optimal weight factors for the next period through a regression analysis model.

8. A digital service resource supply and demand matching system, characterized in that, include: The demand acquisition module is used to acquire digital service demand information from demanders through the demand-side interaction module, and to perform multimodal data fusion and standardization processing on the demand information. The requirement parsing module is used to perform deep semantic parsing on the processed requirement information through a large model, generate a structured requirement profile, and encode the structured requirement profile into a high-dimensional requirement vector. The supply analysis module is used to perform deep semantic understanding of resource information in the supply-side resource library through the large model, generate deep semantic tags, and generate resource vectors based on the deep semantic understanding results and the deep semantic tags. The similarity calculation module is used to calculate the semantic similarity between the demand vector and the resource vector. It combines the semantic similarity with the scenario adaptability, resource reputation and commercial feasibility calculated based on the demand profile and the pre-built industrial knowledge graph to make a multi-factor weighted decision and generate a resource recommendation list. The list display module is used to display the resource recommendation list to the requester through the service management platform.

9. A digital service resource supply and demand matching device, characterized in that, include: The memory is used to store the supply and demand matching program for digital service resources; A processor, configured to implement the steps of the digital service resource supply and demand matching method as described in any one of claims 1-7 when executing the digital service resource supply and demand matching program.

10. A computer-readable storage medium storing a computer program, characterized in that, The readable storage medium stores a digital service resource supply and demand matching program, which, when executed by a processor, implements the steps of the digital service resource supply and demand matching method as described in any one of claims 1-7.

Images