Order automatic generation and targeted sending method and system based on AI and user portrait

Abstract

The application discloses a kind of based on AI and user portrait's procurement demand order automatic generation and directional sending method and system, obtain user relevant data and generate user portrait, resource supplier portrait and transport capacity service provider portrait;Procurement demand text input by user is obtained, and intent classification model and field sequence labeling model are used to identify demand type and extract procurement field;AI analysis field is fused with user portrait, missing field is completed, and field confidence is calculated, according to confidence threshold value to determine writing confirmation or supplement, and generate purchase order;Through multi-path recall, obtain candidate object, filter target object and send purchase order;According to transaction and performance result, generate feedback signal, update weight and form closed loop;The application generates purchase order quickly by based on AI and user portrait automatic understanding user procurement demand, and sends purchase order to matched supplier or service provider, improves procurement demand publishing efficiency and supplier touch accuracy.

Description

Technical Field

[0001] This invention relates to the fields of e-commerce and intelligent information processing technology, specifically to a method and system for automatic order generation and targeted delivery based on AI and user profiling. Background Technology

[0002] In the existing GasConnect.com marketplace, which targets natural gas and related energy trading scenarios, users typically need to fill in information such as product or service type, quantity, pick-up location, unloading location, delivery terminal, expected delivery time, vehicle or tanker requirements, payment method, and the scope of resource suppliers or carriers when purchasing gas products such as LNG, liquefied natural gas, CNG, and LPG, or LNG transportation services. Because these goods and services are characterized by regional supply, terminal delivery, transportation capacity scheduling, tanker scheduling, and price fluctuations, publishing procurement needs is not as simple as entering fields for ordinary commodity orders.

[0003] In existing e-commerce systems, users primarily rely on manual input or simple copying from historical orders when posting procurement requests. This approach has the following problems: First, users need to repeatedly fill in information such as location, frequently used stations, frequently used suppliers, and frequently purchased categories each time they post a request, resulting in low operational efficiency. Second, when users only enter a brief description of their request, the system struggles to automatically identify and complete the purchase order fields, easily leading to missing or non-standard purchase order data. Third, the system typically can only send requests to suppliers or transportation service providers manually selected by the user, failing to automatically determine suitable partners based on data such as the user's historical transaction volume, frequently cooperating suppliers, frequently cooperating fleets, service routes, vehicle availability, and quotation response speed. Fourth, after a request is sent, inconsistencies can easily arise between the sending status, supplier response status, quotation status, carrier confirmation status, and order status, affecting the efficiency of subsequent procurement matching. Fifth, key fields in the purchase order, the scope of supplier or transportation service provider distribution, and quotation responses lack process verification and traceability mechanisms. Summary of the Invention

[0004] This invention provides a method and system for automatic order generation and targeted delivery based on AI and user profiling. By automatically understanding user procurement needs based on AI and user profiling, and combining data such as user location, frequently used stations, historical transaction volume, frequently cooperating suppliers, and frequently cooperating transportation service providers, the system quickly generates purchase orders and sends them to matching resource suppliers or LNG transportation service providers. This improves the efficiency of procurement demand release and the accuracy of supplier outreach, thereby solving the problems in the background technology.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows: A method for automatically generating and targeted sending orders based on AI and user profiling, characterized by performing the following steps via computer equipment: S1: Acquire user-related data and perform cleaning and standardization processing to generate user profiles, resource supplier profiles, and transportation service provider profiles; S2: Obtain the procurement requirement text input by the user, identify the requirement type of the procurement requirement text and extract the procurement field using a hybrid parsing method; S3: The procurement fields obtained by AI parsing are merged with the user profile. The missing fields in the procurement fields are filled in using the user profile. The field confidence of each procurement field is calculated. Based on the comparison result of the field confidence and the preset threshold, it is determined whether the field is automatically written, prompted to confirm by the user, or asked to supplement by the user. Finally, a structured procurement order is generated. S4: Based on the structured purchase order, combined with the resource supplier profile and the transportation service provider profile, candidate objects are recalled, a targeted sending score is calculated, and the target resource supplier or target transportation service provider is selected based on the targeted sending score. S5: Send the structured purchase order to the target resource supplier or target transportation service provider, generate purchase order number and sending batch number, record sending status, receipt status and quotation status, and perform status compensation operation when sending fails or receipt times out. S6: Generate feedback signals based on the procurement process results corresponding to the quoted status and sending status, and use the feedback signals to update each profile and adjust the field extraction weights and matching score weights.

[0006] Preferably, S1 specifically includes the following sub-steps: S11: Obtain user-related data from internal systems and external data sources. The user-related data includes basic user information, enterprise authentication information, user location, delivery address, frequently used stations, historical transaction orders, historical transaction volume, transaction categories, transaction prices, payment or settlement status, frequently cooperating resource suppliers, frequently cooperating transportation service providers, supplier quotation records, carrier quotation records, message receipt records, performance records, regional market prices, category basic data, station basic data, route distance data, and third-party market data. S12: Perform deduplication, missing value completion, outlier identification, and coding standardization on the collected data. Map category names, regional codes, station codes, supplier codes, and capacity service provider codes from different sources to a unified code, and convert different units of measurement into standard units that the system can recognize. S13: Generate a user profile based on the cleaned data. The user profile includes the user's location, frequently used stations, frequently purchased product categories, historical transaction volume, frequently cooperating resource suppliers, and frequently cooperating transportation service providers. S14: Generate resource supplier profiles and transportation service provider profiles. The resource supplier profiles include available product categories, service areas, historical transaction volume, and performance ratings. The transportation service provider profiles include available transport areas, frequently used routes, vehicle availability status, and performance ratings.

[0007] Preferably, S2 specifically includes the following sub-steps: S21: Use a domain dictionary to segment and identify entities in the procurement requirement text, and identify business entities, including LNG, liquefied natural gas, CNG, LPG, station, vehicle type, tanker volume and settlement method; S22: Input the procurement requirement text into the intent classification model and calculate the probability of each requirement type. , where x is the input procurement requirement text. ; The type with the highest probability is used as the model prediction result; when the highest probability is lower than the preset probability threshold, the system switches to the rule engine for a fallback judgment. S23: The rule engine calculates the keyword scores for resource procurement and transportation service, as well as the rule matching score, according to a preset dictionary. The calculation formula is:

[0008] in, The weight of the k-th keyword. This is an indicator function for whether the k-th keyword is matched; when the difference between the resource procurement score and the transportation service score is greater than the classification threshold, it is determined to be a resource procurement demand; when the difference between the transportation service score and the resource procurement score is greater than the classification threshold, it is determined to be a transportation service demand; when the difference between the two scores is less than or equal to the classification threshold, the user is prompted to confirm the demand type. S24: Extract procurement fields from the procurement requirement text using the BIO annotation method through the field sequence annotation model. The procurement fields include procurement category, quantity, pick-up location, unloading location, delivery station, and expected delivery time.

[0009] Preferably, S3 specifically includes the following sub-steps: S31: When there are missing fields in the procurement fields, read the corresponding frequently used data from the user profile, and select the most matching value according to the frequency of use or comprehensive sorting to fill in the procurement order; S32: Obtain the model's probability of extracting field f. Rules for scoring Consistency between this field and the user profile The stability of this field in historical orders and the degree of field conflict ; S33: Calculate the field confidence score using the following formula. :

[0010] in, For the field confidence normalization function, For bias terms, to These are the weighting coefficients for each factor; S34: Calculate the field confidence scores With the preset first threshold Second threshold Comparison :when When the system automatically writes this field into the purchase order; when When the system prompts the user for confirmation or modification before the purchase order is submitted; When this happens, the system marks the field as missing or abnormal and prompts the user to complete it.

[0011] Preferably, the candidate object recall in S4 specifically includes the following sub-steps: S41: Simultaneously employ multiple recall methods to recall candidate objects from resource supplier profiles and capacity service provider profiles. The recall methods include recall of historical cooperation objects, recall of regional or station coverage objects, recall of category or service type matching objects, recall of route coverage objects, recall of profile similarity objects, and recall of recently active and qualified objects. S42: Take the union of the candidate sets obtained from multiple recall methods:

[0012] in, A recall assembly for historical partners. For the recall set of objects covered by the region or site, Recall a set of objects matching the category or service type. For the recall set of objects covered by the line, For the image similarity object recall set, The set of recently active and qualified objects is retrieved, and then the union of the results C is deduplicated to obtain the final set of candidate objects.

[0013] Preferably, the calculation of the targeted delivery score in S4 specifically includes the following sub-steps: S43: For candidate object B, obtain its set of normalized eigenvalues. The normalized feature values ​​include cooperation strength, regional or route coverage, category or service type matching degree, supply capacity or vehicle availability, response time and performance score; S44: The strength of cooperation Calculate using the following formula:

[0014] in, This represents the number of transactions between user u and candidate B in the past a days. This represents the transaction amount over the past day (a). The time since the most recent transaction. The preset time decay factor; The repurchase rate is defined as the proportion of transactions between user u and candidate B in their historical transaction history where the time interval between two consecutive transactions is less than a preset threshold. to The weighting coefficients and , This represents the eigenvalue normalization function; S45: Price Deviation Calculate using the following formula:

[0015] in, Provide quotes to suppliers. This refers to the user's historical transaction price or regional market reference price. S46: Calculate the targeted delivery score according to the following formula. :

[0016] in, Let be the weight coefficient of the i-th normalized feature. Let i be the normalized feature value of candidate object B. The weighting coefficient for risk deduction. This is a risk deduction item for candidate B; The risk deduction items are obtained by combining factors such as expired qualifications, abnormal transportation records, default records, abnormal price deviations, and delayed receipts. S47: Sort the candidate objects from high to low according to the targeted delivery score, and select a preset number of the top-ranked candidate objects as the target resource supplier or target transportation service provider.

[0017] Preferably, S5 specifically includes the following sub-steps: S51: When a purchase order fails to be sent, the system will automatically retry sending it. After the number of retries reaches the preset limit, the system will switch to a different message channel to resend the order. The message channels include internal messages, SMS, WeChat Work, and open APIs. S52: When the reply from a resource supplier or transportation service provider times out or the quotation times out, the system records the timeout status and triggers a supplementary resend or manual intervention reminder; S53: When there is a discrepancy between the sending status, receipt status, quotation status, carrier confirmation status and order status, the system performs a status alignment operation and records the status change log. S54: Generate summary values ​​for key fields before transaction commit. If unauthorized modification of the key fields is detected, the commit is interrupted and an audit log is recorded. The key fields include purchase order number, category, quantity, pick-up location, unloading location, delivery station, target object set, and quotation amount.

[0018] Preferably, S6 specifically includes the following sub-steps: S61: Obtain feedback signals for this procurement process When the user adopts the recommendation, the supplier responds promptly, the quote is accepted, the order is completed and fulfillment is normal, Take a positive value; when a user modifies a recommendation field, deletes a recommendation object, a supplier fails to respond within a timeout period, or an order is canceled or there is a fulfillment error. Take the negative value; S62: Update the weight of field completion or rule matching according to the following formula. :

[0019] Where t is the time step index, representing the current iteration round, and t+1 represents the next iteration round; This indicates that the feedback signal is generated in round t. The weight of the rule for completing the i-th field at the current time. The preset learning rate, An indicator function for determining whether the completion rule for the i-th field is triggered in this recommendation. and These are the preset lower and upper limits for the individual weights, respectively; S63: Update the feature weights in the matching score according to the following formula. :

[0020] in, Let be the weight of the i-th normalized feature at time t. Let i be the i-th normalized eigenvalue; S64: Update the values ​​of each field in the user profile, resource supplier profile, and transportation service provider profile based on the updated field completion weight and matching score weight.

[0021] An AI- and user profiling-based system for automatic order generation and targeted delivery includes: The data acquisition module is used to obtain the data needed to build user profiles from internal systems and external data sources; The data cleaning and standardization module is used to receive the data collected by the data acquisition module and perform deduplication, missing value completion, outlier identification and encoding standardization on the data. The user profile generation module is used to receive the data processed by the data cleaning and standardization module and generate user profiles. The resource supplier and transportation service provider profiling module is used to receive the data processed by the data cleaning and standardization module and generate resource supplier profiles and transportation service provider profiles. The AI ​​demand parsing module is used to obtain the procurement demand text input by the user, identify the demand type of the procurement demand text and extract the procurement fields. The automatic purchase order generation module is used to receive the purchase fields parsed by the AI ​​demand parsing module and the user profile generated by the user profile generation module, merge the purchase fields and the user profile, fill in the missing fields and generate a structured purchase order; The field confidence judgment module is used to receive the structured purchase order generated by the purchase order automatic generation module, calculate the confidence of each purchase field in the structured purchase order, determine the field writing, confirmation or supplementation method according to the threshold, and return the judgment result to the purchase order automatic generation module. The matching and scoring module is used to receive the structured purchase order generated by the purchase order automatic generation module, the user profile generated by the user profile generation module, and the resource supplier profile and transportation service provider profile generated by the resource supplier and transportation service provider profile module, to recall candidate objects and calculate targeted sending scores, and to filter target resource suppliers or target transportation service providers. The purchase order targeted sending module is used to receive the structured purchase order generated by the purchase order automatic generation module and the target object filtered by the matching and scoring module, send the purchase order to the target object, and record the sending process and status receipt. The status compensation and data verification module is used to receive the sending status and receipt status recorded by the purchase order targeted sending module, handle sending failure, receipt timeout and status inconsistency anomalies, perform summary verification and auditing of key fields, and return the anomaly handling results and verification results to the purchase order targeted sending module and the purchase order automatic generation module. The feedback learning module is used to receive the recorded results from the purchase order targeted sending module and the transaction results and performance results from external input, generate feedback signals, and transmit the feedback signals to the user profile generation module, the resource supplier and transportation service provider profile module and the matching scoring module respectively, so as to update the weights of each profile and the matching score.

[0022] Preferably, when recalling candidate objects, the matching and scoring module uses a recall set for multi-way recall, and after deduplicating the recall results, they are sorted by the scoring submodule. The feedback learning module updates the field completion rule weights, candidate object recall weights, and matching score weights according to the learning rate and time decay coefficient based on user confirmation, user modification, supplier response, final transaction, order cancellation, on-time delivery, and delivery anomaly events, forming a closed-loop optimization of profile supplementation, candidate recall, score sending, and feedback learning. The state compensation and data verification module generates summary values ​​for key fields before transaction commit. If unauthorized modification of key fields is detected, the commit is interrupted and an audit log is recorded.

[0023] As can be seen from the above technical solution compared with the prior art, the present invention has the following beneficial effects: 1. This invention automatically completes purchase order fields by using user profiles. Users only need to enter a brief description of their needs to generate a complete purchase order. This solves the problem that users have to fill in a large number of fields such as product type, quantity, pick-up location, unloading location, delivery station, expected delivery time, vehicle requirements, settlement method and supplier range every time they publish a request. This significantly reduces the amount of manual input and improves the efficiency of publishing purchase requests.

[0024] 2. This invention expands the coverage of candidate objects through a multi-path recall strategy. It calculates targeted sending scores based on multiple dimensions such as cooperation strength, regional or route coverage, category matching degree, supply or vehicle availability, response time, performance score and risk deduction. Purchase orders are sent to the target object with the highest comprehensive matching degree first. This solves the problem of supplier selection relying on human experience, makes the selection results more scientific and objective, and improves the transaction success rate.

[0025] 3. This invention continuously updates the field completion weights and matching score weights by using the learning rate and time decay coefficient, and synchronously updates the data of each profile, forming a closed-loop optimization mechanism. This solves the problem that the system's recommendation logic is fixed and cannot be optimized by itself, and achieves continuous improvement in recommendation performance over time. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the method steps in an embodiment of the present invention; Figure 2 This is a schematic diagram of the data change process in S3 of this embodiment of the invention; Figure 3 This is a flowchart illustrating the process of generating user profiles and supplier profiles in an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0028] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention, but should not be used to limit the scope of the present invention.

[0029] This invention provides an AI-based and user profiling-based automatic order generation and targeted delivery system, comprising: The data acquisition module is used to obtain the data needed to build user profiles from internal systems and external data sources; The data cleaning and standardization module is used to receive the data collected by the data acquisition module and perform deduplication, missing value completion, outlier identification and encoding standardization on the data. The user profile generation module is used to receive the data processed by the data cleaning and standardization module and generate user profiles. The resource supplier and transportation service provider profiling module is used to receive the data processed by the data cleaning and standardization module and generate resource supplier profiles and transportation service provider profiles. The AI ​​demand parsing module is used to obtain the procurement demand text input by the user, identify the demand type of the procurement demand text and extract the procurement fields. The automatic purchase order generation module is used to receive the purchase fields parsed by the AI ​​demand parsing module and the user profile generated by the user profile generation module, merge the purchase fields and the user profile, fill in the missing fields and generate a structured purchase order; The field confidence judgment module is used to receive the structured purchase order generated by the purchase order automatic generation module, calculate the confidence of each purchase field in the structured purchase order, determine the field writing, confirmation or supplementation method according to the threshold, and return the judgment result to the purchase order automatic generation module. The matching and scoring module is used to receive the structured purchase order generated by the purchase order automatic generation module, the user profile generated by the user profile generation module, and the resource supplier profile and transportation service provider profile generated by the resource supplier and transportation service provider profile module, to recall candidate objects and calculate targeted sending scores, and to filter target resource suppliers or target transportation service providers. The purchase order targeted sending module is used to receive the structured purchase order generated by the purchase order automatic generation module and the target object filtered by the matching and scoring module, send the purchase order to the target object, and record the sending process and status receipt. The status compensation and data verification module is used to receive the sending status and receipt status recorded by the purchase order targeted sending module, handle sending failure, receipt timeout and status inconsistency anomalies, perform summary verification and auditing of key fields, and return the anomaly handling results and verification results to the purchase order targeted sending module and the purchase order automatic generation module. The feedback learning module is used to receive the recorded results from the purchase order targeted sending module and the transaction results and performance results from external input, generate feedback signals, and transmit the feedback signals to the user profile generation module, the resource supplier and transportation service provider profile module and the matching scoring module respectively, so as to update the weights of each profile and the matching score.

[0030] Example: In this embodiment, in the gas product procurement and transportation service transaction scenario of a certain gas-connected e-commerce platform, when users purchase gas products such as LNG, liquefied natural gas, CNG, and LPG, or purchase LNG transportation services, they typically need to fill in information such as product or service type, purchase quantity, pick-up location, unloading location, delivery terminal, expected delivery time, vehicle or tanker requirements, settlement method, and the scope of resource suppliers or carriers. Traditionally, users need to manually fill in or copy historical orders each time, which is cumbersome and prone to errors. Therefore, this invention introduces an AI-based and user-profile-based method and system for automatic order generation and targeted delivery to solve the above problems.

[0031] The following example illustrates the purchasing request posting process of a typical user on the Qihe.com online marketplace. Figure 1 The specific embodiments of the present invention are described in detail below.

[0032] S1: Acquire user-related data and perform cleaning and standardization processing to generate user profiles, resource supplier profiles, and transportation service provider profiles.

[0033] In this embodiment, the system first obtains various types of data required for building the profile from the internal system of the Qihe.com e-commerce platform and external data sources, such as... Figure 3 As shown.

[0034] S11: Obtain basic user information, enterprise authentication information, user location, delivery address, frequently used stations, historical transaction orders, historical transaction volume, transaction categories, transaction prices, payment or settlement status, frequently cooperating resource suppliers, frequently cooperating transportation service providers, supplier quotation records, carrier quotation records, message receipt records, and performance records from internal systems; at the same time, obtain regional market prices, basic category data, basic station data, route distance data, and third-party market data from external data sources.

[0035] For example, the system can extract information such as the category, quantity, price, counterparty, and delivery station of all transactions completed by a user within the past year from their historical transaction orders.

[0036] S12: The system cleans and standardizes the collected raw data. Specifically, it deduplicates user basic information, historical transaction orders, supplier quotation records, carrier quotation records, and performance records, deleting duplicate records; it completes missing fields, identifies and labels outliers; and it maps product category names from different sources to a unified code. For example, "LNG" and "liquefied natural gas" are uniformly mapped to the same category code, "CNG" and "compressed natural gas" are uniformly mapped to the same category code, and "LPG" and "liquefied petroleum gas" are uniformly mapped to the same category code; at the same time, regional codes, station codes, supplier codes, and transportation service provider codes are mapped to standard coding systems respectively; and different units of measurement are converted into standard units that the system can recognize. For example, units such as tons, kilograms, cubic meters, and train trips can be uniformly converted into the system's preset standard units of measurement.

[0037] S13: Generate user profiles based on the cleaned data. Specifically, the system extracts the user's location, frequently used stations (statistically counting the usage frequency of each delivery station and selecting the station with the highest frequency as the frequently used station), frequently purchased categories (statistically counting the purchase frequency and quantity of each category), historical transaction volume (statistically counting the total transaction volume of the past 30 days, 90 days, and 180 days respectively), frequently cooperating resource suppliers (statistically counting the number of times of cooperation and transaction amount of each supplier, sorted by comprehensive score), and frequently cooperating transportation service providers (statistically counting the number of times of cooperation and total transportation volume of each transportation service provider, sorted by comprehensive score) from the cleaned data, generates a complete user profile, and stores it.

[0038] S14: Generate resource supplier profiles and transportation service provider profiles. Resource supplier profiles should include at least the following: available product categories (the available product categories and corresponding supply capabilities registered by the supplier in the system), service area (the geographical area covered by the supplier and its delivery radius), historical transaction volume (the total transaction volume of the supplier in the past 30, 90, and 180 days), and performance score (calculated based on factors such as delivery timeliness, product quality consistency, and after-sales service response speed in historical transactions). The profile of a transport service provider should include at least the transportable area (the operating area and routes registered by the transport service provider), frequently used routes (high-frequency routes based on historical transport records), vehicle availability status (the number of vehicles registered by the transport service provider in the system, driver scheduling and current vehicle location information), and performance score (calculated based on factors such as on-time rate, cargo integrity rate, and abnormal event occurrence rate in historical transport).

[0039] Compared to existing technologies that rely solely on users manually filling in or simply copying historical orders, this embodiment constructs a multi-dimensional profile system encompassing user transaction habits, supplier service capabilities, and transportation capacity providers' capabilities through multi-source data fusion and standardized processing. This lays a data foundation for subsequent intelligent purchase order generation and precise matching. Users no longer need to repeatedly fill in information such as location, frequently used stations, and frequently purchased product categories each time they post a request; the system can automatically fill in these fields based on the profile.

[0040] S2: Obtain the procurement requirement text input by the user, identify the requirement type of the procurement requirement text using a hybrid parsing method, and extract the procurement fields.

[0041] In this embodiment, when a user logs into the Qihe.com online mall and enters the quick demand posting page, the system reads the generated user profile and waits for the user to input their purchasing needs.

[0042] S21: The system uses a domain dictionary to perform word segmentation and entity recognition on the user-input procurement requirement text. The domain dictionary pre-includes business entities such as LNG, liquefied natural gas, CNG, LPG, station name, vehicle type, tanker volume, and settlement method, as well as their synonyms and aliases. After segmenting the text, the system matches the segmentation results with the domain dictionary to identify the business entities in the text.

[0043] For example, when a user enters "deliver a truckload of LNG to a frequently used station tomorrow afternoon", the system recognizes "LNG" as the product category entity, "tomorrow afternoon" as the time entity, and "frequently used station" as the location entity.

[0044] S22: Input the procurement demand text into the intent classification model and calculate the probability P(y|x) of each demand type, where x is the input procurement demand text, and y∈{resource procurement, transportation service, mixed demand}. The intent classification model uses a language model fine-tuned from the historical procurement text corpus of Qihe.com Mall to map the input text into three categories: resource procurement, transportation service, or mixed demand.

[0045] For example, when a user enters "find an old supplier to quote the price of one truckload of LNG," the model calculates a probability of 0.92 for resource procurement, 0.05 for transportation services, and 0.03 for mixed demand, using resource procurement as the predicted outcome. When the highest probability falls below a preset probability threshold (e.g., 0.70), the system switches to the rules engine for a fallback decision. For instance, when a user enters short text like "needs a truck and cargo" and the model's probabilities for each type are low, the rules engine takes over the processing.

[0046] S23: The rule engine calculates the resource procurement keyword score and the transportation service keyword score according to the preset category dictionary, action dictionary, location dictionary, time dictionary, vehicle dictionary and settlement dictionary respectively.

[0047] Rule matching score The calculation formula is:

[0048] in, The weight of the k-th keyword. This is an indicator function for whether the k-th keyword is matched; when the difference between the resource procurement score and the transportation service score is greater than the classification threshold, it is determined to be a resource procurement demand; when the difference between the transportation service score and the resource procurement score is greater than the classification threshold, it is determined to be a transportation service demand; when the difference between the two scores is less than or equal to the classification threshold, the user is prompted to confirm the demand type. For example, when the text contains keywords such as "quotation," "price," and "gas source," the resource procurement score is higher; when it contains keywords such as "haul," "transport," and "tanker," the transportation capacity service score is higher.

[0049] S24: Extract procurement fields from the procurement requirement text using a field sequence labeling model and the BIO labeling method. The field sequence labeling model adopts a sequence labeling model based on BIO labeling, labeling each character in the text with B (Begin), I (Inside), or O (Outside), and labeling the corresponding field type. The system extracts fields such as procurement category, quantity, pick-up location, unloading location, delivery station, and expected delivery time from the labeling results.

[0050] For example, for the input text "Deliver a truckload of LNG to a frequently used site tomorrow afternoon," the system labels "LNG" as the category field, "a truckload" as the quantity field, "tomorrow afternoon" as the expected delivery time field, and "frequently used site" as the delivery site field. For non-natural language input methods, such as when a user uploads a request via image, the system first converts the image into text using OCR before executing the above parsing process; when a user inputs via voice, the system first converts the audio into text using speech-to-text before executing the parsing process.

[0051] Compared to existing technologies that rely on manual filling in of each procurement field, this embodiment uses a hybrid parsing approach combining a domain dictionary, intent classification model, and field sequence labeling model. This allows the system to automatically identify the type of demand and extract key procurement fields from the user's natural language descriptions, quick templates, historical order repurchases, shopping cart data, image recognition results, or speech recognition results. Even if the input text is short or non-standard, the system can ensure the robustness of the parsing through a combination of rule fallback and model prediction.

[0052] S3: Merge the procurement fields obtained from AI parsing with the user profile, use the user profile to complete the missing fields in the procurement fields, calculate the field confidence score for each procurement field, and determine whether to automatically write the field, prompt the user for confirmation, or require the user to supplement it based on the comparison result of the field confidence score and a preset threshold. Figure 2 As shown, a structured purchase order is finally generated.

[0053] In this embodiment, the AI ​​demand parsing module transmits the procurement fields parsed in step S2 to the automatic purchase order generation module, while the automatic purchase order generation module simultaneously reads the user profile data from the user profile generation module.

[0054] S31: When there are missing fields in the procurement field, read the corresponding frequently used data from the user profile, and select the most matching value to fill in the procurement order according to the frequency of use or comprehensive sorting.

[0055] Specifically, when the delivery site field is missing in the procurement request text entered by the user, the system reads the frequently used site data from the user profile, sorts them from high to low frequency of use, and adds the most frequently used site as the delivery site to the procurement order. When the user's input procurement request text is missing a range of candidate resource suppliers, the system reads data of regularly cooperating resource suppliers from the user profile, sorts them comprehensively according to historical transaction volume and most recent cooperation time, and selects the top-ranked suppliers to supplement the candidate range of the procurement order. When the user's input purchase request text lacks a range of candidate transportation service providers, the system reads the data of regularly cooperating transportation service providers from the user profile and adds them to the purchase order; when the user's input purchase request text lacks the expected delivery time, the system reads the commonly used delivery time from the user profile and generates a recommended delivery time based on the current time.

[0056] For example, if a user simply enters "find quotes from existing suppliers", the system will automatically complete the list of frequently used delivery sites and regular suppliers based on the user's profile.

[0057] S32: Obtain the model's probability of extracting field f. Rules for scoring Consistency between this field and the user profile The stability of this field in historical orders and the degree of field conflict ; Specifically, among them, The confidence probability for the field sequence labeling model to identify this field; This is the score given by the rules engine for matching this field; This indicates the degree of match between the field value and the corresponding field in the user profile; This indicates the frequency and consistency of this field in the user's historical orders; This indicates whether there is a logical conflict between the value of this field and other determined field values.

[0058] S33: Calculate the field confidence score using the following formula. :

[0059] in, For the field confidence normalization function, For bias terms, to These are the weighting coefficients for each factor; For example, if the model has a high probability of extracting the "LNG" category (0.95), and this category is highly consistent with the frequently purchased categories in the user profile, and this category appears consistently in historical orders, then the confidence level is high. Conversely, if the delivery station extracted by the model is inconsistent with the frequently used stations in the user profile, and there is an unreasonable distance conflict between the station and the pickup location, the confidence level of this field will decrease due to the conflict penalty.

[0060] S34: Calculate the field confidence scores With the preset first threshold Second threshold Comparison :when When the system automatically writes this field into the purchase order; when When the system prompts the user for confirmation or modification before the purchase order is submitted; When this happens, the system marks the field as missing or abnormal and prompts the user to complete it.

[0061] For example, if a user has multiple frequently used depots in the past 90 days with similar usage frequencies, the stability of the delivery depot field is low, and Conf (delivery depot) may fall between T2 and T1. The system will mark this field as pending confirmation. If multiple fleets exist for frequently used routes with significant differences in vehicle scheduling, the capacity service provider field will be marked as pending confirmation. After the user confirms or adds information, the system updates the confidence model parameters for this field based on the user's actions.

[0062] Compared to existing technologies where users need to fill in all fields one by one, which can easily lead to low-quality purchase order data due to missing or non-standard fields, this embodiment automatically completes missing fields through user profiling and uses a confidence level judgment mechanism to classify the automatically generated fields. High-confidence fields are written directly, medium-confidence fields prompt for confirmation, and low-confidence fields require supplementation. This ensures both the efficiency of purchase order generation and the accuracy of key fields, avoiding obstacles to subsequent procurement processes due to automatic completion errors.

[0063] S4: Based on the structured purchase order, combined with the resource supplier profile and the transportation service provider profile, candidate objects are recalled, a targeted sending score is calculated, and the target resource supplier or target transportation service provider is selected based on the targeted sending score.

[0064] In this embodiment, after the structured purchase order is generated, the matching and scoring module receives purchase order data, user profile data, resource supplier profile data, and transportation service provider profile data, and performs candidate object recall and scoring screening.

[0065] S41: Simultaneously employ multiple recall methods to recall candidate entities from both the resource supplier profile and the transportation service provider profile. Specific recall methods include: Historical Partner Recall – Retrieve suppliers or transportation service providers with historical transaction records with the user; Regional or Terminal Coverage Recall – Retrieve suppliers or transportation service providers covering the user's location, pick-up location, or delivery terminal; Category or Service Type Matching Recall – Retrieve suppliers or transportation service providers capable of providing the purchased category or service type; Route Coverage Recall – Retrieve transportation service providers covering routes between pick-up and unloading locations or suppliers with delivery radius coverage capabilities; Profile Similarity Recall – Retrieve other suppliers or transportation service providers frequently cooperated with by the user whose profile similarity is higher than a preset threshold; Recently Active and Qualified Retrieval – Retrieve suppliers or transportation service providers with recent quotation or transportation records and whose qualifications are still valid. This multi-pronged recall strategy avoids overlooking new suppliers or fleets due to relying solely on historical partnerships.

[0066] S42: Take the union of the candidate sets obtained from multiple recall methods:

[0067] in, A recall assembly for historical partners. For the recall set of objects covered by the region or site, Recall a set of objects matching the category or service type. For the recall set of objects covered by the line, For the image similarity object recall set, The system first retrieves a set of recently active and qualified candidates, then deduplicates the union of these sets to obtain the final candidate set. For example, if a supplier is listed in both historical partners and regional coverage groups, the system retains only one record.

[0068] S43: For candidate object B, obtain its set of normalized eigenvalues. The normalized feature values ​​include cooperation strength, regional or route coverage, category or service type matching degree, supply capacity or vehicle availability, response time and performance score; S44: The strength of cooperation Calculate using the following formula:

[0069] in, This represents the number of transactions between user u and candidate B in the past a days. This represents the transaction amount over the past day (a). The time since the most recent transaction. The time decay factor is a preset factor; by introducing the time decay factor, more recent transactions contribute more to the strength of cooperation, thus more accurately reflecting the current cooperative relationship.

[0070] The repurchase rate is defined as the proportion of transactions between user u and candidate B in their historical transaction history where the time interval between two consecutive transactions is less than a preset threshold. to The weighting coefficients and , This represents the eigenvalue normalization function; S45: Price Deviation Calculate using the following formula:

[0071] in, Provide quotes to suppliers. This refers to the user's historical transaction price or regional market reference price; price deviation measures the degree of deviation between the supplier's quotation and the market reference price, and if the deviation is too large, it is used as the basis for risk deduction.

[0072] S46: Calculate the targeted delivery score according to the following formula. :

[0073] in, Let be the weight coefficient of the i-th normalized feature. Let i be the normalized feature value of candidate object B. The weighting coefficient for risk deduction. This is a risk deduction item for candidate B; The risk deduction items are obtained by combining factors such as expired qualifications, abnormal transportation records, default records, abnormal price deviations, and delayed receipts. Specifically, risk deductions will be added when a candidate's hazardous chemical transportation qualification is nearing or has expired; when a candidate has an abnormal transportation record; when a candidate has a default record; when a candidate's quoted price deviates significantly from market prices; and when a candidate has a high number of historical timeouts in its delivery receipts. The weights of each risk factor can be configured according to business rules.

[0074] S47: Sort the candidates according to their targeted delivery scores from highest to lowest, and select a predetermined number of the top-ranked candidates as target resource suppliers or target transportation service providers. For example, for resource procurement needs, the system selects the top-scoring suppliers as target suppliers; for transportation service needs, the system selects the top-scoring transportation service providers as target providers. If a supplier or transportation service provider has a high number of transactions in the past 180 days, large transaction amounts, short time since the most recent transaction, high repurchase rate, and covers the user's region and procurement categories, then its cooperation strength and coverage scores are high, and it is likely to be prioritized.

[0075] Compared to existing technologies where users need to manually select suppliers or transportation service providers based on experience, this embodiment ensures the coverage of candidates through multi-channel recall, achieves accurate matching through multi-dimensional feature scoring, and automatically filters out entities with qualification issues or performance risks through risk deduction. This makes the screening results more scientific, objective, and comprehensive, reduces the manual screening workload for users, and improves the accuracy of supplier outreach.

[0076] S5: Send the structured purchase order to the target resource supplier or target transportation service provider, generate purchase order number and sending batch number, record sending status, receipt status and quotation status, and perform status compensation operation when sending fails or receipt times out.

[0077] In this embodiment, the purchase order targeted sending module receives the structured purchase order generated by the purchase order automatic generation module and the target object list filtered by the matching and scoring module, and then executes the targeted sending of the purchase order.

[0078] S51: When a purchase order fails to send, the system automatically retryes sending. Specifically, the system generates a unique purchase order number and batch number, encapsulates the purchase order content into a message, and prioritizes sending it via internal messaging. If internal messaging fails, the system automatically retryes. After the number of retries reaches a preset limit, it switches to the SMS channel, WeChat Work channel, or open API channel to resend. Each messaging channel is prioritized; if a message fails to send through the current channel, it automatically degrades to the next channel. If all messaging channels fail, the system records the failure status and triggers a manual intervention alert.

[0079] For example, when the target supplier's internal message is unread and fails to be delivered after multiple retries, the system automatically switches to the SMS channel to send the procurement notification.

[0080] S52: When a resource supplier or transportation service provider's response or quotation times out, the system records the timeout status and triggers a supplementary delivery or manual intervention reminder. The system has preset response and quotation waiting time windows. If the target entity does not respond or quote within the window, the system automatically records the timeout status and feeds the timeout information back to the matching and scoring module to update the response timeliness characteristic value of that entity. Simultaneously, the system can trigger a supplementary delivery, sending the purchase order to the second-highest-scoring candidate in the candidate list, or trigger a manual intervention reminder for operations personnel to contact the entity offline.

[0081] S53: When there is a discrepancy between the sending status, receipt status, quotation status, carrier confirmation status, and order status, the system performs a status alignment operation and records the status change log.

[0082] For example, if the system records the sending status as "sent" but the receipt status as "timeout," and the supplier has confirmed receipt of the purchase order but has not processed the receipt in the system, the system will synchronize the receipt status to "received" through a status alignment operation. All status changes are recorded in a complete audit log, including the change time, operation source, and status values ​​before and after the change.

[0083] S54: Generate summary values ​​for key fields before committing the transaction. If unauthorized modification of key fields is detected, the commit is interrupted and an audit log is logged. Key fields include purchase order number, category, quantity, pick-up location, unloading location, delivery station, target object set, and quoted amount.

[0084] Specifically, the system concatenates key fields and calculates a hash value as the first digest value when the purchase order is generated; before sending the purchase order to the target object, it calculates the hash value again as the second digest value; the two are compared for consistency. If they do not match, it is determined that the key fields of the purchase order have been modified without authorization after generation and before sending, the sending process is interrupted and an audit log is recorded. After the target object receives the receipt or quotation, the system generates a third digest value for the key fields in the receipt or quotation, compares it with the key fields of the purchase order at the time of sending, and if they do not match, an abnormal receipt is recorded and manual review is triggered.

[0085] Compared to existing technologies where the status of purchase orders is difficult to track after they are sent, failures cannot be automatically recovered, and data lacks anti-tampering mechanisms, this embodiment ensures the reliability of the transmission through multi-channel automatic retries, improves the procurement response rate through timeout detection and alternative supplementation, solves the problem of inconsistent status at each stage through a status alignment mechanism, and prevents unauthorized data tampering through key field summary verification. This significantly improves the reliability, traceability, and data security of the purchase order transmission process.

[0086] S6: Generate feedback signals based on the procurement process results corresponding to the quoted status and sending status, and use the feedback signals to update each profile and adjust the field extraction weights and matching score weights.

[0087] In this embodiment, after the procurement process is completed, the feedback learning module receives the recorded results from the purchase order targeted sending module and the transaction results and performance results from external input, generates feedback signals, and updates the weights and profile data of each module.

[0088] S61: Obtain feedback signals for this procurement process When the user adopts the recommendation, the supplier responds promptly, the quote is accepted, the order is completed and fulfillment is normal, Take a positive value; when a user modifies a recommendation field, deletes a recommendation object, a supplier fails to respond within a timeout period, or an order is canceled or there is a fulfillment error. Take a negative value; the absolute value of the feedback signal can reflect the strength of the feedback. For example, take a larger positive value when the order is completed and the performance evaluation is excellent, and take a larger negative value when the supplier fails to respond within the time limit.

[0089] S62: Update the weight of field completion or rule matching according to the following formula. :

[0090] Where t is the time step index, representing the current iteration round, and t+1 represents the next iteration round; This indicates that the feedback signal is generated in round t. The weight of the rule for completing the i-th field at the current time. The preset learning rate, An indicator function for determining whether the completion rule for the i-th field is triggered in this recommendation. and These are preset lower and upper limits for individual weights to prevent excessive weight drift. For example, if the delivery site automatically recommended by the system is changed to another site multiple times by the user, the weight of the delivery site completion rule will decrease over time; if the system-recommended regular supplier is adopted by the user and a deal is ultimately made, the weight of that supplier recommendation rule will increase.

[0091] S63: Update the feature weights in the matching score according to the following formula. :

[0092] in, Let be the weight of the i-th normalized feature at time t. Let be the i-th normalized feature value. If a supplier responds promptly multiple times and has a high success rate, the system increases the matching weight of features such as response timeliness and performance rating in scenarios with similar users, similar product categories, or similar routes. Updating feature weights allows the scoring model to continuously adapt to dynamic changes in business.

[0093] S64: Update the values ​​of each field in the user profile, resource supplier profile, and capacity service provider profile based on the updated field completion weight and matching score weight.

[0094] For example, newly completed order data is included in historical transaction records, and the user's transaction volume, average purchase frequency, and comprehensive ranking of frequently cooperating suppliers and transportation service providers over the past 30 / 90 / 180 days are recalculated; the transaction volume and performance score of newly acquired suppliers or transportation service providers are updated simultaneously; newly added risk events (such as overdue receipts and default records) are recorded in the risk record field of the corresponding profile. Updates to all profile fields record the data source and update time.

[0095] Through the above steps, the system completes the acquisition of feedback signals, the updating of weights, and the iterative optimization of user profiles. Compared with existing technologies where the system recommendation logic is fixed and cannot self-optimize based on actual transaction results, this embodiment uses a feedback learning mechanism to continuously transmit user confirmation or modification behavior, supplier response behavior, and order completion and fulfillment results back to the system. This allows for continuous iterative optimization of field completion rules, candidate recall weights, and matching score weights. User profiles and supplier profiles are dynamically updated with transactions, achieving a closed-loop evolution effect of "becoming more accurate with use." The efficiency and matching accuracy of subsequent procurement demand releases continue to improve as the system runs longer.

[0096] In summary, the specific implementation of this invention forms a complete and self-optimizing solution for the automatic generation and targeted delivery of procurement orders through six interconnected steps: data collection and profile generation, AI-powered demand analysis, automatic purchase order generation and confidence assessment, candidate retrieval and matching scoring, targeted purchase order delivery and status compensation, and feedback learning and weight updates. When users publish procurement needs on the Qihe.com e-commerce platform, they only need to enter a brief natural language description or select a shortcut template, and the system can automatically complete the entire process of purchase order generation, supplier screening, targeted delivery, and status tracking, significantly improving the efficiency of procurement demand publication and the accuracy of supplier outreach.

[0097] It is understood that the systems, devices, and storage media provided in the embodiments of the present invention correspond to the methods provided in the embodiments of the present invention, and the explanations, examples, and beneficial effects of the relevant content can be referred to the corresponding parts of the above methods.

[0098] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another.

[0099] For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media.

[0100] The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid state disks (SSDs)).

[0101] It should be noted that in this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0102] Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0103] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0104] The embodiments of the present invention are given for the purposes of illustration and description. Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for automatic order generation and targeted delivery based on AI and user profiling, characterized in that, Perform the following steps using a computer device: S1: Acquire user-related data and perform cleaning and standardization processing to generate user profiles, resource supplier profiles, and transportation service provider profiles; S2: Obtain the procurement requirement text input by the user, identify the requirement type of the procurement requirement text and extract the procurement field using a hybrid parsing method; S3: The procurement fields obtained by AI parsing are merged with the user profile. The missing fields in the procurement fields are filled in using the user profile. The field confidence of each procurement field is calculated. Based on the comparison result of the field confidence and the preset threshold, it is determined whether the field is automatically written, prompted to confirm by the user, or asked to supplement by the user. Finally, a structured procurement order is generated. S4: Based on the structured purchase order, combined with the resource supplier profile and the transportation service provider profile, candidate objects are recalled, a targeted sending score is calculated, and the target resource supplier or target transportation service provider is selected based on the targeted sending score. S5: Send the structured purchase order to the target resource supplier or target transportation service provider, generate purchase order number and sending batch number, record sending status, receipt status and quotation status, and perform status compensation operation when sending fails or receipt times out. S6: Generate feedback signals based on the procurement process results corresponding to the quoted status and sending status, and use the feedback signals to update each profile and adjust the field extraction weights and matching score weights.

2. The method for automatic order generation and targeted delivery based on AI and user profiling as described in claim 1, characterized in that: S1 specifically includes the following sub-steps: S11: Obtain user-related data from internal systems and external data sources. The user-related data includes basic user information, enterprise authentication information, user location, delivery address, frequently used stations, historical transaction orders, historical transaction volume, transaction categories, transaction prices, payment or settlement status, frequently cooperating resource suppliers, frequently cooperating transportation service providers, supplier quotation records, carrier quotation records, message receipt records, performance records, regional market prices, category basic data, station basic data, route distance data, and third-party market data. S12: Perform deduplication, missing value completion, outlier identification, and coding standardization on the collected data. Map category names, regional codes, station codes, supplier codes, and capacity service provider codes from different sources to a unified code, and convert different units of measurement into standard units that the system can recognize. S13: Generate a user profile based on the cleaned data. The user profile includes the user's location, frequently used stations, frequently purchased product categories, historical transaction volume, frequently cooperating resource suppliers, and frequently cooperating transportation service providers. S14: Generate resource supplier profiles and transportation service provider profiles. The resource supplier profiles include available product categories, service areas, historical transaction volume, and performance ratings. The transportation service provider profiles include available transport areas, frequently used routes, vehicle availability status, and performance ratings.

3. The method for automatic order generation and targeted delivery based on AI and user profiling as described in claim 2, characterized in that: S2 specifically includes the following sub-steps: S21: Use a domain dictionary to segment and identify entities in the procurement requirement text, and identify business entities, including LNG, liquefied natural gas, CNG, LPG, station, vehicle type, tanker volume and settlement method; S22: Input the procurement requirement text into the intent classification model and calculate the probability of each requirement type. , where x is the input procurement requirement text. ; The type with the highest probability is used as the model prediction result; when the highest probability is lower than the preset probability threshold, the system switches to the rule engine for a fallback judgment. S23: The rule engine calculates the keyword scores for resource procurement and transportation service, as well as the rule matching score, according to a preset dictionary. The calculation formula is: in, The weight of the k-th keyword. This is an indicator function for whether the k-th keyword is matched; when the difference between the resource procurement score and the transportation service score is greater than the classification threshold, it is determined to be a resource procurement demand; when the difference between the transportation service score and the resource procurement score is greater than the classification threshold, it is determined to be a transportation service demand; when the difference between the two scores is less than or equal to the classification threshold, the user is prompted to confirm the demand type. S24: Extract procurement fields from the procurement requirement text using the BIO annotation method through the field sequence annotation model. The procurement fields include procurement category, quantity, pick-up location, unloading location, delivery station, and expected delivery time.

4. The method for automatic order generation and targeted delivery based on AI and user profiling as described in claim 3, characterized in that: S3 specifically includes the following sub-steps: S31: When there are missing fields in the procurement fields, read the corresponding frequently used data from the user profile, and select the most matching value according to the frequency of use or comprehensive sorting to fill in the procurement order; S32: Obtain the model's probability of extracting field f. Rules hit score Consistency between this field and the user profile The stability of this field in historical orders and the degree of field conflict ; S33: Calculate the field confidence score using the following formula. : in, For the field confidence normalization function, For bias terms, to These are the weighting coefficients for each factor; S34: Calculate the field confidence scores With the preset first threshold Second threshold Comparison :when When the system automatically writes this field into the purchase order; when When the system prompts the user for confirmation or modification before the purchase order is submitted; When this happens, the system marks the field as missing or abnormal and prompts the user to complete it.

5. The method for automatic order generation and targeted delivery based on AI and user profiling as described in claim 1, characterized in that: The candidate object recall process in S4 specifically includes the following sub-steps: S41: Simultaneously employ multiple recall methods to recall candidate objects from resource supplier profiles and capacity service provider profiles. The recall methods include recall of historical cooperation objects, recall of regional or station coverage objects, recall of category or service type matching objects, recall of route coverage objects, recall of profile similarity objects, and recall of recently active and qualified objects. S42: Take the union of the candidate sets obtained from multiple recall methods: in, A recall assembly for historical partners. For the recall set of objects covered by the region or site, Recall a set of objects matching the category or service type. For the recall set of objects covered by the line, For the image similarity object recall set, The set of recently active and qualified objects is retrieved, and then the union of the results C is deduplicated to obtain the final set of candidate objects.

6. The method for automatic order generation and targeted delivery based on AI and user profiling as described in claim 5, characterized in that: The calculation of the targeted transmission score in step S4 specifically includes the following sub-steps: S43: For candidate object B, obtain its set of normalized eigenvalues. The normalized feature values ​​include cooperation strength, regional or route coverage, category or service type matching degree, supply capacity or vehicle availability, response time and performance score; S44: The strength of cooperation Calculate using the following formula: in, This represents the number of transactions between user u and candidate B in the past a days. This represents the transaction amount over the past day (a). The time since the most recent transaction. The preset time decay factor; The repurchase rate is defined as the proportion of transactions between user u and candidate B in their historical transaction history where the time interval between two consecutive transactions is less than a preset threshold. to The weighting coefficients and , This represents the eigenvalue normalization function; S45: Price Deviation Calculate using the following formula: in, Provide quotes to suppliers. This refers to the user's historical transaction price or regional market reference price. S46: Calculate the targeted delivery score according to the following formula. : in, Let be the weight coefficient of the i-th normalized feature. Let i be the normalized feature value of candidate object B. The weighting coefficient for risk deduction. This is a risk deduction item for candidate B; The risk deduction items are obtained by combining factors such as expired qualifications, abnormal transportation records, default records, abnormal price deviations, and delayed receipts. S47: Sort the candidate objects from high to low according to the targeted delivery score, and select a preset number of the top-ranked candidate objects as the target resource supplier or target transportation service provider.

7. The method for automatic order generation and targeted delivery based on AI and user profiling as described in claim 1, characterized in that: S5 specifically includes the following sub-steps: S51: When a purchase order fails to be sent, the system will automatically retry sending it. After the number of retries reaches the preset limit, the system will switch to a different message channel to resend the order. The message channels include internal messages, SMS, WeChat Work, and open APIs. S52: When the reply from a resource supplier or transportation service provider times out or the quotation times out, the system records the timeout status and triggers a supplementary resend or manual intervention reminder; S53: When there is a discrepancy between the sending status, receipt status, quotation status, carrier confirmation status and order status, the system performs a status alignment operation and records the status change log. S54: Generate summary values ​​for key fields before transaction commit. If unauthorized modification of the key fields is detected, the commit is interrupted and an audit log is recorded. The key fields include purchase order number, category, quantity, pick-up location, unloading location, delivery station, target object set, and quotation amount.

8. The method for automatic order generation and targeted delivery based on AI and user profiling as described in claim 1, characterized in that: S6 specifically includes the following sub-steps: S61: Obtain feedback signals for this procurement process When the user adopts the recommendation, the supplier responds promptly, the quote is accepted, the order is completed and fulfillment is normal, Take a positive value; when a user modifies a recommendation field, deletes a recommendation object, a supplier fails to respond within a timeout period, or an order is canceled or there is a fulfillment error. Take the negative value; S62: Update the weight of field completion or rule matching according to the following formula. : Where t is the time step index, representing the current iteration round, and t+1 represents the next iteration round; This indicates that the feedback signal is generated in round t. The weight of the rule for completing the i-th field at the current time. The preset learning rate, An indicator function for determining whether the completion rule for the i-th field is triggered in this recommendation. and These are the preset lower and upper limits for the individual weights, respectively; S63: Update the feature weights in the matching score according to the following formula. : in, Let be the weight of the i-th normalized feature at time t. Let i be the i-th normalized eigenvalue; S64: Update the values ​​of each field in the user profile, resource supplier profile, and transportation service provider profile based on the updated field completion weight and matching score weight.

9. An AI- and user profiling-based automatic order generation and targeted delivery system, used to implement the AI- and user profiling-based automatic order generation and targeted delivery method of any one of claims 1-8, characterized in that, include: The data acquisition module is used to obtain the data needed to build user profiles from internal systems and external data sources; The data cleaning and standardization module is used to receive the data collected by the data acquisition module and perform deduplication, missing value completion, outlier identification and encoding standardization on the data. The user profile generation module is used to receive the data processed by the data cleaning and standardization module and generate user profiles. The resource supplier and transportation service provider profiling module is used to receive the data processed by the data cleaning and standardization module and generate resource supplier profiles and transportation service provider profiles. The AI ​​demand parsing module is used to obtain the procurement demand text input by the user, identify the demand type of the procurement demand text and extract the procurement fields. The automatic purchase order generation module is used to receive the purchase fields parsed by the AI ​​demand parsing module and the user profile generated by the user profile generation module, merge the purchase fields and the user profile, fill in the missing fields and generate a structured purchase order; The field confidence judgment module is used to receive the structured purchase order generated by the purchase order automatic generation module, calculate the confidence of each purchase field in the structured purchase order, determine the field writing, confirmation or supplementation method according to the threshold, and return the judgment result to the purchase order automatic generation module. The matching and scoring module is used to receive the structured purchase order generated by the purchase order automatic generation module, the user profile generated by the user profile generation module, and the resource supplier profile and transportation service provider profile generated by the resource supplier and transportation service provider profile module, to recall candidate objects and calculate targeted sending scores, and to filter target resource suppliers or target transportation service providers. The purchase order targeted sending module is used to receive the structured purchase order generated by the purchase order automatic generation module and the target object filtered by the matching and scoring module, send the purchase order to the target object, and record the sending process and status receipt. The status compensation and data verification module is used to receive the sending status and receipt status recorded by the purchase order targeted sending module, handle sending failure, receipt timeout and status inconsistency anomalies, perform summary verification and auditing of key fields, and return the anomaly handling results and verification results to the purchase order targeted sending module and the purchase order automatic generation module. The feedback learning module is used to receive the recorded results from the purchase order targeted sending module and the transaction results and performance results from external input, generate feedback signals, and transmit the feedback signals to the user profile generation module, the resource supplier and transportation service provider profile module and the matching scoring module respectively, so as to update the weights of each profile and the matching score.

10. The AI-based and user profiling-based automatic order generation and targeted delivery system as described in claim 9, characterized in that: When recalling candidate objects, the matching and scoring module uses a recall set for multi-way recall, and after deduplicating the recall results, it submits them to the scoring submodule for sorting. The feedback learning module updates the field completion rule weights, candidate object recall weights, and matching score weights according to the learning rate and time decay coefficient based on user confirmation, user modification, supplier response, final transaction, order cancellation, on-time delivery, and delivery anomaly events, forming a closed-loop optimization of profile supplementation, candidate recall, score sending, and feedback learning. The state compensation and data verification module generates summary values ​​for key fields before transaction commit. If unauthorized modification of key fields is detected, the commit is interrupted and an audit log is recorded.

Images