Interpretable enterprise service matching method, system and device and storage medium

Abstract

The invention relates to the technical field of data processing, and particularly provides an interpretable enterprise service matching method, system and device and a storage medium, and the method comprises the steps: obtaining enterprise user demand data, generating a structured enterprise demand vector, and analyzing the potential demand intention of an enterprise; dynamically determining a recall strategy combination according to the intention, and recalling an initial candidate case set from a case library according to the recall strategy combination; performing semantic analysis on the candidate cases to obtain matching features of each case; fusing the enterprise demand vector, the potential demand intention and the matching feature, calculating an interpretable matching score, and generating a sorted recommendation list according to the interpretable matching score; and finally, based on the recall strategy combination, the scoring elements and the case attributes, generating an explanatory recommendation reason for the recommendation case. According to the method, the problems of cold start, information cocoon room and decision black box in traditional enterprise service matching are effectively solved.

Description

Explainable enterprise service matching methods, systems, devices, and storage media Technical Field

[0001] This invention belongs to the field of data processing technology, specifically relating to an interpretable enterprise service matching method, system, device, and storage medium. Background Technology

[0002] Existing intelligent matching platforms for enterprise services largely rely on static filtering using service provider tags and keywords submitted by enterprises, or on recommendations based on historical behavioral data such as collaborative filtering. These methods often have significant shortcomings when dealing with scenarios involving highly differentiated enterprise needs and complex, diverse service cases: first, they struggle to handle the "cold start" problem for newly registered enterprises; second, their matching dimensions are too singular, easily leading to "information cocoons" and a lack of cross-domain, insightful, and high-quality case recommendations; and third, the matching process is like a "black box," with opaque decision-making logic, making it difficult for enterprise users to understand the reasons for recommendations, thus reducing trust and willingness to adopt them. Therefore, there is an urgent need for an intelligent matching method that can deeply understand the dynamic intentions of enterprises, achieve a balance between accuracy and diversity, and is explainable. Summary of the Invention

[0003] In view of the above-mentioned shortcomings of the prior art, the present invention provides an interpretable enterprise service matching method, system, device and storage medium to solve the above-mentioned technical problems.

[0004] In a first aspect, the present invention provides an interpretable enterprise service matching method, comprising: acquiring enterprise user demand data; generating a structured enterprise demand vector based on the demand data; and analyzing the enterprise's potential demand intent based on the enterprise demand vector; determining a recall strategy combination for candidate service cases based on the potential demand intent; and recalling an initial candidate case set from a service case library based on the recall strategy combination; performing semantic analysis on the service cases in the initial candidate case set to obtain matching features for each service case; fusing the enterprise demand vector, the potential demand intent, and the matching features to calculate an interpretable matching score; and generating a sorted recommendation list based on the matching score; and generating explanatory recommendation reasons for the service cases in the recommendation list based on the recall strategy combination, matching score elements, and service case attributes.

[0005] In one optional implementation, acquiring enterprise user demand data and generating a structured enterprise demand vector based on the demand data includes: providing enterprise users with a standardized demand questionnaire through an interactive interface and collecting their submitted questionnaire answers; simultaneously, associating and acquiring the enterprise's registration information and initial behavior data on the platform; parsing and normalizing the demand data using an enterprise profile parser, mapping non-standardized descriptive information to predefined standard tags; and generating a structured demand vector composed of multiple standard tags based on the mapping results; wherein, the parsing and normalization process includes: converting industry descriptions in text form into national standard industry classification codes, and / or mapping personnel quantity ranges to preset enterprise size levels.

[0006] In one optional implementation, analyzing a company's potential demand intentions based on its demand vector includes: inputting the company's demand vector into a preset intention analysis model; using the intention analysis model, based on predefined intention classification rules and the company attributes and demand keywords contained in the company's demand vector, calculating the company's score weights on multiple preset intention dimensions to determine its potential demand intentions; the preset intention dimensions include at least two of the following: industry benchmarking intention, problem-solving intention, cost control intention, and exploratory intention; the intention classification rules are rule-based weight allocation logic, which at least includes: if the digital maturity attribute in the company's demand vector is beginner level, then increase the weights of industry benchmarking intention and exploratory intention; if the company size attribute in the company's demand vector is micro or small, then increase the weight of cost control intention.

[0007] In an optional implementation, based on the potential demand intent, a combination of recall strategies for candidate service cases is determined, and based on the combination of recall strategies, an initial set of candidate cases is recalled from the service case library, including: dynamically selecting at least one primary recall strategy and at least one secondary recall strategy from a pre-built multimodal recall strategy library based on the dominant potential demand intent, and configuring an exploratory recall strategy; based on preset recall quantity quotas at each level, sequentially executing the primary recall strategy, secondary recall strategy, and exploratory recall strategy to obtain a core candidate case subset, an extended candidate case subset, and an exploratory candidate case subset from the service case library, respectively; and prioritizing the inclusion of the core candidate case subset according to the quota. Cases in the candidate case subset are scored and weighted; cases in the expanded candidate case subset and the exploratory candidate case subset are supplemented according to quotas, wherein industry diversity control is applied to the inclusion of the exploratory candidate case subset to avoid excessive industry concentration; all included cases are deduplicated and initially sorted based on their own attribute scores and source level weights to form the initial candidate case set; wherein the multimodal recall strategy library includes at least three of the following strategies: industry semantic cluster recall strategy, problem-solution pattern matching strategy, enterprise profile similarity recall strategy, success pattern migration recall strategy, and time-series hotspot weighted recall strategy.

[0008] In an optional implementation, based on preset recall quotas for each level, the main recall strategy, the auxiliary recall strategy, and the exploratory recall strategy are executed sequentially, including: First, the main recall strategy is executed to obtain the core candidate case subset with high confidence matching; second, the auxiliary recall strategy is executed to perform semantic or attribute expansion on the core candidate case subset to obtain the extended candidate case subset; finally, the exploratory recall strategy is executed to obtain the exploratory candidate case subset with low relevance to the current demand intent but with inspiration based on diversity, novelty, or trend indicators.

[0009] In an optional implementation, the enterprise demand vector, the potential demand intent, and the matching features are fused to calculate an interpretable matching score, and a ranked recommendation list is generated based on the matching score. This includes: dynamically configuring the calculation weights corresponding to different feature dimensions in the matching score formula according to the weights of each intent dimension in the potential demand intent; calculating an initial matching score for each service case in the initial candidate case set based on the configured matching score formula; and performing hierarchical confidence correction on the initial matching score according to the confidence level at which each service case is recalled, to obtain the final interpretable match. Scoring; sorting all service cases according to the final interpretable matching score to generate the recommendation list; wherein, the calculation weight of the dynamically configured matching scoring formula includes: increasing the calculation weight of the industry matching feature dimension in the matching scoring formula when the weight of the industry benchmarking intent dominates in the potential demand intent; increasing the calculation weight of the business process matching feature dimension in the matching scoring formula when the weight of the problem-solving intent dominates in the potential demand intent; and increasing the calculation weight of the cost preference feature dimension in the matching scoring formula when the weight of the cost control intent dominates in the potential demand intent.

[0010] In an optional implementation, explanatory recommendation reasons are generated for service cases in the recommendation list based on the recall strategy combination, matching scoring elements, and service case attributes. This includes: selecting a corresponding basic explanation template from a preset reason template library based on the service case's source recall level; wherein the reason template library contains dedicated templates for different recall levels and different potential demand intentions; filling the selected basic explanation template with the key matching dimensions in the matching scoring elements and the attribute information of the service case to generate a preliminary reason fragment; and from the preliminary reason fragment, determining the recommendation reason based on the contribution of the matching scoring elements. Based on a predetermined priority rule, at least one key reason is selected and refined to form the final natural language explanatory recommendation reason. The predetermined reason template library includes at least: a first type of template for explaining service cases from the high-confidence core level, whose explanation logic focuses on successful practices in the same industry and targeted solutions to pain points; a second type of template for explaining service cases from the extended level, whose explanation logic focuses on cross-domain semantic similarity and solution transferability; and a third type of template for explaining service cases from the exploration level, whose explanation logic focuses on the novelty, trendiness, and diverse value brought by the solution.

[0011] Secondly, the present invention provides an interpretable enterprise service matching system, comprising: a demand acquisition module, used to acquire demand data of enterprise users, generate a structured enterprise demand vector based on the demand data, and analyze the enterprise's potential demand intent based on the enterprise demand vector; a case recall module, used to determine a recall strategy combination of candidate service cases based on the potential demand intent, and recall an initial candidate case set from a service case library based on the recall strategy combination; a semantic analysis module, used to perform semantic analysis on the service cases in the initial candidate case set to obtain matching features of each service case; a scoring calculation module, used to fuse the enterprise demand vector, the potential demand intent, and the matching features to calculate an interpretable matching score, and generate a sorted recommendation list based on the matching score; and a reason generation module, used to generate explanatory recommendation reasons for the service cases in the recommendation list based on the recall strategy combination, matching score elements, and service case attributes.

[0012] Thirdly, an apparatus is provided, comprising: a memory for storing an interpretable enterprise service matching program; and a processor for implementing the steps of the interpretable enterprise service matching method as provided in the first aspect when executing the interpretable enterprise service matching program.

[0013] Fourthly, a computer-readable storage medium is provided, on which an interpretable enterprise service matching program is stored, wherein when the interpretable enterprise service matching program is executed by a processor, it implements the steps of the interpretable enterprise service matching method provided in the first aspect.

[0014] The beneficial effects of this invention lie in the fact that the interpretable enterprise service matching method, system, device, and storage medium provided by this invention effectively solve the cold start, information cocoon, and decision-making black box problems in traditional enterprise service matching through dynamic intent analysis, multimodal fusion recall, and interpretable matching calculation. It can accurately understand the deep needs of enterprises, achieve intelligent discovery of high-quality cross-domain cases, and significantly improve the diversity and novelty of the results while ensuring high relevance of the recommendation results. The final generated explanatory recommendation reasons make the matching logic clear and transparent, greatly enhancing enterprise users' trust and decision-making efficiency, and significantly improving the platform's service intelligence level and user satisfaction. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 is a schematic flowchart of a method according to an embodiment of the present invention.

[0017] Figure 2 is a schematic block diagram of a system according to an embodiment of the present invention.

[0018] Figure 3 is a schematic diagram of the structure of a device provided in an embodiment of the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0021] The interpretable enterprise service matching method provided in this embodiment of the invention is executed by a computer device, and correspondingly, the interpretable enterprise service matching system runs on the computer device.

[0022] Figure 1 is a schematic flowchart of a method according to an embodiment of the present invention. The entity executing the process in Figure 1 can be an interpretable enterprise service matching system. Depending on different requirements, the order of the steps in this flowchart can be changed, and some steps can be omitted.

[0023] As shown in Figure 1, the method includes: S1. Obtaining enterprise user demand data, generating a structured enterprise demand vector based on the demand data, and analyzing the enterprise's potential demand intent based on the enterprise demand vector; S2. Determining a recall strategy combination for candidate service cases based on the potential demand intent, and recalling an initial candidate case set from the service case library based on the recall strategy combination; S3. Performing semantic analysis on the service cases in the initial candidate case set to obtain matching features for each service case; S4. Integrating the enterprise demand vector, the potential demand intent, and the matching features to calculate an interpretable matching score, and generating a sorted recommendation list based on the matching score; S5. Generating explanatory recommendation reasons for the service cases in the recommendation list based on the recall strategy combination, matching score elements, and service case attributes.

[0024] In one embodiment of the present invention, based on step S1, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0025] S101. Obtain enterprise user demand data and generate a structured enterprise demand vector based on the demand data. This includes: First, the system presents a streamlined and standardized demand questionnaire to enterprise users who have completed registration or actively triggered the matching process via a web page or mobile application front-end. This questionnaire typically contains about 10 multiple-choice questions, covering core dimensions such as industry attributes, enterprise size, digital maturity, current pain points, and service type preferences. Simultaneously with the user submitting the questionnaire answers, the system automatically links and extracts the enterprise's existing registration information (such as the industry category and staff size filled in the business registration information) and its initial behavior log data on the platform (such as recently viewed service category pages) through a backend interface. The questionnaire answers, registration information, and behavior logs together constitute the original, multi-source enterprise demand data.

[0026] Subsequently, this data is sent to the backend enterprise profile parser module for processing. This parser embeds a business rule engine and a data mapping table, performing key parsing and normalization operations. For example, it maps user-submitted text industry descriptions (such as "machine tool production") to standard industry codes (such as "C3421 metal cutting machine tool manufacturing") by querying the National Economic Industry Classification (GB / T4754) code library. Simultaneously, based on preset threshold rules, it maps the number of employees (such as 150 people) to discrete size level labels (such as "medium-sized (51-200 people)"). Similar standardized mappings are performed for other dimensions such as digital maturity and preference stages.

[0027] Ultimately, the enterprise profile parser outputs a structured enterprise demand vector V_q. This vector is a machine-readable collection of predefined standard labels, in the form of, for example: {Industry: "C3421 Metal Cutting Machine Tool Manufacturing", Size: "Medium", Maturity: "Early", Pain Points: ["Production Management", "Inventory Management"], Preference Type: "SaaS Software"}. This vector accurately and unambiguously depicts the enterprise's current demand status, providing a unified and structured input for subsequent intent analysis and case matching.

[0028] S102. Analyze the enterprise's potential demand intentions based on the enterprise demand vector, including: after obtaining the structured enterprise demand vector V_q, the system starts the intention analysis module. The core of this module is a preset intention analysis model, which is essentially a rule-based and lightweight statistical decision logic engine.

[0029] The model predefines several intent dimensions to characterize a company's potential motivations, typically including but not limited to: industry benchmarking intent (wanting to refer to successful practices in the same industry), problem-solving intent (seeking direct solutions to specific pain points), cost control intent (highly sensitive to price or return on investment), and exploratory intent (vague needs, hoping to gain innovative inspiration). Each intent dimension has an initial score weight of 0.

[0030] The model receives V_q as input and performs calculations based on a series of predefined intent classification rules. These rules reflect the company's expert knowledge and historical matching data, and are expressed as an "IF-THEN" weight allocation logic. For example: Rule 1: If the "Digital Maturity" attribute value in V_q is "Beginner," then the company is judged to likely lack implementation experience and tend to learn from benchmarks or explore possibilities. Therefore, the system will increase the weights of its "Industry Benchmarking Intent" and "Exploratory Intent" (e.g., by 0.3 points each).

[0031] Rule 2: If the "Enterprise Size" attribute in V_q is "Micro" or "Small", it is determined that the enterprise's budget may be relatively limited. Therefore, the system will significantly increase the weight of its "Cost Control Intent" (e.g., increase it by 0.4 points).

[0032] Rule 3: The model also analyzes the "current pain point" description in V_q or keywords in the free text of the questionnaire. If words such as "cost reduction" or "limited budget" are found, the weights of "cost control intention" and "problem-solving intention" will be further strengthened.

[0033] After calculation according to the above rules, the model outputs an intent weight dictionary, for example: {"Industry Benchmarking Intent": 0.7, "Problem-Solving Intent": 0.2, "Cost Control Intent": 0.1, "Exploratory Intent": 0.3}. The intent with the highest weight ("Industry Benchmarking Intent" in this example) will be identified as the dominant potential demand intent for the company at its current stage. This analysis provides a direct and quantitative basis for selecting subsequent dynamic recall strategies.

[0034] In one embodiment of the present invention, based on step S2, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0035] The intent weight dictionary is fed into the dynamic path generator module. This module performs intelligent strategy planning based on the dominant potential demand intent with the highest weight. The system maintains a multimodal recall strategy library, which contains various encapsulated recall algorithms. For example, the industry semantic cluster recall strategy: utilizes graph embedding technology of industry knowledge graphs to find industry clusters and related cases that are semantically similar to the target industry.

[0036] Problem-Solution Pattern Matching Strategy: Map enterprise pain point keywords to a standard problem classification system and recall historical cases of solving similar problems.

[0037] Enterprise profile similarity recall strategy: Calculate the weighted cosine similarity between enterprises based on multi-dimensional features such as industry, size, and maturity.

[0038] Successful pattern migration recall strategy: Identify and recommend benchmark cases across industries that have similar problem patterns.

[0039] Time-based hot topic weighted recall strategy: Prioritize cases with high recent page views, high inquiry rates, or strong industry trends.

[0040] The dynamic path generator dynamically selects and combines strategies from this library based on the dominant intent. For example, for a company dominated by "industry benchmarking intent," it might select "industry semantic cluster recall" and "success pattern migration recall" as the primary recall strategies, "company profile similarity recall" as the secondary recall strategy, and specify "time-series hotspot weighted recall" as the exploratory recall strategy. Simultaneously, it loads a preset recall quota configuration, for example: {"core circle": 50, "expansion circle": 30, "exploration circle": 20}, meaning a total of approximately 100 cases will be recalled, but belonging to different confidence levels.

[0041] Subsequently, the system executes a layered, progressive recall: First layer (core circle): First, the selected main recall strategy (such as the "industry semantic cluster recall" mentioned above) is executed to quickly filter out the cases with the highest matching degree and the strongest confidence from the case library, and generate a core candidate case subset until the quota (such as 50) is reached.

[0042] The second layer (expanded circle): Next, auxiliary recall strategies (such as "enterprise profile similarity recall") are executed. This strategy usually builds upon the core circle results or searches within a wider range of parameters, aiming to expand from the semantic or attribute level to discover related but not identical cases, forming an expanded subset of candidate cases (e.g., 30).

[0043] The third layer (exploration circle): Finally, the exploration recall strategy is implemented (such as "time-series hot topic weighted recall"). This strategy does not rely entirely on direct matching, but recalls cases based on their novelty, recent popularity, or cross-domain trends, aiming to introduce inspiring content and form a subset of exploration candidate cases (such as 20).

[0044] After the three-layer recall is completed, the mixing and optimization phase begins. A dedicated mixer will: prioritize filling: first, include all cases concentrated in the core circle according to the quota (50 cases), and multiply the original match scores of these cases by a weighting factor greater than 1 (such as 1.2) to highlight their high relevance.

[0045] Supplementation and Diversity Control: Next, supplement cases from the expanded circle set according to the quota (e.g., 30 cases). When including cases from the exploration circle set (e.g., 20 cases), industry diversity control will be implemented. For example, if the included cases are overly concentrated in one or two industries, exploration cases from other industries will be prioritized to break the "information cocoon".

[0046] Deduplication and Initial Ranking: During the inclusion process, global deduplication is performed based on case ID. Finally, all included cases are summed and initially ranked according to their base scores (such as service provider rating and case freshness) and source level weights (core circle weighted scores), forming an initial candidate case set containing a predetermined number (approximately 100) of candidate cases for use in subsequent deep matching stages. This process ensures that the recall results are highly relevant, rich, and possess a certain degree of exploratory value.

[0047] In one embodiment of the present invention, based on step S3, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0048] After obtaining an initial candidate case set containing cases in the core circle, expansion circle, and exploration circle, the system starts the hierarchical semantic encoding module to perform differentiated and refined semantic analysis on cases at different confidence levels in order to extract their matching features.

[0049] This module calls three encoders with different computational complexities based on the source hierarchy of the cases: A deep encoder (for core circle cases): This encoder allocates the most computational resources to high-confidence core circle cases. It performs the following operations: deep industry knowledge graph disambiguation to accurately parse the technical terms and business scenarios in the case descriptions.

[0050] By using a Natural Language Processing (NLP) model, we analyze the implicit "problem-solution" logical patterns in case documents and calculate their pattern matching degree with the enterprise's pain points.

[0051] Extract and quantify the technology stack composition, implementation complexity, and quantifiable business impact of the case (such as percentage improvement in efficiency and cost reduction rate).

[0052] The final output is a high-dimensional, information-rich deep feature vector.

[0053] Standard encoder (for extended loop cases): For extended loop cases, a standard processing flow that balances efficiency and accuracy is adopted: standard industry tag matching and keyword extraction are performed, and a semantic vector representation of the case is generated using a pre-trained sentence vector model (such as BERT).

[0054] Extract standardized attribute features from the structured fields of the case, such as service type and applicable enterprise size.

[0055] Output a standard feature vector that encompasses the core semantics and attributes.

[0056] Lightweight Encoder (for Exploration Circle Cases): For Exploration Circle cases, the focus is on evaluating their novelty and diversity value: quickly extracting the case's innovative tags (such as whether it adopts emerging technologies), industry trend popularity (based on recent search and attention data), and its industry classification.

[0057] Calculate the difference measure between this case and the current mainstream case set.

[0058] Output a low-dimensional, lightweight feature vector that primarily captures the novelty, trend, and diversity contributions.

[0059] To ensure fair comparison of feature vectors across different levels of cases, the system incorporates a vector space alignment projector. This projector uses a pre-trained projection matrix to perform dimensionality reduction and dimensionality increase transformations on the "deep feature vectors" and "lightweight feature vectors," respectively, projecting them all onto a unified semantic feature space containing the "standard feature vectors." After this step, all cases are represented as comparable feature vectors of the same dimension, collectively referred to as the matching feature vectors for each service case, providing standardized input for subsequent interpretable matching score calculations.

[0060] In one embodiment of the present invention, based on step S4, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0061] After obtaining the matching feature vectors for each service case, the system launches the interpretable matching scoring engine. This engine first performs dynamic weight configuration. It reads the intent weight dictionary generated in S102 and identifies the dominant intent (e.g., the one with the highest weight). The engine has a built-in intent-weight mapping table used to map the dominant intent to the specific weight parameters of each feature dimension in the matching scoring formula. The basic form of the matching scoring formula is: Score = α*Industry_Match + β*Process_Match + γ*Preference_Match + δ*Other_Features.

[0062] The mapping rule example is as follows: If the dominant intent is industry benchmarking intent, the system will significantly increase the industry matching dimension weight α (e.g., from the base value of 0.3 to 0.5), while correspondingly reducing the weights of other secondary dimensions to ensure that excellent cases in the same industry receive higher scores.

[0063] If the dominant intent is problem-solving intent, the system will significantly increase the weight β of the business process matching dimension (e.g., from 0.3 to 0.5), and may use "problem-solution pattern matching degree" as the core calculation factor of Process_Match, so that solutions that directly address the pain points of enterprises will be ranked higher.

[0064] If the dominant intention is cost control, the system significantly increases the weight γ of the preference matching dimension (e.g., from 0.2 to 0.4), and when calculating Preference_Match, it focuses on including the matching degree between the case's price range, ROI (Return on Investment) data and the enterprise's cost preferences.

[0065] After the weights are configured, the engine uses the adjusted formula to combine the enterprise demand vector V_q with the matching feature vector of each case to calculate the initial matching score in parallel.

[0066] Subsequently, the engine performs hierarchical confidence adjustments. The system defines adjustment factors (CF) for cases at different recall levels (e.g., core circle CF=1.1, expansion circle CF=1.0, exploration circle CF=0.9). The initial match score of each case is multiplied by its corresponding adjustment factor to obtain the final explainable match score. This step reasonably balances precision (core case scores increase) and exploration (exploration case scores decrease but still have a chance) in the final ranking.

[0067] Finally, the sorting and list generation module sorts all candidate cases from highest to lowest according to their final scores, and extracts the top N (e.g., Top 10) to generate the final recommendation list. This list is output as a structured data object, where each recommended case entry contains its ID, final score, and scores for each dimension used to explain the score, providing a data foundation for the next step of generating recommendation reasons. The weight configuration logic and correction factor application throughout the entire scoring process are recorded as an important part of the interpretability output.

[0068] In one embodiment of the present invention, based on step S5, a possible embodiment will be given below, and its specific implementation will be described in a non-limiting manner.

[0069] After generating the sorted recommendation list, the system activates an explanatory recommendation reason generator. At its core is a structured reason template library. This library pre-loads various explanatory templates, organized along two dimensions: first, based on the case's source recall level (core circle, extended circle, exploration circle); and second, based on potential demand intent (e.g., industry benchmarking, problem solving). For example: The first type of template (core circle): This template logic focuses on precise matching, with a typical sentence like: "This solution belongs to the same industry as you (core circle case.industry) and has successfully solved the '[enterprise demand vector.core pain point]' type of problem. Its [case attributes.core technology / effect] highly aligns with your needs." The second type of template (extended circle): This template logic focuses on transferable similarity, with a typical sentence like: "Although this solution originates from the [extended circle case.source industry] field, the '[case attributes.core problem pattern]' it solves aligns with your '[enterprise demand vector.pain point' aspect'." "The challenges are logically highly similar and offer valuable lessons." The third type of template (Exploration Circle): This template focuses on inspiration and trends, with a typical sentence structure like: "This is an innovative solution from the [Exploration Circle Case. Innovative Tag]. Although its [Exploration Circle Case. Industry] field differs from yours, its [Case Attribute. Emerging Technology / Model] is becoming a focus trend in the [Enterprise Demand Vector. Industry] field, providing you with cutting-edge ideas." When the generator works, it first selects the most suitable basic explanation template from the template library based on the recall level of the case to be explained and the dominant intent of this round of matching.

[0070] Next, data population is performed. The system extracts detailed attributes of the case (such as industry and key solution points) from the backend and extracts key matching dimensions (such as industry matching degree 0.95 and process matching degree 0.88) and their contribution values ​​from the matching score records. This structured data is automatically populated into the reserved variable positions in the template to generate one or more preliminary reason fragments.

[0071] Next, the reasons are optimized and synthesized. Based on preset rules (e.g., prioritizing the matching dimension with the highest contribution, or emphasizing reasons most relevant to the dominant intent), the system intelligently selects 1 to 3 of the most persuasive key reasons from all preliminary reason fragments. These reasons are fluently combined using conjunctions and then refined with simple natural language (e.g., ensuring sentence fluency and clear pronoun references) to ultimately form a coherent and easily understood explanatory natural language recommendation reason.

[0072] Finally, the system encapsulates the sorted recommendation list, the explanatory reasons for each case, and the key data indexes used to generate the reasons into a structured recommendation result response, which is then returned to the front-end application for display. This process ensures that each recommendation is supported by clear, reasonable, and personalized logic.

[0073] In some embodiments, the interpretable enterprise service matching system may include multiple functional modules composed of computer program segments. The computer programs of each program segment in the interpretable enterprise service matching system may be stored in the memory of a computer device and executed by at least one processor to perform the function of interpretable enterprise service matching (see Figure 1 for details).

[0074] In this embodiment, the interpretable enterprise service matching system can be divided into multiple functional modules according to the functions it performs, as shown in Figure 2. A module, as referred to in this invention, is a series of computer program segments that can be executed by at least one processor and perform a fixed function, stored in memory. In this embodiment, the functions of each module will be detailed in subsequent embodiments.

[0075] The system comprises the following modules: a demand acquisition module, which acquires demand data from enterprise users, generates a structured enterprise demand vector based on the demand data, and analyzes the enterprise's potential demand intent based on the enterprise demand vector; a case retrieval module, which determines a combination of retrieval strategies for candidate service cases based on the potential demand intent, and retrieves an initial set of candidate cases from the service case library based on the combination of retrieval strategies; a semantic analysis module, which performs semantic analysis on the service cases in the initial set of candidate cases to obtain the matching features of each service case; a scoring calculation module, which integrates the enterprise demand vector, the potential demand intent, and the matching features to calculate an interpretable matching score, and generates a sorted recommendation list based on the matching score; and a reason generation module, which generates explanatory recommendation reasons for the service cases in the recommendation list based on the combination of retrieval strategies, matching score elements, and service case attributes.

[0076] Figure 3 illustrates that the interpretable enterprise service matching method provided in this application embodiment can be applied to a device. Those skilled in the art will understand that the device structure involved in the embodiments of this invention does not constitute a limitation on the device. A device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. In the embodiments of this invention, the device includes, but is not limited to, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of this application described and / or claimed herein.

[0077] The device 300 may include a processor 310, a memory 320, and a communication unit 330. These components communicate via one or more buses. Those skilled in the art will understand that the server structure shown in the figure does not constitute a limitation of the present invention. It may be a bus topology or a star topology, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0078] The memory 320 can be used to store execution instructions of the processor 310. The memory 320 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. When the execution instructions in the memory 320 are executed by the processor 310, the device 300 is able to perform some or all of the steps in the above method embodiments.

[0079] The processor 310 serves as the control center of the storage device, connecting various parts of the electronic device via various interfaces and lines. It executes software programs and / or modules stored in the memory 320, and calls data stored in the memory to perform various functions of the electronic device and / or process data. The processor can be composed of integrated circuits (ICs), such as a single packaged IC or multiple packaged ICs with the same or different functions connected together. For example, the processor 310 may consist only of a central processing unit (CPU). In this embodiment of the invention, the CPU may have a single processing core or include multiple processing cores.

[0080] The communication unit 330 is used to establish a communication channel, enabling the storage device to communicate with other devices. It can receive user data sent by other devices or send user data to other devices.

[0081] The present invention also provides a computer storage medium, wherein the computer storage medium may store a program, which, when executed, may include some or all of the steps provided in the embodiments of the present invention. The storage medium may be a magnetic disk, an optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0082] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, or any other medium capable of storing program code. It includes several instructions to cause a computer device (which may be a personal computer, a server, or a second device, network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0083] The same or similar parts between the various embodiments in this specification can be referred to mutually. In particular, the device embodiments are basically similar to the method embodiments, so the description is relatively simple, and the relevant parts can be referred to the description in the method embodiments.

[0084] In the embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or modules may be electrical, mechanical, or other forms.

[0085] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0086] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0087] Although the present invention has been described in detail with reference to the accompanying drawings and preferred embodiments, the present invention is not limited thereto. Various equivalent modifications or substitutions can be made to the embodiments of the present invention by those skilled in the art without departing from the spirit and essence of the invention, and such modifications or substitutions should all be within the scope of the present invention. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should also be covered within the protection scope of the present invention.

Claims

1. An interpretable enterprise service matching method, characterized in that, include: Acquire enterprise user demand data, generate structured enterprise demand vectors based on the demand data, and analyze the enterprise's potential demand intentions based on the enterprise demand vectors; Based on the potential demand intent, a combination of recall strategies for candidate service cases is determined, and based on the combination of recall strategies, an initial set of candidate service cases is recalled from the service case library; semantic analysis is performed on the service cases in the initial set of candidate service cases to obtain the matching features of each service case; By integrating the enterprise demand vector, the potential demand intent, and the matching features, an interpretable matching score is calculated, and a ranked recommendation list is generated based on the matching score. Explanatory recommendation reasons are generated for the service cases in the recommendation list according to the recall strategy combination, matching score elements, and service case attributes.

2. The method according to claim 1, characterized in that, The process involves acquiring enterprise user demand data and generating a structured enterprise demand vector based on this data. This includes: providing standardized demand questionnaires to enterprise users through an interactive interface and collecting their submitted answers; simultaneously, acquiring the enterprise's registration information and initial behavior data on the platform; parsing and normalizing the demand data using an enterprise profile parser, mapping non-standardized descriptive information to predefined standard tags; and generating a structured demand vector composed of multiple standard tags based on the mapping results. The parsing and normalization process includes: converting text-based industry descriptions into national standard industry classification codes, and / or mapping personnel quantity ranges to preset enterprise size levels.

3. The method according to claim 1, characterized in that, Analyzing a company's potential demand intentions based on its demand vector includes: inputting the company's demand vector into a preset intention analysis model; using the intention analysis model, based on predefined intention classification rules and the company attributes and demand keywords contained in the company's demand vector, calculating the company's score weights on multiple preset intention dimensions to determine its potential demand intentions; the preset intention dimensions include at least two of the following: industry benchmarking intention, problem-solving intention, cost control intention, and exploratory intention; the intention classification rules are rule-based weight allocation logic, which at least includes: if the digital maturity attribute in the company's demand vector is beginner level, then increase the weights of industry benchmarking intention and exploratory intention; if the company size attribute in the company's demand vector is micro or small, then increase the weight of cost control intention.

4. The method according to claim 1, characterized in that, Based on the potential demand intent, a combination of recall strategies for candidate service cases is determined, and an initial set of candidate cases is recalled from the service case library based on the recall strategy combination. This includes: dynamically selecting at least one primary recall strategy and at least one secondary recall strategy from a pre-built multimodal recall strategy library based on the dominant potential demand intent, and configuring an exploratory recall strategy; based on preset recall quantity quotas at each level, sequentially executing the primary recall strategy, secondary recall strategy, and exploratory recall strategy to obtain a core candidate case subset, an extended candidate case subset, and an exploratory candidate case subset from the service case library, respectively; and prioritizing inclusion of the core candidate case subset according to the quota. The process involves: identifying cases within the initial candidate case set and assigning them a weighted score; supplementing the expanded candidate case set and the exploratory candidate case set with additional cases according to quotas, wherein industry diversity control is applied to the inclusion of the exploratory candidate case set to avoid excessive industry concentration; deduplicating all included cases and initially ranking them based on their own attribute scores and source hierarchy weights to form the initial candidate case set; wherein the multimodal recall strategy library includes at least three of the following strategies: industry semantic cluster recall strategy, problem-solution pattern matching strategy, enterprise profile similarity recall strategy, success pattern migration recall strategy, and time-series hotspot weighted recall strategy.

5. The method according to claim 4, characterized in that, Based on the preset recall quotas for each level, the main recall strategy, the auxiliary recall strategy, and the exploratory recall strategy are executed sequentially, including: First, the main recall strategy is executed to obtain the core candidate case subset with high confidence matching; second, the auxiliary recall strategy is executed to perform semantic or attribute expansion on the core candidate case subset to obtain the extended candidate case subset; finally, the exploratory recall strategy is executed to obtain the exploratory candidate case subset with low relevance to the current demand intent but with inspiration based on diversity, novelty, or trend indicators.

6. The method according to claim 1, characterized in that, The process involves integrating the enterprise demand vector, the potential demand intent, and the matching features to calculate an interpretable matching score. A ranked recommendation list is then generated based on this score. This includes: dynamically configuring the calculation weights of different feature dimensions in the matching score formula according to the weights of each intent dimension in the potential demand intent; calculating an initial matching score for each service case in the initial candidate case set based on the configured matching score formula; adjusting the initial matching score according to the confidence level at which each service case is recalled to obtain a final interpretable matching score; and ranking all service cases according to the final interpretable matching score to generate the recommendation list. The dynamic configuration of the calculation weights in the matching score formula includes: increasing the calculation weight of the industry matching feature dimension in the matching score formula when the industry benchmarking intent dominates in the potential demand intent; increasing the calculation weight of the business process matching feature dimension in the matching score formula when the problem-solving intent dominates in the potential demand intent; and increasing the calculation weight of the cost preference feature dimension in the matching score formula when the cost control intent dominates in the potential demand intent.

7. The method according to claim 1, characterized in that, Based on the recall strategy combination, matching scoring elements, and service case attributes, explanatory recommendation reasons are generated for service cases in the recommendation list. This includes: selecting a corresponding basic explanation template from a preset reason template library according to the service case's source recall level; wherein, the reason template library contains dedicated templates for different recall levels and different potential needs; combining the key matching dimensions in the matching scoring elements with the attribute information of the service case to fill the selected basic explanation template and generate preliminary reason fragments; from the preliminary reason fragments, selecting at least one key reason for combination and refinement based on the contribution of the matching scoring elements or preset priority rules to form the final natural language explanatory recommendation reasons; the preset reason template library includes at least: a first type of template for explaining service cases from the high-confidence core level, whose explanation logic focuses on successful practices in the same industry and targeted solutions to pain points; a second type of template for explaining service cases from the extended level, whose explanation logic focuses on cross-domain semantic similarity and solution transferability; and a third type of template for explaining service cases from the exploration level, whose explanation logic focuses on the novelty, trendiness, and diverse value of the solution.

8. An interpretable enterprise service matching system, characterized in that, include: The requirement acquisition module is used to acquire the requirement data of enterprise users, generate a structured enterprise requirement vector based on the requirement data, and analyze the enterprise's potential requirement intentions based on the enterprise requirement vector. The case recall module is used to determine the recall strategy combination of candidate service cases based on the potential demand intent, and recall an initial candidate case set from the service case library based on the recall strategy combination; the semantic analysis module is used to perform semantic analysis on the service cases in the initial candidate case set to obtain the matching features of each service case. The scoring calculation module is used to integrate the enterprise demand vector, the potential demand intent and the matching features to calculate an interpretable matching score, and generate a sorted recommendation list based on the matching score; The reason generation module is used to generate explanatory recommendation reasons for service cases in the recommendation list based on the recall strategy combination, matching scoring elements and service case attributes.

9. An interpretable enterprise service matching device, characterized in that, include: Memory for storing interpretable enterprise service matching programs; A processor, configured to implement the steps of the interpretable enterprise service matching method as described in any one of claims 1-7 when executing the interpretable enterprise service matching procedure.

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

Images