A method and device for matching supply and demand of supply service resources based on text features

Abstract

The present application provides a method and device for matching supply and demand of supply service resources based on text features. The method provided by the present application includes: obtaining multiple topics from the demand description text of the supply service resource, matching the description type of the supply service resource; obtaining description data corresponding to each description type, and obtaining multiple supply service resource samples by combination; extracting the first feature of each supply service resource sample and the second feature of the demand description text based on a text representation model, and determining a first category of supply service resource samples based on the similarity between the first feature and the second feature; generating a candidate demand set based on the first category of supply service resource samples; extracting the third feature of each candidate enterprise demand description text based on the text representation model, and determining a second category of supply service resource samples based on the second feature and the third feature; and determining the optimal supply service resource based on the matching degree between the second category of supply service resource samples and the demand description text.

Description

Technical Field

[0001] The present application relates to the field of natural language processing technology, and in particular to a method and device for matching supply and demand of supply service resources based on text features. Background Art

[0002] In modern supply chain management, matching supply service resources is a critical step in ensuring efficient supply chain operations. Supply service resources encompass basic supplier information, product specifications, and potentially related patents. Accurate and efficient matching of supply service resources helps companies quickly identify suppliers and product resources that meet their needs, thereby shortening procurement cycles, reducing costs, and improving overall supply chain efficiency. As supply chains grow in scale, rapidly identifying the optimal supply service resources is a major challenge facing modern supply chain management.

[0003] Traditional supply and service resource matching methods rely primarily on manual review and matching. Specifically, demand description texts are typically manually parsed and compared, with suppliers' basic information, product specifications, and related patent documents reviewed to identify matching resources. With the recent development of natural language processing (NLP) technology, some automated matching systems have emerged, using keyword matching or simple text similarity calculations for preliminary matching. These systems provide partial automation by calculating the similarity between demand description text and supply resource text. However, current matching methods still suffer from the following major drawbacks: Traditional manual matching is time-consuming and labor-intensive, making it difficult to scale with large-scale supply and service resource databases. Existing methods based on keywords or simple similarity calculations, while offering some efficiency improvements, still struggle to rapidly process massive amounts of information. Manual matching relies on human experience and is prone to bias. Automated systems often rely on keyword matching or shallow semantic analysis, struggling to understand the context and intent of user needs and the complex semantics of supply and service resources, resulting in limited relevance and accuracy in matching results. Existing general-purpose large language models lack a deep understanding of supply chain domain context and struggle to accurately capture supplier backgrounds, product characteristics, and their actual relevance to demand, leading to incomplete or inaccurate results.

[0004] Therefore, there is an urgent need for a method that can accurately extract demand information by semantically parsing and reasoning the demand description text, and combine the characteristics of supply service resources to use a large model to quickly, accurately and reliably match supply service resources, thereby improving the quality and efficiency of supply service resource matching. Summary of the Invention

[0005] In view of this, the present application provides a method and device for matching supply and demand of supply service resources based on text features, which is used to accurately extract demand information by semantically parsing and reasoning the demand description text, and combines the characteristics of the supply service resources to use a large model to quickly, accurately and reliably match the supply service resources, thereby improving the quality and efficiency of supply service resource matching.

[0006] Specifically, this application is implemented through the following technical solutions:

[0007] In a first aspect, the present application provides a method for matching supply and demand of provisioning service resources based on text features, the method comprising:

[0008] Obtain multiple topics from the demand description text of the supply service resource, matching the description type of the supply service resource;

[0009] Obtaining description data corresponding to each description type, combining multiple description data of each provisioning service resource as a sample, and obtaining multiple provisioning service resource samples;

[0010] extracting a first feature of each of the supply service resource samples and a second feature of the demand description text based on a text representation model, screening the plurality of supply service resource samples based on similarity between the first feature and the second feature, and determining a first category of supply service resource samples; the text representation model being arranged in a computer;

[0011] generating a candidate demand set according to each supply service resource sample in the first category of supply service resource samples;

[0012] extracting a third feature of each candidate enterprise demand description text in the candidate demand set based on the text representation model, screening the first category supply service resource samples based on each second feature and each third feature, and determining the second category supply service resource samples;

[0013] An optimal provisioning service resource is determined based on the matching degree between the second-category provisioning service resource sample and the demand description text.

[0014] A second aspect of the present application provides a device for matching supply and demand of supply service resources based on text features, the device comprising an acquisition module, an extraction module, a generation module, and a determination module;

[0015] The acquisition module is used to acquire multiple topics from the demand description text of the supply service resource and match the description type of the supply service resource;

[0016] The acquisition module is further configured to acquire description data corresponding to each description type, combine multiple description data of each provisioning service resource as a sample, and obtain multiple provisioning service resource samples;

[0017] The extraction module is configured to extract a first feature of each of the supply service resource samples and a second feature of the demand description text based on a text representation model, filter the plurality of supply service resource samples based on similarity between the first feature and the second feature, and determine a first category of supply service resource samples; the text representation model is disposed in a computer;

[0018] The generating module is configured to generate a candidate demand set based on each supply service resource sample in the first category of supply service resource samples;

[0019] The determining module is configured to extract a third feature of each candidate enterprise demand description text in the candidate demand set based on the text representation model, screen the first category supply service resource samples based on each second feature and each third feature, and determine the second category supply service resource samples;

[0020] The determining module is further configured to determine the optimal provisioning service resource based on the matching degree between the second-category provisioning service resource sample and the demand description text.

[0021] The supply and demand matching method and device of supply service resources based on text features provided by this application gradually optimizes the accuracy and efficiency of supply service resource matching through two screenings. Each screening is aimed at different objects. The first screening starts from the demand description text and matches the demand with the sample. The second screening starts from the matched sample and generates the most matching predicted demand based on the sample. The most matching sample is determined by matching the demand with the demand. Different strategies are adopted to avoid the insufficient model matching accuracy caused by the lack of feature extraction and model matching capabilities. It has hierarchical and collaborative properties, and improves the accuracy of demand and service resource matching. First, the first screening is for all supply service resource samples. Based on the first feature of all supply service resource samples and the second feature of the demand description text, the semantic similarity calculation is used to quickly screen out samples with basic relevance to the demand description text, reducing the screening scope and computing cost, and reducing the sample set from the global scale to the preliminary relevant sample set. At the same time, the combination of fusion features and auxiliary features is used to ensure that the results of the preliminary screening take into account both overall semantic relevance and multi-dimensional detail matching, laying the foundation for screening accuracy. Building on the first screening, the second screening process further refines the first-category supply service resource samples obtained in the first screening. This second screening generates a set of candidate requirements based on the first-category supply service resource samples. This process then performs deep semantic matching, leveraging the characteristics of the candidate requirements set with those of the requirement description text. This dynamic generation of candidate requirements sets allows for a more accurate and precise match of the semantics of enterprise requirements, supplementing semantic details that may have been missed in the first screening. By leveraging the multidimensional semantic features of the candidate requirements set to refine the description of the resource samples, this step improves matching accuracy and avoids the interference of highly relevant but low-precision samples that can result from a single screening step. In summary, this two-step screening approach achieves both efficiency and accuracy. The first screening process reduces the sample size through preliminary filtering, addressing the computational cost of global screening. The second screening process dynamically generates and deeply matches the candidate requirements set, ensuring high-precision matching results. This overall design achieves progressive resource optimization and establishes an efficient and accurate matching mechanism between supply service resources and enterprise requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A flowchart of a method for matching supply and demand of provisioning service resources based on text features provided in Example 1 of the present application;

[0023] Figure 2 This is a structural diagram of the supply and demand matching device for supply service resources based on text features provided in Example 2 of the present application. DETAILED DESCRIPTION

[0024] Exemplary embodiments are described in detail herein, with examples illustrated in the accompanying drawings. When the following description refers to the drawings, identical numerals in different drawings represent identical or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with this application.

[0025] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms "a," "the," and "the" used in this application are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0026] It should be understood that although the terms first, second, third, etc. may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0027] Specific embodiments are given below to introduce the technical solutions of the present application in detail.

[0028] Figure 1 This is a flowchart of the supply and demand matching method for supply service resources based on text features provided in Example 1 of this application. Figure 1 The method provided in this embodiment may include:

[0029] S101. Acquire multiple topics from a demand description text of a provisioning service resource and match the description type of the provisioning service resource.

[0030] Specifically, demand description text refers to the natural language text content used by enterprises to describe their specific needs, including information such as the products and service types, functional characteristics, technical specifications, budgets, application scenarios, etc. The demand description text is used to clarify the specific needs of enterprises in the supply chain in order to match the most appropriate supply service resources.

[0031] Furthermore, the requirement description text contains multiple themes. Each theme is the main semantic unit or key element extracted from the requirement description text. A theme is an abstract representation that reflects the core information of the enterprise's requirements in the text content. A theme is the refinement and decomposition of the requirement description text, which helps the matching system convert the complex requirement description text into structured semantic information, facilitating subsequent matching with supply service resources. The multiple themes may include service objects, service contents, product-related themes, technical requirement themes, functional requirement themes, application scenario themes, budget or procurement restriction themes, time requirement themes, etc.

[0032] Specifically, the supply service resources refer to all the information and data related to the supply chain provided by the supplier. The supply service resources are used to help enterprises meet specific requirements, such as providing products or technical support that meet the requirements. The description types of the supply service resources are the classification of different attributes or information in the supply service resources, and each description type corresponds to a specific attribute or feature of the supply service resources. There are multiple description types, including basic supplier information, supply product specifications, and technical patent information. Among them, the basic supplier information includes the enterprise name, enterprise profile, business scope, honorary qualifications, address, and contact information, etc. The supply product specifications include the product name, product model, technical parameters, application scenarios, product function descriptions, etc. The technical patent information includes the patent name, patent number, patent technical description, patent application field, etc.

[0033] It should be noted that each theme of the requirement description text corresponds to supply service resources of different description types. The description types corresponding to different themes may be different, and the supply service resource data corresponding to different themes may be different.

[0034] In specific implementation, the requirement description text is obtained through relevant enterprise employees or from the enterprise management system, and the requirement description text is preprocessed to remove the stop words (such as "of", "and", etc. words without practical meaning) in the requirement description text, and the requirement description text is normalized and segmented. A pre-trained large language model is used to parse the preprocessed requirement description text, refine and decompose the requirement description text, identify the semantic structure in the requirement description text, and extract multiple themes from the requirement description text in combination with the semantic representation ability of the large prediction model. Furthermore, the extracted multiple themes are classified and filtered, and the similar themes are merged or refined into more distinguishable specific themes to ensure that the themes have clear semantic meanings. Finally, the extracted and processed themes are determined as the multiple themes of the requirement description text.

[0035] Furthermore, based on the attributes of supply service resources, multiple description types of supply service resources (such as basic supplier information, supply product specifications, and technical patent information) are predefined. Based on these defined description types, the specific attributes or information categories of the supply service resources corresponding to each description type are determined. The content corresponding to each topic in the demand description text is analyzed to determine the primary demand attributes expressed by each topic. Semantic matching is performed between the primary demand attributes expressed by each topic and the specific attributes of the supply service resources corresponding to each description type to determine the most relevant description type for each topic.

[0036] For example: If a topic involves the technical parameter requirements of a supply product, the matching description type is "Supply Product Manual"; if a topic involves the service capabilities of a supplier, the matching description type is "Supplier Basic Information"; if a topic involves a patent name, the matching description type is "Technical Patent Information".

[0037] Optionally, the method of obtaining multiple topics from the demand description text of the supply service resource and matching the description type of the supply service resource includes: performing semantic segmentation on the demand description text of the supply service resource to obtain multiple topics; obtaining the supply service history of each service object, and determining the supply service resource usage habits based on the supply service history; predicting the constraint conditions of the supply service resource based on the service content and the supply service resource usage habits; and matching the description type based on the constraint conditions.

[0038] In specific implementation, the large model first parses the requirement description text. After standard preprocessing steps, including text cleaning (stop word and punctuation removal), word segmentation, and syntactic analysis, the processed text is input into the large language model to generate a context-aware vector representation. Next, an embedding-based clustering algorithm (such as K-Means) is used to automatically segment the text into semantically similar subparagraphs. Each subparagraph represents a potential topic, such as "Service Target," "Service Content," "Budget Requirements," and "Delivery Time." For example, if the input requirement description is "My company needs to find a battery management system supplier for a new energy vehicle manufacturing project, requiring fast charging support and a delivery time of less than three months," it will be automatically divided into Topic 1: "Service Target: New Energy Vehicle Manufacturing Project," Topic 2: "Service Content: Battery Management System," Topic 3: "Functional Requirements: Fast Charging Support," and Topic 4: "Time Requirement: Less than Three Months."

[0039] Furthermore, for the extracted "service object" topic, supply service records for similar projects are retrieved by calling an internal historical database or a third-party API. For example, the database can be searched for past procurement records using the keywords "new energy vehicles" and "BMS projects" to obtain information such as historical suppliers, transaction prices, delivery cycles, and performance evaluations. Data analysis techniques are then used to extract features based on this historical supply service data. For example, the DBSCAN clustering algorithm can be used to perform unsupervised clustering of supplier attributes in supply records to identify usage habits. Frequency analysis can also be used to extract common supplier characteristics. Furthermore, for the extracted "service content" topic, a semantic analysis model (such as RoBERTa or T5) is used to extract key constraint terms, such as "fast charging function," "battery protection strategy," and "CAN bus interface compatibility." These key constraint terms are then combined with preference features extracted from historical data (such as supplier certification status and product technical parameters) to construct a feature vector, which is then input into the prediction model. Based on the input features, the prediction model outputs specific constraints for supply service resources, such as "the battery management system must support a fast charging current of ≥150A," "delivery time must not exceed 90 days," and "suppliers must be ISO9001 and ISO26262 certified." Finally, based on the generated constraints, a mapping relationship is established between topics and description types. For example, the functional constraint "fast charging function" is mapped to the supply product manual description type, qualification requirements are mapped to the supplier basic information description type, and technical detail requirements are mapped to the technical patent information description type. Using a semantic matching engine (such as the Siamese Network text matching model), the description type that best meets the constraints of each topic is identified in the supply resource data, achieving precise matching between topics and description types.

[0040] It's important to note that topics and description types correspond one-to-one. Each topic defines what information is described in the demand description text, such as who the initiator is, who the service target is, and what the specific content of the demand is. Each description type describes the performance of the supply service resource in relation to that topic. For example, a semantic fragment in the demand description text describes who the initiator is, while the corresponding type of supply service resource describes the attributes that are more suitable for the type of customer. The descriptive data under the corresponding description type specifically describes the type of customer this supply service resource is more suitable for. In other words, each supply service resource sample includes multiple description data, each description data corresponds to a description type, and the number of description types in the supply service resource sample is consistent with the number of topics in the demand description text.

[0041] S102: Acquire description data corresponding to each description type, combine multiple description data of each provisioning service resource as a sample, and obtain multiple provisioning service resource samples.

[0042] Specifically, each supply service resource sample includes descriptive data for the supply service resource under each description type. In other words, each supply service resource sample includes supply service resource data under each description type. Based on the above description, each supply service resource sample includes supply service resource data corresponding to the supplier's basic information, supply service resource data corresponding to the supply product manual, and supply service resource data corresponding to the technical patent information.

[0043] During implementation, based on the determined description type, collect the corresponding supply service resource data for all description types. For each description type, collect the corresponding supply service resource data. Combine the supply service resource data corresponding to each description type into a complete supply service resource sample. Repeat the above steps to gradually collect and construct multiple supply service resource samples, ensuring that each supply service resource sample covers supply service resource data for all description types.

[0044] For example, for the description type of "supplier basic information", the supply service resource data such as the supplier's company profile and business scope are obtained; for the description type of "supply product manual", the supply service resource data such as product specifications and performance indicators are obtained. Each supply service resource sample contains supply service resource data of multiple description types such as supplier basic information, supply product manuals, and technical patent information. Each supply service resource sample can be represented as , wherein the To provide a sample of service resources, For the basic information of the supplier, For supply product instructions, the For technical patent information.

[0045] S103: extracting a first feature of each of the supply service resource samples and a second feature of the demand description text based on a text representation model, screening the multiple supply service resource samples based on similarity between the first feature and the second feature, and determining a first category of supply service resource samples.

[0046] Specifically, the first feature includes a first fused feature and a plurality of first auxiliary features corresponding to respective description data. The second feature includes a second fused feature and a plurality of second auxiliary features corresponding to respective topics.

[0047] The first fusion feature is the fusion feature of the supply service resource sample, and each first auxiliary feature is the feature corresponding to a single descriptive data of the supply service resource sample. That is, the first fusion feature is the fusion feature of supply service resource data of multiple descriptive types. The first fusion feature is a single feature, and the first auxiliary feature is a single feature of supply service resource data of each descriptive type. The first auxiliary features are multiple, and the number of first auxiliary features is the same as the number of descriptive types. The same descriptive data has different information content in the first fusion feature and the first auxiliary features.

[0048] In a specific implementation, the first feature of each supply service resource sample is extracted based on the text representation model, including: extracting multiple features of the supply service resource data of different description types in each supply service resource sample at a first depth based on the text representation model; extracting multiple features of the supply service resource data of different description types in each supply service resource sample at a second depth based on the text representation model, and determining them as multiple first auxiliary features; the second depth is greater than the first depth.

[0049] Specifically, a text representation model is a model based on natural language processing (NLP) technology that converts text data into machine-processable numerical feature representations. Common text representation models include large models based on deep learning, such as BERT (Bidirectional Encoder Representations from Transformers), GPT (Generative Pre-trained Transformer), and RoBERTa. Text representation models generate high-quality text vector representations by understanding linguistic information such as semantics and context. They support feature extraction at different depths (such as first and second depths), enabling them to capture semantic information from coarse to fine granularity, supporting relationship modeling between descriptive data types.

[0050] Furthermore, the text representation model is deployed on a computer. During feature extraction, the text representation model typically relies on a large model with a large number of parameters and extremely high computational requirements, placing high demands on computer hardware. Specifically, computer hardware primarily relies on high-performance computing resources to support the efficient operation of large models, including but not limited to key components such as graphics processing units (GPUs), tensor processing units (TPUs), and high-bandwidth memory (such as HBM). GPUs, with their powerful parallel computing capabilities, can accelerate matrix and vector operations in deep learning models and are currently the mainstream platform for inference and training of large models. TPUs are specialized hardware optimized for deep neural networks, further improving computational efficiency. Furthermore, the CPU coordinates system operations and handles non-parallel tasks. High-speed cache and memory ensure high-speed data read / write and temporary storage during inference and training of large-scale models. Hard disk systems are used to store model weights and training data. Therefore, the efficient operation of large models relies not only on advanced algorithm design but also on the comprehensive performance of the underlying computer hardware.

[0051] In implementation, the text representation model is deployed on a server, whose central processing unit coordinates the scheduling of various hardware components, including task distribution, memory management, and data preprocessing. Subsequently, the core computational tasks of the large model are offloaded to graphics processing units (GPUs) or tensor processing units (TPUs). These computing accelerators operate with a highly parallel architecture, using numerous computing cores to simultaneously execute neural network operations such as matrix multiplication, activation functions, and self-attention mechanisms. For example, in the Transformer architecture, numerous vector dot products and weight-matrix multiplications are broken down into smaller tasks, each executed in parallel by hundreds or thousands of GPU cores, accelerating forward and backward propagation. During computation, high-speed caches and high-bandwidth graphics memory are used to quickly access intermediate activation values, weight parameters, and gradient information, ensuring efficient data flow between computing units without becoming a bottleneck. A memory controller manages data read and write operations between main memory and graphics memory, optimizing data scheduling across different storage hierarchies. Some platforms also employ data pipelining and tensor parallelism to split the model's computational layers or tensor dimensions across multiple GPUs. Finally, the executed feature vectors are transferred from the accelerator back to the host system for output or further processing via the input and output interfaces controlled by the central processing unit. From the control logic of the central processing unit to the highly parallel computing of the graphics processing unit / tensor processing unit, and then to the efficient data transfer of memory, each hardware module works together through a clear division of labor and communication mechanisms to complete the operation of large models.

[0052] It's important to note that after training the text representation model, it can be used to extract features from the input text. The input data is a sample of supply service resources. Depending on the information content requirements of the output features, features are extracted from different layers. Once the model's parameters and structure are determined, the relationship between the model input and output is corresponding. In other words, there is a correspondence between the input supply service resource samples and the output features, a correspondence constrained by natural laws. Since text data inherently embodies stable semantic structures, linguistic logic, and contextual regularities, the text representation model learns and captures these linguistic regularities through a large corpus of data to achieve a vectorized representation of the input text. Therefore, each piece of input text is mapped to a fixed, high-dimensional semantic feature representation. This mapping is determined by both the model structure and the training data. In other words, the text representation model does not "create features out of thin air" but rather extracts features from the input text that conform to linguistic laws by modeling at the semantic, lexical, and syntactic levels of the language, subject to computational rules and model weights. Therefore, the content, structure, and semantics of the input text determine how the output features are expressed. This is a deterministic process that is jointly constrained by language laws, large model training mechanisms, and computing structures, and has a clear one-to-one correspondence.

[0053] Furthermore, the first depth is smaller than the second depth. The first depth focuses on extracting macroscopic global semantic features of the supply service resource samples, which is used to fuse features of different description types. The second depth focuses on extracting refined local semantic features, which is used to capture the differences between various description features.

[0054] In the specific implementation, first, the supply service resource data of different description types in each supply service resource sample is processed. Specifically, suppose there is a supply service resource sample that contains text data of two description types: "data analysis capability requirements" and "service time limit requirements." At the first depth, the text representation model will extract multiple feature vectors from the text of these two description types. The first depth focuses on extracting macro-global semantic features. That is, by understanding the text content of these description types, the text representation model can extract coarse-grained information of the entire text, such as the relationship between "data analysis capability" and "service time limit," reflecting the overall meaning of the text. For example, assuming that the text "needs to have data analysis capabilities and be able to deliver results within two weeks" is processed, the first depth feature vector generated by the text representation model through BERT may contain the following information: the service described in this text is a "data analysis" task with "time limit requirements." These feature vectors integrate the global semantics of the text and provide a basis for further matching and analysis.

[0055] Next, when processing other descriptive text types within the same supply service resource sample, the text representation model extracts even finer-grained feature vectors at the second depth. Feature extraction at the second depth prioritizes local semantic information, specifically designed to capture subtle differences between descriptive features. For example, after processing the "data analysis capability requirements" text at the second depth, more specific information may be extracted, such as requirements such as "ability to analyze big data," "proficiency in data modeling," and "familiarity with specific analytical tools." For the "service deadline requirements" text, the second depth feature might identify different deadline types, such as "urgent needs" and "routine needs," as well as their corresponding handling methods. These finer-grained feature vectors are referred to as multiple first auxiliary features, which reflect the specific differences between the various descriptive features within the text. Ultimately, the first fused features generated by combining the global features extracted at the first depth and the local features extracted at the second depth are used to describe the comprehensive characteristics of the entire supply service resource sample. These auxiliary features are then used to accurately match and screen resource samples. For example, if a sample supply service resource contains the text "five years of data analysis experience, able to deliver an analysis report within three months," the feature vector at the first depth might indicate that this is a data analysis resource requirement with a clear delivery timeline. The feature vector at the second depth can reveal the specific technical details of the resource sample, such as "proficiency in a specific analysis tool" and "delivery time of three months," providing more details for subsequent precise matching.

[0056] Optionally, the extracting the first feature of each of the supply service resource samples based on the text representation model includes: constructing a cross-feature constraint relationship based on the logical correlation between the supply service resource data of different description types; determining the first weight of the supply service resource data of different description types based on the matching degree between the supply service resource data of different description types and the demand description text; extracting the features of the supply service resource data of different description types based on the text representation model; performing weighted fusion of the features based on the first weight to obtain a first fused feature; and adjusting the first fused feature based on the cross-feature constraint relationship to obtain an adjusted first fused feature.

[0057] In specific implementation, supply service resource data is first preprocessed, and supply resource information is divided into different categories based on preset description types. Next, the logical relationships between different description types are analyzed. Specifically, a supply service resource logical relationship graph is constructed. Using knowledge graph technology, each description type is modeled as a node, and the business logic relationships between them are modeled as edges. For example, a positive correlation is established between the "Supplier Scale" node and the "Delivery Capability" node, and a dependency relationship is established between the "Product Specifications" node and the "Application Scenario" node. The strength of the dependency relationship is assigned based on historical business rules or expert experience (for example, a correlation weight of 0.8 indicates a high correlation). Then, using a large language model (such as BERT), the supply service resource data under each description type is encoded into a fixed-length semantic vector, and the demand description text is also encoded into a semantic vector. For each description type, the cosine similarity formula is used to calculate the similarity between its vector and the demand description text vector. The resulting similarity score is normalized, for example, using the Min-Max normalization method to linearly transform the similarity value to the range [0, 1]. The normalized score is the first weight of the corresponding description type.

[0058] Furthermore, based on the text representation model, data feature vectors are extracted for each description type. Taking the supply product manual as an example, the BERT model can be used to extract sentence- or paragraph-level deep semantic features (e.g., extracting the contextual relationships of the product function description and the entity attributes of the technical parameters), forming a feature matrix. The feature vectors of all description types are then weighted and fused according to a first weight. Specifically, the feature vectors extracted for each description type are multiplied by their corresponding weights, and all weighted feature vectors are concatenated or averaged to form the first fused feature of the supply service resource sample. For example, the supply product manual has a weight of 0.6, the supplier basic information has a weight of 0.3, and the technical patent information has a weight of 0.1. These corresponding features are weighted and fused to form a comprehensive representation. Subsequently, the first fused feature is further adjusted based on the previously constructed logical association knowledge graph. Specifically, if there is a strong positive correlation between two description types (e.g., supplier size and delivery capability), the correlation is enhanced in the fused feature. For example, a weighted linear combination of the two feature vectors is used before injecting the weighted linear combination into the fused feature. If a negative or conflicting relationship exists (such as "low price" and "high-end technical indicators"), an inhibitory factor is introduced into the feature fusion to reduce the contribution of the conflicting features. This is addressed through feature normalization and regularization methods (such as L2 regularization). If a descriptive feature exceeds a reasonable threshold after overall fusion (for example, if product specifications are exaggerated), normalization is applied to the first fused feature to ensure a reasonable distribution in the final feature space. Ultimately, through the combined processing of these methods, an adjusted first fused feature is obtained that considers the importance of each descriptive type and meets the cross-feature logical constraints.

[0059] The method provided in this embodiment, based on a feature fusion approach based on logical association, matching weighting, and feature adjustment, can effectively resolve conflicts and redundancies in information of different description types, enhance the accuracy and relevance of feature representation, thereby improving the matching degree between supply service resources and enterprise needs, and ultimately achieving the goal of optimizing matching results. First, cross-feature constraints are established based on the logical associations between supply service resource data of different description types, ensuring that features of different description types maintain consistency and validity during fusion. For example, the "price" and "specifications" of supply service resources are often interrelated. Reasonable cross-feature constraints can effectively avoid unreasonable conflicts or invalid combinations between these features, thereby improving the accuracy and representativeness of feature fusion. Second, weights are determined based on the matching degree between supply service resource data of different description types and the demand description text, enabling the feature fusion process to prioritize description types that are highly relevant to the demand text, which helps improve the model's matching accuracy with the target demand. Through weighted fusion, the proportion of features of different description types in the fused feature vector is appropriately adjusted based on their importance, thereby enhancing the ability of the final feature to reflect the true attributes of the supply service resource. Furthermore, by extracting features from each description type through a text representation model and then weighting and fusing these features, we can integrate the different information from each description type and avoid biasing the matching results due to a single feature. For example, a supply resource may match demand in terms of "specifications" but not in terms of "price." Using weighted fusion allows for comprehensive consideration of both aspects, resulting in a more comprehensive representation of the resource in the fused features. Finally, by adjusting the first fused feature based on cross-feature constraint relationships, the quality of the feature is further improved, ensuring that the fused feature is not only semantically consistent but also conforms to the actual logical relationships between the various description types. This adjustment effectively optimizes feature representation, making the final feature more accurate and more practical, thereby improving the effectiveness of the entire resource matching system.

[0060] The method provided in this embodiment effectively improves the accuracy and efficiency of supply service resource matching by extracting first fused features and first auxiliary features from supply service resource samples. The first fused features provide a global representation of supply service resources by averaging features extracted at the first depth for data of different description types. This condenses complex semantic information, simplifies feature dimensions, and reduces computational complexity, while ensuring overall semantic integrity. The first auxiliary features, on the other hand, extract multidimensional features at the second depth, retain detailed information from different description types, further enriching the hierarchical semantic representation and enabling each attribute of the supply service resource to be meticulously described and characterized. This comprehensive and detailed representation method can quickly filter out irrelevant samples during the supply service resource matching process while providing higher discrimination in refined screening, ensuring the accuracy and rationality of the final matching results. Together, the first fused features and the first auxiliary features form the basis for multi-level matching between supply service resource samples and demand description text. The first fusion feature serves as the core of the initial screening, quickly completing the global semantic matching of enterprise needs and resources, greatly improving the screening efficiency; while the first auxiliary feature further accurately locates the initial screening results in the subsequent stage by calculating the fine-grained similarity between description types. This design not only adapts to the diverse characteristics of enterprise needs from macro to micro, but also takes into account the computational efficiency and final effect of the matching process. In addition, the hierarchical extraction method of the first fusion feature and the first auxiliary feature in the text representation model extraction enables the matching system to achieve a comprehensive semantic understanding of supply service resources and enterprise needs from the global to the local, further enhancing the intelligence level and adaptability of the matching process.

[0061] Specifically, in combination with the above description, each topic of the demand description text corresponds to multiple description types of supply service resource data, that is, each description type of supply service resource data is determined according to the topic of the demand description text. There may be many description types. In this embodiment, only three description types are taken as examples (supplier basic information, supply product manual, and technical patent information) for introduction. Structured supply enterprise information refers to the basic information of suppliers or enterprises organized in an orderly and standardized manner, including industry, location, main products, honorary qualifications, enterprise scale, output, etc. Functional supply product information refers to detailed descriptions of supply products. These descriptions focus on the characteristics, functions and uses of the products, including name, model, main technical parameters, application scenarios, etc.

[0062] Similarly, the second feature includes a second fusion feature and multiple second auxiliary features corresponding to each topic. The second fusion feature is the fusion feature of each topic in the demand description text, and each second auxiliary feature is the feature of each topic in the demand description text, that is, the second fusion feature is the fusion feature of multiple topics, the second fusion feature is single, the second auxiliary feature is a single feature of each topic, and the second auxiliary feature is multiple. The first category of supply service resource samples refers to the supply service resource samples that are more compatible with the demand description text among multiple supply service resource samples.

[0063] In specific implementation, the text representation model is used to extract multiple feature vectors at a first depth for different topics in the requirement description text. The average of these feature vectors is calculated and used as the second fused feature. Furthermore, the text representation model is used again to extract multiple, more fine-grained feature vectors at a second depth for different topics in the requirement description text, generating multiple second auxiliary features.

[0064] Furthermore, the screening of the multiple supply service resource samples based on the similarity between the first feature and the second feature to determine the first category of supply service resource samples includes: calculating the first similarity between each first fusion feature and the second fusion feature; establishing a matching relationship between the description type and the subject; calculating the first auxiliary similarity between the first auxiliary feature and the second auxiliary feature based on the matching relationship; determining a second weight based on the degree of influence of the semantics of the text fragment corresponding to each subject in the demand description text on the comprehensive semantics of the demand description text; weighting the first auxiliary similarity based on the second weight to obtain the second similarity between the first auxiliary feature of each supply service resource sample and the second auxiliary feature of the demand description text; correcting the first similarity based on the second similarity; and screening the first category of supply service resource samples based on the size of the corrected first similarity.

[0065] Specifically, in combination with the above description, each description type of supply service resource data corresponds to a subject of a demand description text. Therefore, the first auxiliary feature of each description type of supply service resource data also corresponds to a subject of a demand description text.

[0066] It should be noted that the first preset number is not a fixed value but is dynamically variable. Its size is closely related to the demand description and the service content and timeline requirements in the demand description. Service content is a key factor in determining the number of service resource samples to be supplied. Service content includes the required expertise, complexity, and diversity. For highly specialized services, such as technical support or high-end manufacturing, more precise service resource samples are required, so the number of matching samples can be smaller, but the quality requirements for each sample are higher. For services with high diversity requirements, such as integrated services or distributed support, more samples are required to cover the needs of different scenarios. For complex service scenarios, when a single topic involves multiple related service contents, more resource samples are required to ensure comprehensive coverage. The more urgent the timeline, the greater the resource requirement to ensure sufficient options. Urgent service needs require more candidate resource samples to complete the service within a short period of time, allowing for rapid screening and adjustment. For more relaxed service needs, if the timeline requirements are more relaxed, the number of initial resource samples can be reduced, with a focus on precise matching.

[0067] Specifically, first, semantic analysis technology (such as BERT) is used to pre-process the demand description text, including operations such as removing stop words, word segmentation, and syntactic dependency analysis. Subsequently, cluster analysis or keyword extraction methods are used to extract the main semantic topics, such as "service content" and "time limit requirements." Each topic corresponds to a text fragment. After vectorization using BERT, the semantic weight of each topic fragment is determined by calculating the cosine similarity between it and the overall demand description text. After normalization, the proportion of each topic in the overall demand description is obtained. For example, for a demand description: "We need to purchase a batch of communication chips that support high-speed data transmission within two months for use in 5G base station equipment." The extracted topics include: "Communication chip procurement requirements" (service content topic, accounting for 70% of the overall text weight) and "two-month delivery cycle" (time limit requirement topic, accounting for 30% of the overall text weight).

[0068] After obtaining the topic, an initial value for the first preset quantity is determined based on the complexity and diversity of the service content. Specifically, complexity and diversity are measured by analyzing the number of key technical indicators, the density of professional terms, and the span of technical fields within the service content. For example, when the service content involves multiple fields (such as "communication chips," "5G protocols," and "high-speed interface standards") and is rich in technical details, both complexity and diversity are determined to be high, so a higher initial preset quantity (such as 100) is set. If the service content is relatively simple, such as "procurement of standard power modules," a lower initial value (such as 30) is set. Furthermore, timeline requirements are extracted, and the urgency of delivery is determined through a rule engine or keyword recognition (such as "urgent," "delivery within 2 months," etc.). An urgency level is established. For example, "delivery time ≤ 2 months" is considered highly urgent, requiring an increase of the first preset quantity by 20%; "delivery time 3-6 months" is considered moderately urgent, requiring the first preset quantity to remain unchanged; and "delivery time ≥ 6 months" is considered less urgent, requiring a decrease of the first preset quantity by 10%. Taking the above example, since "delivery within two months" is highly urgent, the initial value of 100 is increased by 20%, and the adjusted preset quantity is 120.

[0069] Finally, the adjusted initial values ​​for each theme are combined, the contributions of all themes are aggregated, and the weighted average is calculated to arrive at the final first preset number. For example, if the service content theme's contribution weight is 0.7, the adjusted sample size is 120; if the time limit requirement theme's contribution weight is 0.3, the adjusted sample size is 120. The combined preset number = 120 × 0.7 + 120 × 0.3 = 120. Ultimately, the first preset number for this demand matching process is determined to be 120 supply service resource samples, which serves as the initial number for subsequent sample screening.

[0070] The method provided in this embodiment dynamically determines the first preset quantity based on the service content and time limit requirements in the demand description text, which can effectively balance the matching accuracy and computational efficiency in the resource screening process. First, the basic quantity is set in combination with the complexity and diversity of the service content to ensure that the resource samples screened have the ability to cover all key needs; secondly, the time limit requirement is added as a dynamic adjustment factor so that the screening can adapt to the high-priority processing of urgent needs and avoid delays due to insufficient resources or waste due to redundant resources. At the same time, the dynamic adjustment mechanism can also improve the system's adaptability to different enterprise demand scenarios, provide a sufficient sample basis for subsequent detailed screening, and ensure that the screening results are both high-quality and broad, thereby improving the accuracy of the final match and the availability of the service.

[0071] In specific implementation, the above steps extract primary textual features from the requirement description text. For example, "technical support" is extracted as the service content and "urgent" as the time limit requirement from the requirement description text. Subsequently, for each supply service resource sample, relevant features are extracted from its description, such as the service type (e.g., "technical support") and the time limit requirement (e.g., "urgent"). The similarity between the first fused feature of each supply service resource sample and the second fused feature of the requirement description text is then calculated. For example, using the cosine similarity algorithm, a preliminary similarity score is calculated for each resource sample using "technical support" and "urgent" as keywords. Next, a matching relationship is established based on the first auxiliary feature of each resource sample (e.g., service type, field, etc.) and the corresponding topic in the requirement description text. For example, if the requirement description text mentions "technical support," all supply service resource samples related to "technical support" are clustered together. For these samples, the first auxiliary similarity is calculated by calculating the similarity between each first auxiliary feature and the second auxiliary feature. Suppose that "technical support" is mentioned in the demand text, and in a certain supply service resource sample, the first auxiliary feature of the sample is "computer technical support". Calculate the similarity between it and "technical support" in the demand text, for example, using word embedding technology to calculate the similarity between the two and obtain the corresponding first auxiliary similarity value.

[0072] After obtaining the first auxiliary similarity, the semantic weight of each topic in the requirement description is analyzed. For example, "technical support" may occupy a larger proportion in the requirement text, while time requirements (such as "urgent") may account for a smaller proportion. Semantic analysis techniques are used to calculate the weight of each topic, and the first auxiliary similarity is weighted based on this weight. For example, if "technical support" accounts for 70% of the importance of the requirement text and "urgent" accounts for 30%, the similarity is adjusted based on this ratio to obtain a weighted second similarity value. Based on this second similarity, the initial similarity is further revised. For example, if a supply service resource sample has a high similarity with the requirement text in the first step, but after weighting, it is found that the time requirement "urgent" is more important, the matching score of this resource sample is adjusted, and the initial similarity is revised. Finally, all supply service resource samples are sorted in ascending order based on the revised similarity value, ensuring that the resource sample that best meets the requirement is ranked first.

[0073] The method provided in this embodiment firstly calculates similarity, helping to quantify the degree of match between the supply service resources and the demand description text, making the matching results more accurate and measurable. By comparing the similarity between the first fused feature and the second fused feature, the present application can first grasp the overall semantic fit between the two main elements, thereby ensuring that the resource samples that are closest to the enterprise's needs are selected from the outset. Furthermore, calculating the similarity between each first auxiliary feature and the second auxiliary feature of the corresponding topic in the enterprise's needs further refines the matching evaluation and emphasizes the refined treatment of specific description types and topics. The weighted similarity calculation ensures that the importance of different description types in the overall matching is appropriately reflected, and the weights are dynamically adjusted based on the degree of semantic influence. This can prevent the influence of irrelevant or low-relevance features on the final results, thereby improving accuracy and relevance. Secondly, based on this similarity calculation, the most matching supply service resource samples can be effectively screened and sorted according to the similarity value, ensuring that the first-category supply service resource samples ultimately selected have the highest degree of fit with the enterprise's needs. By correcting similarities, the sample selection priority can be dynamically adjusted, making the matching process more flexible and accurate. Ultimately, this similarity calculation mechanism not only enhances the method's intelligence but also significantly improves the accuracy and efficiency of the matching process. It avoids the crude screening inherent in traditional methods, accurately reflects the specific requirements of enterprise needs, and provides high-quality resource matching results. These steps work together to ensure the optimal selection of supply service resources and meet enterprise needs.

[0074] S104: Generate a candidate demand set according to each supply service resource sample in the first category of supply service resource samples.

[0075] Specifically, the candidate demand set includes candidate enterprise demand description texts generated according to description data of various description types of a supply service resource sample.

[0076] In a specific implementation, the generation of a candidate demand set based on each supply service resource sample in the first category of supply service resource samples includes: determining a first prompt word of a large language model based on each supply service resource sample in the first category of supply service resource samples and feature extraction requirements; the large language model generates a plurality of initial segments based on the first prompt word and a first fusion feature of each supply service resource sample; the large language model generates a plurality of modified segments based on the first prompt word and a plurality of first auxiliary features of each supply service resource sample; matching a plurality of corresponding target modified segments based on the initial segment, processing the initial segment based on the plurality of target modified segments, generating a plurality of processed candidate demand texts, and obtaining a candidate demand set.

[0077] Specifically, the initial segment is a text segment generated by the large language model based on the first fused feature of the supply service resource sample. The initial segment is preliminary, basic text information used to describe the relevant features of the supply service resource sample and conduct preliminary matching. The modified segment is a text segment generated by the large language model based on the first auxiliary feature of the supply service resource sample. The modified segment further expands, supplements, and optimizes the initial segment. The modified segment further refines the features of the supply service resource sample and emphasizes its matching degree with enterprise needs or specific attributes.

[0078] In specific implementation, multiple descriptive data types are first extracted from historical service resource samples, including "service type," "service deadline," "resource availability," and "service area." Based on this descriptive data and the requirements submitted by the requester, "prompt words" are generated for the large language model. For example, if a requester requires cybersecurity technical support and the time limit is 24 hours, the generated prompt words might include "Provide cybersecurity technical support, with an urgent response and a 24-hour resolution. Service areas include cloud computing and cybersecurity, and are available 24 / 7." This prompt word, along with the fused features of each service resource sample, is then input into the large language model. Based on these inputs, the large language model generates multiple initial segments containing the basic description that matches the requirements. For example, the generated initial segment might be "Provide cybersecurity technical support services, ensuring a 24-hour response and resolution." Furthermore, based on multiple auxiliary features of each service resource sample (such as service advantages and historical success stories), the large language model generates multiple modified segments to supplement and expand upon the initial segment. The resulting modified snippet might be "We have a world-leading technical team that can quickly identify and resolve customers' cybersecurity issues" or "We provide 24 / 7 technical support to ensure that customers' issues are quickly responded to and resolved at any time."

[0079] Next, a semantic matching algorithm, such as cosine similarity or the BERT model, is used to calculate the semantic similarity between the initial and modified segments. Based on this similarity, the most suitable modified segment is selected to complement the initial segment. For example, the modified segment "We have a world-leading technical team that can quickly identify and resolve customers' cybersecurity issues" closely matches the initial segment "We provide cybersecurity technical support services, ensuring a 24-hour response and resolution." Therefore, it is selected as the target modified segment for processing. After combining the target modified segment with the initial segment, the text content is optimized to form a processed segment, such as "We provide efficient cybersecurity technical support services, ensuring a 24-hour response and resolution of all customer issues. Our technical team possesses world-leading technical capabilities, enabling us to quickly identify and resolve various cybersecurity issues, and provide 24 / 7 real-time monitoring and technical support." Finally, the processed segments are aggregated into a candidate demand set, which is then ranked according to the specific needs of the demander. Each segment in the candidate demand set contains precise service content and timeframe requirements, meeting the actual needs of the demander. This ultimately generates a sample set of service resources for the demander to select.

[0080] Optionally, semantic parsing is performed on each generated initial segment. Combining the structure and content of the initial segment, the target sub-segment corresponding to the modified segment in the initial segment is determined to ensure that the semantics of the sub-segment located in the initial segment are associated with the semantics of the modified segment. Furthermore, a deep semantic analysis is performed on each modified segment to extract its core information, including supplementary descriptions, limiting conditions, or additional details. Refined information of the modified segment is obtained through semantic feature extraction technology. The semantic content of the modified segment is integrated into the corresponding target sub-segment, and the refined information in the modified segment is integrated with the core content of the target sub-segment for semantic expansion or enhancement. A language generation model is then used to generate a new target sub-segment that is fluent and consistent with the contextual logic. The regenerated target sub-segment is replaced with the corresponding position in the original initial segment, maintaining the overall sentence structure intact while optimizing the expression integrity and semantic accuracy of the initial segment. Finally, all processed initial segments are spliced ​​or adjusted according to the coherence of the semantic content to generate multiple processed candidate requirement texts, forming a candidate requirement set.

[0081] The method provided in this embodiment, firstly, generates multiple initial and modified segments using a large language model when generating a candidate demand set. These segments construct a multi-faceted description of supply service resources based on different feature dimensions (such as fused features and auxiliary features). The initial segment provides a basic demand description, while the modified segments further refine the specific attributes and details of the description. This multi-layered generation process ensures a more complete and comprehensive demand description while avoiding the bias or incomplete information that may result from a single description approach. Secondly, the modified segments are used to continuously refine the initial segment, not only ensuring a more precise and fluent expression of the candidate demand but also increasing the contextual adaptability of the text, making it more tailored to specific enterprise needs. This processing can be customized based on the context of the actual demand, thereby optimizing matching quality and reducing the matching of irrelevant resources. Secondly, this method for generating a candidate demand set not only enhances the system's intelligence and adaptability but also provides a richer set of candidates for subsequent screening and matching. By using the candidate demand set as a basis for further screening (for example, by comparing similarity with the demand description text to determine the optimal supply service resource), supply service resources that are highly aligned with the actual demand can be more efficiently identified. In summary, this setup, based on generating and processing candidate demand sets, allows the system to be more flexible in describing supply resources, providing a variety of candidate texts that meet the requirements based on different input features (such as fused features and auxiliary features). This diversity provides a rich foundation for screening samples of the second category of supply service resources. Further similarity calculations can be used to more accurately select the optimal matching supply resources, further improving the overall system performance.

[0082] S105. Extract the third feature of each candidate enterprise demand description text in the candidate demand set based on the text representation model, screen the first category supply service resource samples based on each second feature and each third feature, and determine the second category supply service resource samples.

[0083] Specifically, combined with the above description, the third feature includes a third fusion feature and multiple third auxiliary features. The third fusion feature is the fusion feature of the demand description texts of each candidate enterprise in the candidate demand set, and each third auxiliary feature is the feature of the demand description text of each candidate enterprise in the candidate demand set.

[0084] In a specific implementation, a text representation model is used to extract multiple feature vectors at a first depth from each candidate enterprise demand description text in the candidate demand set, and the average of these feature vectors is calculated, which is then determined as a third fused feature. Furthermore, the text representation model is again used to extract multiple, more fine-grained feature vectors at a second depth from each candidate enterprise demand description text in the candidate demand set to obtain multiple third auxiliary features. A third similarity is calculated between the third fused feature of each candidate enterprise demand description text and the second fused feature of the demand description text; a corresponding topic is determined based on the description type corresponding to each third auxiliary feature, and a fourth auxiliary similarity is calculated between each third auxiliary feature and the second auxiliary feature of the corresponding topic. The fourth auxiliary similarity is weighted based on the third weight corresponding to each topic to obtain a fourth similarity between each candidate enterprise demand description text and the demand description text; the third similarity is corrected based on the fourth similarity; and each supply service resource sample in the first category of supply service resource samples is sorted in ascending order according to the magnitude of the corrected third similarity, and the supply service resource samples that are ranked at the top by a preset number are determined as second category supply service resource samples.

[0085] S106: Determine the optimal provisioning service resource based on the matching degree between the second-category provisioning service resource sample and the demand description text.

[0086] In a specific implementation, the method of determining the optimal supply service resource based on the matching degree between the second-category supply service resource sample and the demand description text includes: determining the supply service resource and the corresponding first confidence level by the large language model based on the first fusion feature of each supply service resource sample and the second fusion feature of the demand description text; determining each first fusion feature and the first auxiliary feature in the second-category supply service resource sample to establish a matching relationship between the description type and the subject; calculating the second auxiliary confidence level between the corresponding first auxiliary feature and the second auxiliary feature based on the matching relationship by the large language model; determining a fourth weight based on the degree of influence of the semantics of the text fragment corresponding to each subject in the demand description text on the comprehensive semantics of the demand description text; weighting the second auxiliary confidence level based on the fourth weight to obtain the second confidence level between the second-category supply service resource sample and the demand description text; revising the first confidence level based on the second confidence level; and determining the optimal supply service resource according to the size of the revised first confidence level.

[0087] Specifically, first, the first fusion feature of each supply service resource sample is extracted from the second category of supply service resource samples and compared with the second fusion feature of the demand description text. The input first fusion feature and second fusion feature are processed through the large language model, and the matching status of each supply service resource sample and the demand description text is returned, and a first confidence level is assigned to each sample.

[0088] Next, for each provisioning service resource sample, a matching relationship between description types and topics is established based on the description type corresponding to each first auxiliary feature and its relationship with the topics in the requirement description. For example, suppose a provisioning service resource sample contains descriptions of "data processing capability" and "service response time," each corresponding to a different description type. A matching relationship is established between each description type and a topic in the requirement description (such as "technical requirements" and "time requirements"). This process uses a large language model to match the first auxiliary features with the relevant parts of the requirement description (second auxiliary features), generating a second auxiliary confidence score for each pair of corresponding features. Based on this, the semantic influence of each topic is identified through in-depth analysis of the requirement description (using keyword extraction and topic modeling). For example, if the requirement description contains multiple topics, such as "functional requirements" and "time requirements," and analysis reveals that "time requirements" has a greater semantic influence within the entire text, then its semantic influence will be assigned a higher weight. This weight, known as the fourth weight, is used to adjust the second auxiliary confidence score. The fourth weight is combined with the second auxiliary confidence level to produce a weighted calculation to determine the second confidence level between each provisioning service resource sample and the requirement description. By correcting the first confidence level between the second-category provisioning service resource sample and the requirement description, a corrected first confidence level is obtained. This correction process adds the second confidence level to the first confidence level and adjusts the weights of the two using an algorithmic model generated by the large language model to form a final confidence level. This value reflects the degree of match between each provisioning service resource sample and the requirement description.

[0089] Finally, all supply service resource samples are sorted in ascending order based on the modified first confidence level. The top samples are considered the optimal supply service resource samples, and the top second preset number of supply service resource samples are selected from them. Ultimately, the sorted and filtered supply service resource samples are determined to be the optimal supply service resources that meet all requirements in the demand description.

[0090] Optionally, the fourth weight is determined based on the degree of influence of the semantics of the text fragment corresponding to each topic in the requirement description text on the comprehensive semantics of the requirement description text, including: extracting the comprehensive feature representation of the requirement description text and the feature representation of the text fragment corresponding to each topic based on the text representation model; using a semantic similarity calculation model to calculate the similarity between the feature representation of the text fragment corresponding to each topic and the comprehensive feature representation of the requirement description text, and determining the similarity as the initial weight; based on the semantic correlation between each topic and other topics, adjusting the initial weight, and determining the adjusted weight as the weight corresponding to each topic.

[0091] In specific implementation, the text representation model is first used to process the requirement description text to extract its comprehensive feature representation, namely, the semantic vector of the entire text. For example, suppose the requirement description text is "We need an expert with strong data analysis skills who can provide a data report within two weeks and train the team." During this process, the text representation model converts the text into a vector representing its overall semantics. Next, the text segments corresponding to each topic in the requirement description text are separated. For example, the above text can be divided into two topic segments: "Expert with strong data analysis skills" and "Able to provide a data report within two weeks and train the team." Each segment is processed using the same text representation model to extract its semantic vector representation. Based on this, a semantic similarity calculation model (such as cosine similarity or Euclidean distance) is used to calculate the similarity between the feature representation of each text segment corresponding to the topic and the comprehensive feature representation of the requirement description text. For example, suppose the similarity between the semantic vector of the first segment and the semantic vector of the entire requirement description text is 0.85, and the similarity between the semantic vector of the second segment and the semantic vector of the entire requirement description text is 0.90. The resulting similarity value serves as the initial weight for each topic segment, indicating the strength of each topic segment's semantic relevance to the overall text. This initial weight reflects the importance of each topic in the requirements description. To further optimize the weights, consider the semantic relevance between topics. Suppose the first topic segment, "Experts with strong data analysis skills," and the second topic segment, "Able to provide data reports and train the team within two weeks," have some semantic overlap. In particular, both are related to "data analysis skills" and "training skills." To this end, the semantic similarity between the topic segments can be recalculated to quantify their relevance. For example, a semantic similarity of 0.75 between the two segments indicates a strong relevance. Based on this analysis, a weight adjustment function (e.g., a weighting factor based on semantic relevance) is set to adjust the initial weights. If the semantic relevance between two topic segments is high, their weights can be appropriately increased, indicating that these segments have a greater impact on the overall semantics of the requirements description. For example, if the initial weights are 0.85 and 0.90, the adjusted weights might be 0.95 and 1.00, respectively, if the semantic relevance is strong. Ultimately, the adjusted weight will serve as the weight of the influence of the text fragment corresponding to each topic on the comprehensive semantics of the demand description text.

[0092] Corresponding to the aforementioned embodiment of a method for matching supply and demand of provisioning service resources based on text features, the present application also provides an embodiment of an apparatus for matching supply and demand of provisioning service resources based on text features.

[0093] Figure 2This is a structural diagram of the supply and demand matching device for supply service resources based on text features provided in Example 2 of this application. Figure 2 The device provided in this embodiment includes an acquisition module 210, an extraction module 220, a generation module 230 and a determination module 240;

[0094] The acquisition module 210 is configured to acquire multiple topics from the demand description text of the provisioning service resource, matching the description type of the provisioning service resource;

[0095] The acquisition module 210 is further configured to acquire description data corresponding to each description type, combine multiple description data of each provisioning service resource as a sample, and obtain multiple provisioning service resource samples;

[0096] The extraction module 220 is configured to extract a first feature of each of the supply service resource samples and a second feature of the demand description text based on a text representation model, filter the plurality of supply service resource samples based on similarity between the first feature and the second feature, and determine a first category of supply service resource samples; the text representation model is disposed in a computer;

[0097] The generating module 230 is configured to generate a candidate demand set based on each provisioning service resource sample in the first category of provisioning service resource samples;

[0098] The determining module 240 is configured to extract a third feature of each candidate enterprise demand description text in the candidate demand set based on the text representation model, and screen the first category supply service resource samples based on each second feature and each third feature to determine the second category supply service resource samples;

[0099] The determining module 240 is further configured to determine an optimal provisioning service resource based on a matching degree between the second-category provisioning service resource sample and the demand description text.

[0100] The device of this embodiment can be used to perform Figure 1 The steps, specific implementation principles and implementation processes of the method embodiment shown are similar and will not be repeated here.

[0101] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0102] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present application scheme. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0103] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for matching supply and demand of supply service resources based on text features, characterized in that: The method comprises: Obtain multiple topics from the demand description text of the supply service resource, matching the description type of the supply service resource; Obtaining description data corresponding to each description type, combining multiple description data of each provisioning service resource as a sample, and obtaining multiple provisioning service resource samples; extracting a first feature of each of the supply service resource samples and a second feature of the demand description text based on a text representation model, screening the plurality of supply service resource samples based on similarity between the first feature and the second feature, and determining a first category of supply service resource samples; the text representation model being arranged in a computer; generating a candidate demand set according to each supply service resource sample in the first category of supply service resource samples; extracting a third feature of each candidate enterprise demand description text in the candidate demand set based on the text representation model, screening the first category supply service resource samples based on each second feature and each third feature, and determining the second category supply service resource samples; An optimal provisioning service resource is determined based on the matching degree between the second-category provisioning service resource sample and the demand description text.

2. The method according to claim 1, characterized in that The first feature includes a first fusion feature and a plurality of first auxiliary features corresponding to respective description data; The second feature includes a second fusion feature and a plurality of second auxiliary features corresponding to respective topics.

3. The method according to claim 2, characterized in that The extracting the first feature of each of the provisioning service resource samples based on the text representation model includes: Based on the logical association between supply service resource data of different description types, cross-feature constraint relationships are constructed; Determining first weights of the supply service resource data of different description types based on the matching degree between the supply service resource data of different description types and the demand description text; Extract the features of supply service resource data of different description types based on text representation model; Performing weighted fusion on the features based on the first weight to obtain a first fused feature; The first fused feature is adjusted based on the cross-feature constraint relationship to obtain an adjusted first fused feature.

4. The method according to claim 2, characterized in that The screening of the plurality of provisioning service resource samples based on the similarity between the first feature and the second feature to determine a first category of provisioning service resource samples includes: Calculating a first similarity between each first fused feature and the second fused feature; Establish a matching relationship between description type and topic; calculating a first auxiliary similarity between the first auxiliary feature and the second auxiliary feature according to the matching relationship; Determining a second weight based on the influence of the semantics of the text fragment corresponding to each topic in the demand description text on the comprehensive semantics of the demand description text; Weighting the first auxiliary similarity based on the second weight to obtain a second similarity between the first auxiliary feature of each supply service resource sample and the second auxiliary feature of the demand description text; modifying the first similarity based on the second similarity; The first category of supply service resource samples is screened according to the size of the corrected first similarity.

5. The method according to claim 1, characterized in that Generating a candidate demand set according to each supply service resource sample in the first category of supply service resource samples includes: determining a first prompt word of a large language model based on each supply service resource sample in the first category of supply service resource samples and feature extraction requirements; The large language model generates a plurality of initial segments based on the first prompt word and the first fused feature of each provisioning service resource sample; The large language model generates a plurality of modified segments based on the first prompt word and a plurality of first auxiliary features of each supply service resource sample; Based on matching the initial segment with corresponding multiple target modified segments, the initial segment is processed based on the multiple target modified segments to generate multiple processed candidate requirement texts, and a candidate requirement set is obtained.

6. The method according to claim 1, characterized in that The extracting the first feature of each of the provisioning service resource samples based on the text representation model includes: Extracting multiple features of supply service resource data of different description types at a first depth in each supply service resource sample based on a text representation model; Based on the text representation model, multiple features of the supply service resource data of different description types in each supply service resource sample at a second depth are extracted and determined as multiple first auxiliary features; the second depth is greater than the first depth.

7. The method according to claim 2, characterized in that The determining of the optimal provisioning service resource based on the matching degree between the second-category provisioning service resource sample and the demand description text includes: The large language model determines the supply service resource and the corresponding first confidence level based on the first fused feature of each supply service resource sample and the second fused feature of the demand description text; Determine each first fusion feature and first auxiliary feature in the second-category supply service resource sample, and establish a matching relationship between the description type and the subject; The large language model calculates a second auxiliary confidence between the corresponding first auxiliary feature and the second auxiliary feature based on the matching relationship; Determining a fourth weight based on the influence of the semantics of the text fragment corresponding to each topic in the demand description text on the comprehensive semantics of the demand description text; Weighting the second auxiliary confidence level based on the fourth weight to obtain a second confidence level between the second type of provisioning service resource sample and the demand description text; Modifying the first confidence level based on the second confidence level; The optimal provisioning service resource is determined according to the magnitude of the corrected first confidence level.

8. The method according to claim 7, characterized in that The determining of the fourth weight based on the influence of the semantics of the text fragment corresponding to each topic in the demand description text on the comprehensive semantics of the demand description text includes: Extract comprehensive feature representation of the requirement description text and feature representation of the text fragment corresponding to each topic based on the text representation model; Calculate the similarity between the feature representation of the text segment corresponding to each topic and the comprehensive feature representation of the demand description text using a semantic similarity calculation model, and determine the similarity as an initial weight; Based on the semantic relevance between each topic and other topics, the initial weight is adjusted, and the adjusted weight is determined as the weight corresponding to each topic.

9. The method according to claim 1, characterized in that Get multiple topics from the requirement description text of the provisioning service resource, matching the description type of the provisioning service resource, including: Perform semantic segmentation on the demand description text of the supply service resources to obtain multiple topics; Acquiring the provisioning service history of each service object, and determining the provisioning service resource usage habits based on the provisioning service history; Predicting constraints on provisioning service resources based on service content and usage habits of the provisioning service resources; Description types are matched based on the constraints.

10. A device for matching supply and demand of supply service resources based on text features, characterized in that: The device includes an acquisition module, an extraction module, a generation module and a determination module; The acquisition module is used to acquire multiple topics from the demand description text of the supply service resource and match the description type of the supply service resource; The acquisition module is further configured to acquire description data corresponding to each description type, combine multiple description data of each provisioning service resource as a sample, and obtain multiple provisioning service resource samples; The extraction module is configured to extract a first feature of each of the supply service resource samples and a second feature of the demand description text based on a text representation model, filter the plurality of supply service resource samples based on similarity between the first feature and the second feature, and determine a first category of supply service resource samples; the text representation model is disposed in a computer; The generating module is configured to generate a candidate demand set based on each supply service resource sample in the first category of supply service resource samples; The determining module is configured to extract a third feature of each candidate enterprise demand description text in the candidate demand set based on the text representation model, screen the first category supply service resource samples based on each second feature and each third feature, and determine the second category supply service resource samples; The determining module is further configured to determine the optimal provisioning service resource based on the matching degree between the second-category provisioning service resource sample and the demand description text.

Images