A new category of procurement prediction method and system for a centralized procurement platform

Abstract

This invention relates to the field of computer technology and discloses a method and system for predicting new product categories in a centralized procurement platform. The method involves defining demand placeholders containing functional specifications on the buyer's client side, which are then anonymously aggregated by the server and matched against the functional profiles of candidate new products. When a match is found to be successful, a simulation suggestion is sent to the source client, and weighted intention signals generated by the client based on the user's selected tiered intention levels are obtained. Finally, the server calculates one or more weighted intention signals based on the weight coefficients corresponding to each intention level to obtain the total weighted predicted demand and procurement risk coefficient for the candidate new products. This invention resolves the contradiction between the accuracy of demand forecasting and the protection of user business privacy. While effectively protecting data privacy, it quantifies user subjective intentions, significantly improving the accuracy and reliability of forecasts and providing the platform with an objective basis for procurement decisions.

Description

Technical Field

[0001] This invention relates to the field of computer technology, specifically to a method and system for predicting new product categories in a centralized procurement platform. Background Technology

[0002] With the digital transformation of enterprise procurement models, centralized procurement platforms, acting as a bridge connecting numerous buyers and suppliers, are playing an increasingly important role in improving procurement efficiency and reducing procurement costs. To maintain platform vitality and market competitiveness, centralized procurement platforms need to continuously introduce new product categories that align with market trends and potential user needs.

[0003] However, in the decision-making process of introducing new product categories, platform operators generally face a core challenge: how to accurately and reliably predict the market demand for a candidate new product. Traditional decision-making methods often rely on macro-level market research reports or extrapolation analysis of existing sales data. The demand information obtained by these methods is usually lagging, imprecise, and has a weak correlation with the actual and future purchasing intentions of specific buyers within the platform.

[0004] To obtain more direct demand signals, platforms may attempt to communicate with purchasing users or initiate demand surveys. However, in practice, information such as future project plans and bills of materials held by purchasing users often involve their core trade secrets, and users are generally unwilling or unable to fully disclose such sensitive data to the platform operator. This information barrier makes it difficult for platforms to collect accurate, specific, and highly certain future demand data, often putting new product introduction decisions in a dilemma: if inventory is purchased based on uncertain market signals, there may be huge risks of inventory backlog and capital tied up; if market opportunities are missed due to insufficient signals, the platform's competitiveness and user stickiness will be weakened.

[0005] Therefore, how to obtain and quantify the future purchasing intentions of purchasing users while effectively protecting their business privacy, so as to make more accurate predictions of market demand for new product categories, is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] The technical problem that this invention aims to solve is that when introducing new product categories, existing centralized procurement platforms struggle to accurately predict market demand while protecting the business secrets of purchasing users, leading to risks such as procurement decision-making errors, inventory backlogs, or missed business opportunities for platform operators.

[0007] To address the aforementioned technical problems, the first aspect of this invention provides a method for predicting new product categories in a centralized procurement platform. This method anonymizes the private demand planning of the purchaser on the client side and interacts with the server, quantifying and classifying the user's purchasing intentions to ultimately form a weighted quantitative prediction of candidate new products, thereby providing data support for the platform operator's procurement decisions.

[0008] Specifically, the method includes: Retrieve requirement placeholders defined by the purchasing client, which include functional specifications and required quantities; On the server side, the candidate new product feature profiles entered by the platform operator are matched and calculated with the required placeholders; When a match is found to be successful, a simulated suggestion containing the candidate new product information is sent to the source purchaser client of the demand placeholder. The system obtains the tiered intention levels selected by users for the candidate new products from the purchasing client and generates a weighted intention signal that includes the demand quantity and the tiered intention levels. On the server side, based on the intention level weight coefficients corresponding to the tiered intention levels, weighted intention signals for one or more candidate new products are calculated to obtain the total weighted predicted demand for the candidate new products.

[0009] In one possible implementation, to improve the accuracy and efficiency of matching, the demand placeholder may further include a budget unit price range and a set of key constraints, and the candidate new product feature profile may further include an estimated unit price and structured functional specifications. Accordingly, the matching calculation process may first perform constraint filtering, that is, comparing whether the estimated unit price is within the budget unit price range, and comparing whether the structured functional specifications meet the set of key constraints.

[0010] In one possible implementation, to further improve the accuracy of functional matching, after the aforementioned constraint filtering, the matching calculation can also perform semantic similarity evaluation on the functional specifications of the candidate new product functional profile and the demand placeholders. This evaluation process can use a pre-trained natural language processing model to map the text description into feature vectors, and perform quantitative comparison by calculating the cosine similarity between the two feature vectors. A successful match is determined only when the cosine similarity is greater than or equal to a preset matching threshold.

[0011] In one possible implementation, to ensure user anonymity while pushing simulated suggestions, this invention employs an anonymous communication mechanism based on a temporary association identifier. When the purchasing client uploads a requirement placeholder, the server generates and returns a temporary association identifier, and establishes a mapping relationship between the requirement placeholder and the temporary association identifier internally. The purchasing client, on the other hand, establishes an internal association relationship between its requirement placeholder and the temporary association identifier. Subsequently, the server can target and send simulated suggestions to the correct source purchasing client based on this temporary association identifier, without obtaining any user identity information.

[0012] In one possible implementation, the tiered intention levels are used to quantify a user's purchasing intentions. This may include at least two levels of increasing commitment strength, for example: a first intention level indicating that the user has initial interest in the candidate product; a second intention level with a commitment strength higher than the first intention level, indicating that the user has already considered the candidate product as their preferred option in their purchasing plan; and a third intention level with a commitment strength higher than the second intention level, indicating that the user agrees to have their demand quantity aggregated with the demand quantities of other buyers.

[0013] In one possible implementation, the specific process of calculating the total weighted predicted demand is as follows: for each weighted intention signal collected for the candidate new product, the demand quantity is multiplied by the intention level weight coefficient corresponding to its intention level to obtain a weighted demand value; then all weighted demand values ​​for the candidate new product are summed to obtain the total weighted predicted demand.

[0014] In one possible implementation, to assess the reliability of the prediction results, the present invention may also calculate a procurement risk coefficient for the candidate new product. This calculation process may include: first, calculating the total unweighted demand for the candidate new product, which is the arithmetic sum of the demand quantities in one or more weighted intention signals regarding the candidate new product; then, deriving the procurement risk coefficient by calculating the relationship between the total weighted predicted demand and the total unweighted demand.

[0015] In one possible implementation, in order to enable the prediction model to have self-optimization capabilities, the present invention can also dynamically adjust the weight coefficient of the intention level based on the conversion rate of demand at different intention levels in historical data that is ultimately converted into actual purchase orders.

[0016] A second aspect of the present invention provides a new product category procurement forecasting system for a centralized procurement platform, used to execute any of the aforementioned methods. The system includes a procurement client and a server.

[0017] The purchasing client includes: The virtual project bill of materials management module is used by users to define requirement placeholders that include functional specifications and required quantities. The interactive simulation and intention grading module is used to receive simulation suggestions sent by the server and generate a weighted intention signal that includes the number of demands and the tiered intention levels selected by the user.

[0018] The server communicates with the purchasing client, including: An anonymous demand aggregation engine is used to obtain the demand placeholders defined by the virtual project bill of materials management module; The demand-feature matching and push engine is used to match and calculate the candidate new product feature profiles entered on the server with the demand placeholders, and generate and send the simulation suggestions to the interactive simulation and intention classification module when the match is successful. The weighted forecasting and risk assessment engine is used to obtain the weighted intention signal from the purchasing client and calculate the total weighted forecast demand of the candidate new products based on the intention level weight coefficient.

[0019] This invention provides a method and system for predicting new product categories in a centralized procurement platform. It offers the following advantages: 1. This invention effectively protects the privacy of the purchasing user's business planning by performing private data management on the client side and uploading only anonymized and structured requirement placeholders to the server.

[0020] 2. By introducing a tiered intention level and an intention level weighting coefficient, this invention transforms users' vague purchasing intentions into quantifiable and weighted predictive basis, significantly improving the accuracy and reliability of demand forecasting.

[0021] 3. By calculating the total weighted forecast demand and procurement risk coefficient, this invention provides platform operators with multi-dimensional decision support data, which helps to formulate more reasonable procurement and inventory strategies and reduce market risks. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the architecture of a new product category procurement forecasting system for a centralized procurement platform according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a new product category procurement forecasting method for a centralized procurement platform according to an embodiment of the present invention.

[0023] Among them, 10. Purchaser client; 110. Virtual project bill of materials management module; 120. Interactive simulation and intention classification module; 130. Local data anonymization and secure communication module; 20. Server; 210. Anonymous demand aggregation engine; 220. Demand-function matching and push engine; 230. Weighted prediction and risk assessment engine; 240. Platform operation and management backend. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0025] See attached document Figure 1 , Figure 1 This is a schematic diagram of the architecture of a new product category procurement forecasting system for a centralized procurement platform according to an embodiment of the present invention. The new product category procurement forecasting system for a centralized procurement platform provided by the present invention is logically divided into a procurement client 10 deployed on the procurement user side and a server 20 deployed on the centralized procurement platform side.

[0026] The purchasing client 10 is used to provide purchasing users with a private demand planning environment and interactive interface, which may include: a virtual project bill of materials management module 110, an interactive simulation and intention classification module 120, and a local data anonymization and secure communication module 130.

[0027] The server 20 is used to anonymously aggregate, match, and quantitatively predict user needs across the entire platform. It may include: an anonymous demand aggregation engine 210, a demand-function matching and push engine 220, a weighted prediction and risk assessment engine 230, and a platform operation and management backend 240.

[0028] In a specific embodiment of the present invention, the overall workflow of the new product category procurement forecasting method for the centralized procurement platform is described in the appendix. Figure 2 , Figure 2 This is a flowchart illustrating the method, which includes the following steps: S201, the purchasing user can create a virtual project bill of materials for its future projects through the virtual project bill of materials management module 110 in the purchasing client 10, and define a demand placeholder containing information such as functional specifications and quantity when materials not provided by the platform are needed.

[0029] S202, the anonymous demand aggregation engine 210 within the server 20 initiates an anonymous data extraction request to one or more purchasing client 10s. Upon responding to the request, the local data anonymization and secure communication module 130 of the purchasing client 10 extracts only the demand placeholder data and removes all identity information, then uploads it.

[0030] S203, the anonymous demand aggregation engine 210 collects the received anonymous demand placeholder data and builds a global, aggregated pool of future demands on the server 20.

[0031] S204, the platform operator inputs the functional profile of candidate new products through the platform operation management backend 240. The demand-function matching and push engine 220 obtains this profile and performs matching calculations with the demand placeholders in the aggregated demand pool.

[0032] S205, when a successful match is determined, the demand-function matching and push engine 220 generates a simulated suggestion and sends it to the source purchaser client 10 of the demand placeholder.

[0033] S206, the interactive simulation and intention classification module 120 of the purchasing client 10 receives simulation suggestions and, after being triggered by the user, performs replacement simulation calculations locally for the user to evaluate.

[0034] S207 If the user performs a save operation after evaluation, the interactive simulation and intention grading module 120 will guide the user to select a tiered level of their purchasing intention, and generate a weighted intention signal containing new product identification, required quantity and intention level based on the user's selection.

[0035] S208, the local data anonymization and secure communication module 130 securely uploads the weighted intention signal to the server 20.

[0036] S209, the weighted prediction and risk assessment engine 230 of the server 20 receives and aggregates the weighted intention signals for the candidate new product, and calculates the weighted predicted total demand and procurement risk coefficient of the new product based on the preset weight coefficient.

[0037] S210, the weighted prediction and risk assessment engine 230 outputs the calculation results to the platform operation and management backend 240, providing objective data support for the platform operator's subsequent procurement decisions.

[0038] The steps of the method of the present invention will be described in detail below.

[0039] In one specific embodiment of the present invention, step S201 transforms the unstructured future procurement plan within the purchasing user's internal system into a standardized data format recognizable by the system, ensuring the privacy of user data throughout the process. This step is executed by the virtual project bill of materials management module 110 deployed within the purchasing client 10.

[0040] This module provides users with an independent and private project planning environment. Users can create virtual project bills of materials for one or more future procurement projects within this environment. Users can add existing product categories from the centralized procurement platform to the bill of materials and indicate the planned purchase quantity and expected delivery time.

[0041] The Virtual Project Bill of Materials Management module 110 provides a mechanism for defining material requirements that do not yet exist on the platform. This mechanism is implemented by creating requirement placeholders, and the specific creation process may include the following steps: S1101, when a user is planning their project materials using the virtual project bill of materials management module 110, it is identified that a certain material requirement cannot be met by the product catalog currently provided by the platform.

[0042] S1102, the user initiates an instruction to create a new requirement entry through the user interface provided by this module.

[0043] S1103, the virtual project bill of materials management module 110 responds to this instruction, presenting a structured data entry interface to the user. Through this interface, the user defines the various attributes of the requirement placeholders. The data structure of a requirement placeholder can be represented as follows: The definition is as follows: ; in, A unique identifier representing a demand placeholder, generated by the system during creation, used for unique indexing within the client; This field represents a functional specification description. It can be natural language text, such as a pressure sensor that can operate in an environment from -40℃ to 85℃ and has an IP67 protection rating; or it can be a structured set of key-value pairs, such as {computing power: >100TOPS, power consumption: <150W}. This design allows it to accommodate both precise technical parameters and general functional requirements. This represents the budget unit price range, which is a numerical pair that includes the lowest acceptable price and the highest acceptable price. This is used to define the user's cost expectations; This indicates the required quantity, which is a specific numerical value representing the planned purchase quantity of the material. This represents a demand time window, which is a pair of dates containing the expected start date and end date for delivery. This is used to indicate the urgency of the need; This represents a set of key constraints, which is a collection of one or more key-value pairs used to define non-functional requirements that must be met. For example, {Compatibility: Must be compatible with Siemens S7-1200 series PLC, Certification: Must pass CE certification}.

[0044] In this embodiment, the privatized management of the virtual project bill of materials (BOM) refers to all virtual project BOM data created by the user, including all defined requirement placeholders. All data is stored in user-controlled storage space. This storage space can be the local storage of the device where the purchasing client 10 is located, or a private cloud storage account designated by the user. The programming of the virtual project bill of materials management module 110 ensures the isolation of this data, preventing direct read access from the server 20 or any other external entity, thereby protecting the confidentiality of the user's business plan.

[0045] Through the above steps, the scattered and non-public future needs of the purchasing users are transformed into structured demand placeholder data objects, laying the data foundation for subsequent anonymous aggregation and analysis, while simultaneously protecting user data privacy.

[0046] In one embodiment of the present invention, step S202 aims to establish a secure and anonymous demand data transmission channel from the purchaser client 10 to the server 20, which is implemented by the local data anonymization and secure communication module 130 deployed in the purchaser client 10.

[0047] The specific implementation of this step may include the following steps: S1211, the local data anonymization and secure communication module 130 periodically initiates a polling request to a preset and publicly accessible signaling endpoint of the server 20 to query whether a valid data collection cycle is currently in progress. The anonymization request aggregation engine 210 of the server 20 responds to this request, returning a status signal indicating "allow submission" or "pause submission". This client-initiated polling mechanism can effectively adapt to diverse network environments on the user side.

[0048] S1212, upon receiving the "Allow Submission" signal from the server 20, the local data anonymization and secure communication module 130 performs data extraction operations locally on the client. This module interacts with the virtual project bill of materials management module 110, traversing one or more user-privately managed virtual project bills of materials, and extracting all defined requirement placeholder data objects from them. .

[0049] S1213, the local data anonymization and secure communication module 130 performs local anonymization processing on the extracted demand placeholder data. This module creates a temporary data packet, copying only each demand placeholder during the generation of this data packet. Payload data, i.e., functional specification description Budget unit price range Quantity required Demand Time Window and key constraint set During this process, any identifying information that can be used to trace the source of the data, such as user identification, project name, internal ID of the virtual project bill of materials, and the client-side unique identifier of the demand placeholder itself, will be considered. All of these were discarded and not included in the final generated data packet.

[0050] Through this anonymization process, the resulting data packet is merely a pure set of requirements specifications, completely unrelated to any specific purchasing user, project, or equipment.

[0051] S1214, the local data anonymization and secure communication module 130 transmits the anonymized data packet to the server 20. This transmission process is implemented through one or more of the following methods: First, the entire communication process is built on top of the standard Transport Layer Security (TLS) protocol, ensuring the confidentiality and integrity of the data at the transmission link layer. Second, the data packet is sent to a dedicated anonymous data receiving endpoint on the server 20. This endpoint is independent of the authentication endpoint that handles regular business such as user login and order placement, thereby avoiding data source association caused by information such as session tokens or authentication credentials at the application layer. Third, before transmission, the payload of the data packet itself can be asymmetrically encrypted using the public key of the server 20 to further enhance data security.

[0052] In one embodiment of the present invention, step S203 aggregates the discrete and anonymous demand data collected from each purchasing client 10 into a structured global dataset. This process is performed by an anonymous demand aggregation engine 210 deployed in the server 20.

[0053] This step can be implemented in the following ways: S1221, the anonymous demand aggregation engine 210 receives data packets uploaded by one or more purchasing client 10s through its dedicated anonymous data receiving endpoint. This receiving endpoint is designed not to perform any user authentication or session state checks, and each data packet it receives is treated as an independent data submission.

[0054] S1222, the anonymous request aggregation engine 210 parses and verifies the received data packets. This process first deserializes the data packets, converting them from a transmission format (e.g., JSON) back into data objects that the program can process. Subsequently, the engine performs structural validation on each data object to confirm whether it conforms to predefined request placeholders. The data structure is checked, and the data type and format of each field are verified to be valid. Any object that does not conform to the structure definition or contains invalid data will be discarded to ensure the quality and consistency of data in the aggregation requirement pool.

[0055] S1223, the verified anonymous demand placeholder data is persistently stored to build the server-side aggregated demand pool. In one specific implementation, this aggregated requirement pool can be a dedicated database table or a collection of documents. Each validated anonymous requirement placeholder is stored as a separate record in this database. The structure of this record maps to the payload of the requirement placeholder, containing a functional specification description. Budget unit price range Quantity required Demand Time Window and key constraint set Fields such as...

[0056] When storing data in the database, the anonymous demand aggregation engine 210 generates and assigns a completely new, unique identifier for each new record, valid only within the server. This identifier is used for indexing and tracing in subsequent server-side processing. It is unrelated to any identifier generated by the client, thus completely severing the possibility of tracing back to the data source at the data level. At this point, the aggregated demand pool... At the data level, this can be considered as including... A collection of records: ; in, This represents the total number of valid anonymous demand placeholders collected, with each element representing the total number of placeholders. This represents a record in the database.

[0057] S1224, to support subsequent efficient demand matching queries, the anonymous demand aggregation engine 210 optimizes the indexing of the aggregated demand pool storage. This applies to numeric or date fields, such as budget unit price ranges. Quantity required and demand time window Establish standard database indexes. For functional specification descriptions... This core text field is then used to build a dedicated text retrieval engine. One specific implementation is to use inverted index technology to build a full-text search engine, enabling the system to quickly retrieve relevant descriptions of needs based on keywords. Another, more advanced implementation is to use a pre-trained natural language processing model (such as the BERT model) to process the data during storage. The text is converted into a high-dimensional feature vector, and this vector is stored along with the record. At the same time, a vector index (such as an index based on KD-tree or HNSW algorithm) is built for these feature vectors to support subsequent efficient vector nearest neighbor search based on semantic similarity.

[0058] In one embodiment of the present invention, step S204 transforms the candidate new products identified and planned for introduction by the platform operator into standardized data objects identifiable within the system, providing input for subsequent demand-function matching calculations. This step is executed collaboratively by the platform operator through the platform operation management backend 240 and the demand-function matching and push engine 220.

[0059] The specific implementation of this step may include the following steps: S2111: After conducting market research or contacting suppliers, the platform operator identifies one or more candidate new products with market potential.

[0060] S2112, the operator creates a functional profile for each candidate new product through a dedicated interface provided by the platform's operation management backend 240. This interface provides structured forms to guide the operator in standardizing the input of various information about the candidate new products.

[0061] S2113, During the data entry process, the system organizes this information into a data object called "Candidate New Product Feature Profile," which can be represented as... The structure of this data object is designed to facilitate subsequent aggregation with placeholder requirements in the requirement pool. Perform direct comparison and matching. The definition is as follows: ; in, A unique identifier representing a candidate new product, generated by the system when creating the profile, used for unique indexing within server 20; Indicates the product name or model number of the candidate new product; This field represents the functional specifications of the new product. It provides a comprehensive description, and its internal structure is similar to that of requirement placeholders. In and Corresponding to the fields, one specific implementation method is, It can contain two parts: a long text description for semantic matching; and a structured set of key-value pairs for precise parameter and constraint matching. For example, the set of key-value pairs could be {power consumption: 120W, protection level: IP67, certification: CE}. This represents the estimated unit price of the candidate new product. This is a precise value used for subsequent cost simulation calculations on the client side. This refers to the set of other ancillary attributes of the new product, which may include supplementary information such as supplier information, estimated delivery date, product image links, and links to detailed technical documentation.

[0062] Functional profile defined in the above manner This transforms product information into machine-readable digital objects. Its standardized data structure, especially... The field design enables the subsequent matching engine to precisely align supply (candidate new products) and demand (demand placeholders) across dimensions such as functional specifications and key constraints, forming the technical foundation of the entire interactive forecasting method. After the definition is complete, this... The data object is submitted to the requirements-feature matching and push engine 220 to initiate the next stage of the matching process.

[0063] In one specific embodiment of the present invention, the requirement-function matching algorithm based on constraints and semantic similarity is implemented by the requirement-function matching and push engine 220 within the server 20, which aggregates requirements from the pool. In the middle, a functional profile of a given candidate new product is presented. Filter out all highly relevant anonymous request placeholders .

[0064] In one specific implementation, this matching algorithm may include a processing flow consisting of two consecutive stages: a constraint filtering stage and a semantic similarity evaluation stage. This design aims to significantly narrow down the candidate range for subsequent complex calculations through efficient hard-condition screening.

[0065] The specific implementation of this algorithm may include the following steps: S2121, Execute the constraint filtering phase. For a given candidate new product feature profile... The requirement-feature matching and push engine iterates through and aggregates the requirement pool 220 times. Each demand placeholder record in And perform hard constraint checks. Requirement placeholder records. This stage is considered passed only after all of the following checks are passed: Price constraint check: Estimated unit price of candidate new products Budget unit price range with demand placeholders (Right now ) for comparison. Only when When it is within this interval ( Only after this process was completed did the inspection pass.

[0066] Key parameter constraint check: Profiling the features of candidate new products The structured key-value pairs defined in the document, along with the requirement placeholder records. Key constraint set in The engine will perform a matching process, checking each match individually. Each requirement in the document. For example, if If it contains {Certification: CE}, the engine will check. Does the attribute also contain a completely matching CE certification item? Must meet All constraints in the [context].

[0067] Any placeholder record that fails any of the above checks will be excluded at this stage and will not be included in subsequent calculations.

[0068] S2122, performs a semantic similarity evaluation phase on the candidate demand placeholders that have passed the constraint filtering phase. This phase focuses on processing unstructured functional specification description text to determine the degree of functional matching between demand and supply. The engine extracts functional profiles of candidate new products. Text description in and each candidate requirement placeholder record Text description in .

[0069] Subsequently, the engine uses a pre-trained natural language processing model, such as BERT, to map the two text descriptions into high-dimensional real-valued vectors, i.e., feature vectors. Let the feature vector of the candidate product be... The feature vector of the demand placeholder is .

[0070] For the specific implementation of converting text descriptions into feature vectors through pre-trained natural language processing models (such as the BERT model based on the Transformer architecture), those skilled in the art can use open-source libraries or call mature cloud service interfaces to complete it. The internal network structure and training process are well-known technologies in the field and will not be described in detail here.

[0071] The engine quantifies the closeness of two feature vectors in the semantic space by calculating the cosine similarity between them. Cosine similarity The calculation formula is as follows: ; in, Indicates candidate new products With demand placeholders The semantic similarity score between them has a range of [-1, 1], and the closer the value is to 1, the more similar the semantics are. This indicates the text describing the features of the candidate new product. The corresponding feature vector; Placeholder text indicating requirements and function descriptions The corresponding feature vector; Represents the dot product operation of vectors; It represents the Euclidean norm (or L2 norm) of a vector.

[0072] S2123, make the final determination based on the preset matching threshold. Requirement placeholder record. Ultimately determined to be related to the candidate new product For a successful match to occur, two conditions must be met simultaneously: firstly, the... It has successfully passed the constraint filtering stage in S2121; secondly, its semantic similarity score calculated in S2122 Greater than or equal to the system's preset matching threshold .

[0073] The matching threshold This is a parameter that can be configured in the platform's operation and management backend 240, and its value is usually set in the range of (0,1]. The operator can adjust this threshold according to business needs to achieve a balance between matching accuracy and recall rate. A higher threshold means a stricter matching standard.

[0074] In one specific embodiment of the present invention, step S205 sends the successfully matched candidate new product information back to the purchasing client 10 that initially submitted the anonymity request, without compromising the anonymity design of the system throughout the process.

[0075] This step can be implemented in the following ways: S2211, when the demand-feature matching and push engine 220 determines a candidate new product feature profile With a placeholder record of a specific demand in the aggregated demand pool Upon successful matching, the engine immediately generates a "simulation suggestion" data object. This data object contains all the information needed for the client to perform simulation calculations and demonstrations, and may include a unique identifier for the candidate new product. Product Name Estimated unit price and detailed functional specifications .

[0076] S2212, to ensure that this simulated suggestion is sent to the correct source client in a targeted manner, the present invention employs an anonymous communication mechanism based on temporary associated identifiers. In method step S202, when the local data anonymization and secure communication module 130 of the purchasing client 10 uploads anonymous demand data to the server 20, the anonymous demand aggregation engine 210 of the server 20 receives and assigns an internal server identifier to the demand data. At the same time, a one-time, random, temporary association identifier unrelated to the user's identity will be generated for each piece of request data submitted in this instance. The server 20 maintains an internally time-sensitive mapping table to record... and The corresponding relationship. Then, server 20 will... This is returned as part of the response to the purchasing client 10. The purchasing client 10 then establishes and maintains its internal requirement placeholder identifiers locally. The server returned The relationships between them.

[0077] This step establishes a temporary, anonymous communication channel between the client and server for a single request submission.

[0078] S2213, After the simulation suggestions are generated, the demand-function matching and push engine 220 records the matched demands. The corresponding temporary association identifier is retrieved from the internal mapping table. The engine will... Packaged together with the simulation suggestion data object.

[0079] Subsequently, server 20 delivers the suggestion to the client using a hybrid push and polling approach. One specific implementation involves server 20 sending a wake-up notification with empty content or only a general instruction (such as "You have a new purchasing suggestion") to the client application on the target device via a publicly available push notification service.

[0080] S2214, Upon receiving this wake-up notification or during its own background polling cycle, the local data anonymization and secure communication module 130 of the purchasing client 10 will proactively initiate a query request to the dedicated endpoint of the server 20. This request carries all currently valid temporary association identifiers held by the client. List. After receiving the request, server 20 retrieves these... Are there any corresponding simulation suggestions to be issued? If so, return the packaged simulation suggestion data object as a response to the client.

[0081] This server-side wake-up and client-initiated retrieval method ensures timely information delivery while avoiding the server directly establishing a push channel to specific users, thus technically guaranteeing the anonymity and security of the push process.

[0082] In one specific embodiment of the present invention, step S206 provides the purchasing user with a secure and quantifiable decision support environment to assess the impact of accepting candidate new product recommendations. This step is implemented by the interactive simulation and intention grading module 120 deployed within the purchasing client 10.

[0083] The specific implementation of this step may include the following steps: S2221, the interactive simulation and intention classification module 120 receives a temporary association identifier from the server 20 through the local data anonymization and secure communication module 130. The module provides simulated suggested data objects. Within the locally maintained relationships, locate the corresponding original requirement placeholder in the user's private virtual project bill of materials. .

[0084] S2222, to ensure the integrity and security of the user's original project planning data, the interactive simulation and intention grading module 120 creates a sandbox environment before performing any calculations. One specific implementation is that this module creates a placeholder for the requirement in the client device's memory. A deep copy of the virtual project's bill of materials. All subsequent replacement simulations and calculations are performed on this temporary in-memory copy without modifying the original bill of materials data stored locally or in a private cloud.

[0085] S2223, in this sandbox environment, the interactive simulation and intention-leveling module 120 performs a replacement simulation operation. This operation replaces the original requirement placeholder. The relevant fields are used to simulate the candidate new products carried in the suggested data object. Information is replaced. For example, the item name in the bill of materials is updated from a descriptive placeholder name to the product name of the candidate new product. More importantly, the unit price on which cost calculations are based has changed from the original budgeted unit price range. Updated to the exact estimated unit price of the candidate new product .

[0086] S2224, after the replacement is completed, the interactive simulation and intention-leveling module 120 performs a benefit assessment calculation, the core of which is to quantify the cost changes brought about by this replacement. This module calculates the estimated costs before the replacement. Compared with the new cost of replacement And draw the cost difference. Its calculation formula can be defined as follows: ; ; ; in, This represents the estimated total cost based on the original demand placeholder budget, using a budgeted unit price range. The median value is used as the calculation benchmark; This indicates the new estimated total cost after replacing the candidate product; Indicates cost difference, if This indicates cost savings. This indicates an increase in costs; This indicates the quantity of demand defined in the demand placeholder; and These represent the lower and upper limits of the original budget unit price range, respectively; This indicates the estimated unit price of the candidate new product.

[0087] S2225, the interactive simulation and intention grading module 120 presents the results of the benefit assessment to the purchasing user through its user interface. One specific presentation method is to display key information from the original demand placeholders and key information from candidate new products side-by-side, and prominently indicate the calculated cost difference. The interface also provides interactive controls, such as a save control for accepting the simulation results and a cancel control for rejecting and closing the suggestion.

[0088] If the user chooses to cancel, the system will clear the sandbox environment created this time and all its temporary data, and the process will terminate. If the user chooses to save, it indicates that they have initially approved of the candidate product, and the system will use this as a trigger to enter the subsequent tiered intention locking process.

[0089] In one specific embodiment of the present invention, step S207 transforms the user's subjective intention into a quantifiable predictable basis. This step is performed by the interactive simulation and intention grading module 120 within the purchasing client 10.

[0090] The specific implementation of this step may include the following sub-steps: S3111, the trigger condition for this process is that after the user completes the benefit evaluation in step S2225, the user activates the saved interactive control on the user interface.

[0091] S3112, upon receiving the trigger signal, the interactive simulation and intention grading module 120 further presents a selection component for intention grading on the user interface. This component provides the user with a set of preset, tiered intention level options with different levels of commitment.

[0092] S3113, In this embodiment, the set of tiered intention level options may include three specific levels, each level corresponding to a business meaning and commitment strength, defined as follows: Attention: This is the lowest level of interest. Selecting this option indicates that the user is interested in the candidate product and agrees that the system will record this interest so that they can receive status updates or related information pushes about the new product in the future. This level does not represent any purchase commitment; its main purpose is to provide the platform with a wide range of potential interest signals.

[0093] Soft Lock-in: This represents a medium level of intent. Selecting this option indicates that the user has already chosen this candidate product as the preferred option in the current project plan's bill of materials. This demonstrates a strong purchasing inclination, but the user retains the right to change their choice at the final purchasing decision stage due to other factors (such as project changes or the emergence of a better option). This level provides the platform with a more reliable demand signal than simply "attention."

[0094] Joint Signature: This is the highest level of intent. Selecting this option indicates that the user has a very strong purchasing intention and agrees to aggregate the quantity of their confirmed demand with the demand quantities of other purchasing users who have also selected joint signature. This aggregated demand forms a collective, bulk purchasing intention visible to the supplier or platform. The user understands that selecting this option may help the platform conduct more effective preliminary business negotiations, thereby securing more favorable purchase prices or terms for all participating users.

[0095] S3114, the user interface of the interactive simulation and intention grading module 120 explains the meaning of each intention level and waits for the user to make a selection. This module captures the user's final selection and obtains the selected intention level.

[0096] S3115, after obtaining the user's intention level selection, the interactive simulation and intention grading module 120 immediately generates a "weighted intention signal" data object locally on the client. This data object contains all the core information required for subsequent weighted prediction, and its structure can be defined as follows: : ; in, A unique identifier representing the candidate new product that the user has confirmed their intention to purchase; This indicates the number of requests confirmed by the user, which is derived from the original request placeholder. This indicates the user's selected level of intent, which can be an enumeration value or an integer. For example, 1 represents attention, 2 represents soft lock, and 3 represents joint signature.

[0097] This weighted intention signal Once generated, it is transmitted to the local data anonymization and secure communication module 130 in preparation for the subsequent upload steps.

[0098] In one specific embodiment of the present invention, step S208 transmits the purchasing intention confirmed by the user locally on the client to the server 20 in a manner that ensures both anonymity and data security. This step is performed by the local data anonymization and secure communication module 130 within the purchasing client 10.

[0099] The specific implementation of this step may include the following steps: S3121, the local data anonymization and secure communication module 130 receives a locally generated weighted intention signal data object from the interactive simulation and intention classification module 120. The structure of this data object has been defined in the preceding steps, and it contains only the non-identification information necessary for subsequent predictions: a unique identifier for the candidate novel. Demand quantity and the user's selected level of intent. .

[0100] S3122, Local Data Anonymization and Secure Communication Module 130 performs this weighted intention signal The object undergoes serialization, such as converting it into a JSON-formatted data packet for network transmission. During this process, the network request generated by the module strictly excludes any information that could be associated with the user's identity. Specifically, this upload request does not contain the user's login status session token, authentication credentials, or any other form of user identifier.

[0101] S3123, the local data anonymization and secure communication module 130 sends this anonymized data packet to a dedicated, authentication-free API endpoint on the server 20 for receiving weighted intent signals. This endpoint is architecturally separate from endpoints that handle platform business logic requiring authentication, such as user orders and account management. By submitting data to a public and unsecured endpoint, the association between data and specific user sessions is further severed at the application layer.

[0102] S3124, the entire data transmission process takes place on a secure communication channel. The communication between the local data anonymization and secure communication module 130 and the server 20 is established on the Transport Layer Security (TLS) protocol, which ensures the confidentiality of the weighted intention signal data packets during transmission, prevents data eavesdropping, and also guarantees data integrity, preventing data tampering.

[0103] Through the above steps, the specific purchasing intentions originating from a particular user are transformed into completely anonymous signals and fed into the server, providing data input for the next stage of global weighted prediction calculation.

[0104] In one specific embodiment of the present invention, step S209 assigns a numerical value reflecting the likelihood of procurement conversion to each tiered intention level. This step is executed by the demand aggregation and prediction engine 230 of the server 20, and the management of its weight coefficients is completed through the platform operation management backend 240.

[0105] The specific implementation of this step may include the following steps: S3211, the demand aggregation and forecasting engine 230, during calculation, will perform calculations for each defined intention level. (For example, attention, soft lock, and joint authorship) Configure a corresponding intention level weight coefficient, denoted as . The weighting coefficient is a real number between 0 and 1. Its value directly reflects the estimated probability that the quantity of demand with this level of intent will eventually be converted into actual purchase orders.

[0106] S3212, the weighting coefficients are set according to the principle of monotonically increasing. That is, the higher the level of commitment, the larger the corresponding weighting coefficient must be. If we consider... , and Let each represent a weighting coefficient for one of the three intention levels. Then, they must satisfy the following relationship: ; A specific example of setting initial weighting coefficients could be: ; ; ; In this example, "follow" only represents initial interest and has a low conversion probability; "soft lock" represents the user's preferred option and has a high conversion probability; while "joint signature" represents a strong collective purchasing intention and has a very high conversion probability.

[0107] S3213, these intention level weighting coefficients can be dynamically configured and managed through the platform operation management backend 240. The platform operator can adjust these weighting coefficients according to the evolution of business and the accumulation of data.

[0108] S3214, the adjustment of weighting coefficients is based on statistical analysis of historical data. The demand aggregation and forecasting engine 230, or a separate data analysis module, periodically analyzes completed procurement projects. The system tracks purchase orders initially converted from candidate new product suggestions and traces back to their intention level before the order was formed. By calculating the ratio of the total intended demand to the final actual purchase volume at a specific intention level, the actual conversion rate of that intention level over a past period can be obtained. This conversion rate can then serve as a direct basis for adjusting the corresponding weighting coefficients. For example, if the system finds that 65% of all demands marked as soft-locked were ultimately converted into actual orders in the past quarter, the platform operator can adjust the weighting coefficients accordingly. The value was adjusted from 0.6 to 0.65 to make the prediction model more closely reflect the platform's actual business situation.

[0109] Through this closed-loop adjustment mechanism based on historical data feedback, the prediction model of this invention can achieve self-optimization, and its prediction accuracy will gradually improve as the platform operates for longer and the amount of data accumulates.

[0110] In one specific embodiment of the present invention, step S209 further aggregates all anonymous, discrete weighted intention signals into a specific total demand forecast for a particular candidate new product. This forecast calculation process is executed by the demand aggregation and forecasting engine 230 of the server 20.

[0111] The specific implementation of this predicted value calculation process may include the following steps: S3221, the calculation process of the demand aggregation and prediction engine 230 is triggered. This trigger can be a periodic automatic scheduling task, such as automatically updating the prediction of all active candidate new products during the daily system off-peak period; or it can be an instant calculation request for one or more specific candidate new products initiated by the platform operator through the platform operation management backend 240.

[0112] S3222, for a given candidate new product, the new product is characterized by its unique identifier. As determined, the demand aggregation and forecasting engine 230 retrieves all data related to the target demand from its dedicated intention signal database. Associated weighted intention signals These signals were anonymously uploaded and aggregated by the various purchasing clients in the preceding steps.

[0113] S3223, the demand aggregation and forecasting engine 230 performs a weighted intention signal on each retrieved signal. Process it. For a signal... The engine first parses out the number of requirements it contains. With intention level Subsequently, the engine adjusts the settings based on the level of intent. The corresponding intention level weight coefficient can be found from its internal configuration. .

[0114] In S3224, the engine calculates the total weighted predicted demand for the candidate new product by performing a weighted summation calculation. The formula for this calculation is defined as follows: ; in, Indicates that the identifier is The total weighted predicted demand for the candidate new products is the target output calculated in this step. This indicates that the candidate new product The total number of weighted intention signals collected; This represents the index used in the summation calculation, traversing from 1 to N; Indicates the first The quantity of demand contained in a weighted intention signal; Indicates the first The level of intent contained in a weighted intention signal; Indication and Intention Level The corresponding intention level weighting coefficient.

[0115] The calculation process of this formula is as follows: the engine traverses all relevant intention signals, multiplies the demand quantity of each signal by the weight coefficient corresponding to its intention level, and then sums all these product results. The final sum is the predicted value of future market demand for the candidate new product.

[0116] S3225, the calculated total weighted forecast demand Persistent storage is performed, and the candidate product is associated with it. The records are linked together. This prediction result is then pushed or displayed on the corresponding interface of the platform operation and management backend 240, providing direct quantitative basis for the platform operator's decisions on inventory preparation, supplier negotiation and marketing.

[0117] In one specific embodiment of the present invention, the platform operator is further provided with key auxiliary indicators, in addition to the total weighted predicted demand, for assessing the reliability of the prediction results. The calculation and application of this risk coefficient are performed by the demand aggregation and prediction engine 230 of the server 20, and are displayed and interpreted in the platform operation management backend 240.

[0118] The definition, calculation, and application of this procurement risk factor may specifically include the following steps: S3301, the purpose of introducing the procurement risk coefficient is to quantify the total weighted forecast demand. The uncertainty lies in the forecasted demand. A higher forecasted demand, if primarily composed of low-commitment levels of intent (e.g., concerns), carries a relatively higher risk of translating into actual purchases. Conversely, if the forecasted demand is primarily composed of high-commitment levels of intent (e.g., joint signatures), its reliability is higher and the procurement risk is lower. This embodiment addresses this by designing a procurement risk coefficient. To measure this structural risk.

[0119] S3302, the procurement risk coefficient calculation model is based on the comparison between the total weighted forecast demand and the total unweighted demand. For a given candidate new product... Its procurement risk coefficient The calculation formula is defined as follows: ; in, Indicates candidate new products The procurement risk coefficient has a value range of 0 to 1, including 0 and 1; This represents the total weighted forecast demand for the candidate product, which has been calculated in the previous steps. This represents the total unweighted demand for the candidate product, which is the simple arithmetic sum of the demand quantities from all relevant intent signals. The formula for its calculation is: ;in, The total number of intention signals. For the first The required number of signals.

[0120] S3303 provides an interpretation of the calculation results for the procurement risk coefficient. The closer the value is to 0, the higher the proportion of high-weight intentions in the total predicted demand, the stronger the certainty of the prediction result, and the lower the procurement risk. The closer the value is to 1, the more it indicates that the total predicted demand is mainly composed of low-weight intentions, the greater the uncertainty of the prediction results, and the higher the procurement risk.

[0121] S3304 applies the calculated procurement risk coefficient to the platform's operational decisions. The demand aggregation and forecasting engine 230 will... Total weighted forecast demand Both indicators are displayed together on the decision dashboard of the platform's operation and management backend. Platform operators can use these two indicators to make decisions, and specific application scenarios include: Inventory and stocking strategies: For High and For a low number of potential new products, the platform can develop a more proactive inventory preparation plan to meet the high-certainty market demand. For High but For products with higher demand, the platform can adopt a more conservative inventory strategy, such as negotiating more flexible tiered procurement agreements with suppliers to avoid the risk of demand falling short of expectations.

[0122] Supplier negotiation support: A lower level of support is needed when conducting business negotiations with suppliers of candidate new products. Values ​​are powerful data demonstrating the authenticity and reliability of market demand. Platforms can use this metric to leverage negotiations for more favorable purchase prices, payment terms, or supply guarantees.

[0123] Precision Marketing and Conversion: For For products with higher levels of interest, the system can identify them as key targets for conversion. Platform operators can then plan targeted marketing campaigns or push notifications for users who have submitted low-level interest (such as following), aiming to guide them to upgrade their interest to soft lock or joint signature, thereby effectively reducing the overall procurement risk of the product.

[0124] In one specific embodiment of the present invention, step S210 transforms the quantitative prediction results calculated in the preceding steps into a business decision-making basis that the platform operator can execute. This step is mainly implemented through the platform operation management backend 240 of the server 20.

[0125] The specific implementation of this step may include the following steps: S4001, the demand aggregation and forecasting engine 230, after completing calculations for one or more candidate new products, will calculate the total weighted forecast demand. With procurement risk coefficient , and push to the platform operation and management backend 240.

[0126] S4002, the platform operation management backend 240, provides a visual presentation to the platform operator through its built-in decision support dashboard. This dashboard displays the core forecasting metrics for each candidate new product in the form of a list or cards. For each candidate new product, the interface clearly lists at least its product name and total weighted forecast demand. and procurement risk coefficient To enhance readability, dashboards can also employ data visualization techniques, such as using bar charts to visually compare the forecasted demand for different products, or using color coding to indicate the level of procurement risk.

[0127] S4003: The platform operator formulates procurement and inventory strategies based on the data presented on the decision support dashboard. This decision-making process can be conducted according to the following rules: When a candidate new product High value, and its A low value indicates a strong and certain market demand for the new product. Platform operators can use this to develop proactive procurement and inventory strategies, such as placing large initial orders with suppliers to ensure sufficient inventory to meet market demand.

[0128] When a candidate new product The value is high, but its A high value indicates that while there is broad market interest in the new product, the commitment strength of most intentions is low, and there is significant uncertainty in demand. In this case, platform operators should adopt a prudent procurement strategy to mitigate risk, such as conducting only small-batch trial purchases or negotiating tiered procurement agreements with suppliers, linking purchase volume to the actual conversion of subsequent intentions.

[0129] When a candidate new product A low value indicates insufficient market acceptance of the product, regardless of its risk coefficient. The platform operator may then decide to postpone the introduction of the new product or place it under low-priority observation.

[0130] S4004, the forecast results are used as data support in business negotiations with suppliers. When the platform operator is dealing with a supplier with high... With low When negotiating with suppliers for potential new products, this data can serve as reliable proof of market demand. This provides the platform with strong negotiating leverage to secure more favorable purchase prices, better payment terms, or higher supply priority.

[0131] S4005, the forecast results are used to guide precise marketing conversion campaigns. When platform operators find that a product has a high [percentage missing] on the dashboard... When the value is determined, the system analyzes the composition of the intent and identifies that the risk mainly stems from a large number of low-commitment intents at the attention level. Based on this insight, platform operators can plan and execute targeted marketing campaigns, such as explaining the potential price advantages of joint signatures to users through platform announcements or event pages, to incentivize users with relevant attention to upgrade to a higher commitment level, thereby proactively managing and reducing market demand uncertainty.

[0132] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting new product category procurement on a centralized procurement platform, characterized in that, Includes the following steps: Retrieve requirement placeholders defined by the purchasing client, which include functional specifications and required quantities; On the server side, the candidate new product feature profiles entered by the platform operator are matched and calculated with the required placeholders; When a match is found to be successful, a simulated suggestion containing the candidate new product information is sent to the source purchaser client of the demand placeholder. The system obtains the tiered intention levels selected by users for the candidate new products from the purchasing client and generates a weighted intention signal that includes the demand quantity and the tiered intention levels. On the server side, based on the intention level weight coefficients corresponding to the tiered intention levels, weighted intention signals for one or more candidate new products are calculated to obtain the total weighted predicted demand for the candidate new products.

2. The method for predicting new product categories for centralized procurement platforms according to claim 1, characterized in that, The demand placeholders also include the budget unit price range and the set of key constraints; The candidate new product functional profile also includes the estimated unit price and structured functional specifications; The matching calculation includes: Compare whether the estimated unit price is within the budgeted unit price range, and compare whether the structured functional specifications meet the set of key constraints.

3. The method for predicting new product categories for centralized procurement platforms according to claim 2, characterized in that, The matching calculation also includes: Once the estimated unit price and the set of key constraints are both met, a semantic similarity assessment is performed on the functional specifications in the candidate new product functional profile and the functional specifications in the demand placeholders.

4. The method for predicting new product categories for centralized procurement platforms according to claim 3, characterized in that, The semantic similarity assessment includes: A pre-trained natural language processing model is used to map the functional specifications into feature vectors. Calculate the cosine similarity between the two feature vectors, and determine that the match is successful when the cosine similarity is greater than or equal to a preset matching threshold.

5. The method for predicting new product categories for centralized procurement platforms according to claim 1, characterized in that, The step of sending a simulated suggestion containing the candidate new product information to the source buyer's client of the demand placeholder includes: When the purchasing client uploads the requirement placeholder, the server generates and returns a temporary association identifier, and establishes a mapping relationship between the requirement placeholder and the temporary association identifier within the server. The purchasing client establishes an association between its internal demand placeholders and the temporary association identifier locally; Once the simulated suggestion is generated, the server queries the corresponding temporary association identifier based on the matched demand placeholder, and sends the simulated suggestion to the source purchaser's client through the temporary association identifier.

6. The method for predicting new product categories for centralized procurement platforms according to claim 1, characterized in that, The tiered intention level includes at least two levels of varying commitment strength, from low to high, and the levels include: The levels include: The first level of intent indicates that the user has initial interest in the candidate new products; The second level of intent, which is stronger than the first level of intent, indicates that the user has regarded the candidate new product as the preferred option in their procurement plan. The third level of intent, which represents a stronger commitment than the second level of intent, indicates that the user agrees to use their required quantity for aggregate calculations with the required quantities of other buyers.

7. The method for predicting new product categories for centralized procurement platforms according to claim 1, characterized in that, The specific steps for calculating the total weighted forecast demand are as follows: For each weighted intention signal collected for the candidate new products, the number of demands is multiplied by the intention level weight coefficient corresponding to its intention level to obtain the weighted demand value. The total weighted predicted demand is obtained by summing up all the weighted demand values ​​for the candidate new products.

8. The method for predicting new product categories for centralized procurement platforms according to claim 7, characterized in that, The method further includes: Calculate the total unweighted demand for the candidate new products, where the total unweighted demand is the arithmetic sum of the demand quantities in one or more weighted intention signals for the candidate new products; The procurement risk coefficient of the candidate new product is obtained by calculating the relationship between the total weighted predicted demand and the total unweighted demand.

9. The method for predicting new product categories for centralized procurement platforms according to claim 1, characterized in that, The method further includes: Based on historical data on the conversion rate of demand at different intention levels into actual purchase orders, the weighting coefficient of the intention level is dynamically adjusted.

10. A new product category procurement forecasting system for a centralized procurement platform, used to execute the new product category procurement forecasting method for a centralized procurement platform as described in any one of claims 1-9, characterized in that, This includes both the purchasing client and the server. The purchasing client includes: The virtual project bill of materials management module is used by users to define requirement placeholders that include functional specifications and required quantities. The interactive simulation and intention grading module is used to receive simulation suggestions sent by the server and generate a weighted intention signal containing the number of demands and the tiered intention levels selected by the user. The server communicates with the purchasing client, including: An anonymous demand aggregation engine is used to obtain the demand placeholders defined by the virtual project bill of materials management module; The demand-feature matching and push engine is used to match and calculate the candidate new product feature profiles entered on the server with the demand placeholders, and generate and send the simulation suggestions to the interactive simulation and intention classification module when the match is successful. The weighted forecasting and risk assessment engine is used to obtain the weighted intention signal from the purchasing client and calculate the total weighted forecast demand of the candidate new products based on the intention level weight coefficient.

Images