Reverse order full-link identification and risk control method and device, equipment and storage medium

Abstract

The invention relates to the technical field of information processing, in particular to a reverse order full-link identification and risk control method, device and equipment and a storage medium, and the method comprises the steps: regularly pulling order basic information of a merchant from a plurality of open APIs, and employing a CRF model to identify a waybill reverse label in the order basic information, cross-platform reverse order data streams are automatically processed, the multi-source heterogeneous data integration capability is achieved, the data processing efficiency is improved, accurate recognition and standardized management of reverse order customer information are achieved, sending and receiving information corresponding to reverse waybill numbers is inquired, whole-process processing of order data is covered, the data management quality is improved, and the data management efficiency is improved. Calling a naive Bayesian classification model to calculate a first risk probability of the sending and receiving information, improving the intelligent degree of risk identification, fusing the first risk probability and the historical total return rate of the user to obtain a second risk probability, and generating abnormal order interception information when the second risk probability is greater than a preset risk threshold value. And risk control decision-making efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and in particular to a method, apparatus, equipment, and storage medium for full-link identification and risk control of reverse orders. Background Technology

[0002] In the context of e-commerce reverse order processing, merchants need to accurately identify customer information to optimize operations and control risks. The current manual processing mode is inefficient and prone to errors. Especially when facing reverse orders from multiple platforms, there is a lack of automated support for tracking number extraction and customer information integration. It is currently difficult to automate the processing of cross-platform reverse order data streams. In particular, the tracking number extraction and customer information integration stages rely heavily on manual rule configuration, resulting in low processing efficiency, low fault tolerance, inability to adapt to dynamic business scenarios, and a lack of multi-source heterogeneous data integration capabilities.

[0003] Meanwhile, the screening of risky keywords still relies on manual experience, which is time-consuming and labor-intensive, making it difficult to detect abnormal orders in a timely manner. The lack of semantic-level NLP model support results in insufficient intelligence in risk identification, making it impossible to achieve real-time anomaly detection. This leads to a persistently high rate of missed detections for high-risk orders and a continuous accumulation of operational risks.

[0004] Moreover, existing technologies have shortcomings in data cleaning, waybill processing, and information identification, and cannot cover the entire process of data acquisition, waybill processing, information identification, and storage. Data quality is uncontrollable and management is fragmented, leading to chaotic data management for merchants, resulting in a broken technology chain and data silos. This not only fails to meet the needs of efficient order processing, but also makes it difficult to support the closed loop of risk control decisions, and cannot achieve automated, accurate identification and standardized management of reverse order customer information. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the present invention aims to provide a method, device, equipment, and storage medium for full-link identification and risk control of reverse orders, which has the ability to integrate multi-source heterogeneous data, automatically process cross-platform reverse order data streams, improve data processing efficiency, cover the entire process of order data processing, improve data management quality, realize automated and accurate identification and standardized management of reverse order customer information, improve the intelligence level of risk identification, and improve the efficiency of risk control decisions.

[0006] The first aspect of this invention provides a method for full-link identification and risk control of reverse orders, comprising: periodically pulling basic order information from multiple open API interfaces, and deduplicating duplicate data of the basic order information from the same e-commerce platform; using a preset CRF model to identify reverse waybill tags in the basic order information, and extracting reverse waybill numbers from the corresponding orders based on the reverse waybill tags; calling a third-party logistics platform API to query the sender and receiver information corresponding to the reverse waybill number, and calling a preset Naive Bayes classification model to calculate a first risk probability of the sender and receiver information; matching the sender and receiver information with user identities from multiple e-commerce platforms, and obtaining the historical total return rate of the user identity in the multiple e-commerce platforms; fusing the first risk probability and the historical total return rate to obtain a second risk probability, and generating abnormal order interception information when the second risk probability is greater than a preset risk threshold.

[0007] Optionally, in a first implementation of the first aspect of the present invention, the step of periodically retrieving basic order information from multiple open API interfaces and deduplicating duplicate data of the basic order information from the same e-commerce platform includes: configuring a distributed task scheduling system to trigger multiple parallel crawler tasks at preset time intervals, wherein the multiple crawler tasks are respectively connected to the open API interfaces of different e-commerce platforms; retrieving basic order information through the crawler tasks; constructing a unique key based on the order ID, the creation time, and the update time; and using a Redis Bloom filter based on the unique key to deduplicate the basic order information retrieved from the same e-commerce platform.

[0008] Optionally, in a second implementation of the first aspect of the present invention, the step of using a preset CRF model to identify the waybill reverse label in the order basic information and extracting the reverse waybill number from the corresponding order based on the waybill reverse label includes: collecting operation log text of historical orders, labeling the tags representing reverse logistics in the operation log text to obtain training data; training the CRF model using the training data; extracting word features, prefix and suffix features, part-of-speech features, and context window features of the order basic information as input features of the CRF model, and outputting the model recognition result; when the model recognition result is that a waybill reverse label is identified, extracting the waybill number that conforms to the preset regular expression rule from the order basic information as the reverse waybill number.

[0009] Optionally, in the third implementation of the first aspect of the present invention, the step of calling a third-party logistics platform API to query the sender and receiver information corresponding to the reverse waybill number, and calling a preset Naive Bayes classification model to calculate the first risk probability of the sender and receiver information, includes: calling the corresponding third-party logistics platform API according to the logistics company to which the reverse waybill number belongs; using the third-party logistics platform API to query the sender and receiver information corresponding to the reverse waybill number, wherein the sender and receiver information includes the sender's name, sender's address, sender's phone number, receiver's name, receiver's address, and receiver's phone number; constructing an input feature vector for the Naive Bayes classification model, wherein the input feature vector includes the anomaly degree of the sender and receiver address keywords and the anomaly degree of the sender and receiver name keywords; inputting the input feature vector into the Naive Bayes classification model, and outputting the first risk probability corresponding to the sender and receiver information.

[0010] Optionally, in the fourth implementation of the first aspect of the present invention, the step of matching the user identity of multiple e-commerce platforms based on the mailing and receiving information and obtaining the historical total return rate of the user identity across multiple e-commerce platforms includes: constructing a unique user identifier based on the core hash values ​​of the sender's phone number, sender's name, and sender's address in the mailing and receiving information; using the unique user identifier to query the user's registration ID on different e-commerce platforms in a cross-platform user profile database; aggregating the total number of orders and total number of returned orders generated by all the registered IDs on all associated platforms within a preset time period; and calculating the historical total return rate based on the total number of orders and the total number of returned orders.

[0011] Optionally, in a fifth implementation of the first aspect of the present invention, the step of fusing the first risk probability and the historical total return rate to obtain a second risk probability, and generating abnormal order interception information when the second risk probability is greater than a preset risk threshold, includes: weighted fusing the first risk probability and the historical total return rate to obtain a second risk probability; setting a risk threshold according to the merchant's risk preference; and generating abnormal order interception information when the second risk probability is greater than the risk threshold, wherein the abnormal order interception information includes an order ID, a reverse tracking number, risk probability details, and suggested actions.

[0012] Optionally, in the sixth implementation of the first aspect of the present invention, after fusing the first risk probability and the historical total return rate to obtain a second risk probability, and generating abnormal order interception information when the second risk probability is greater than a preset risk threshold, the method further includes: sending the abnormal order interception information to a management terminal so that the management terminal generates and displays an interception page based on the interception information; collecting the final processing result of the abnormal order interception information; merging the basic order information corresponding to the reverse waybill number, the second risk probability, the abnormal order interception information, and the final processing result into a risk control record; encrypting the risk control record to obtain an encrypted risk control record; and uploading the encrypted risk control record to the blockchain.

[0013] A second aspect of the present invention provides a reverse order end-to-end identification and risk control device, comprising: a deduplication module for periodically pulling order basic information from multiple open API interfaces and deduplicating duplicate data of the order basic information from the same e-commerce platform; an identification and extraction module for using a preset CRF model to identify the waybill reverse label in the order basic information and extracting the reverse waybill number from the corresponding order based on the waybill reverse label; a query and calculation module for calling a third-party logistics platform API to query the sender and receiver information corresponding to the reverse waybill number and calling a preset Naive Bayes classification model to calculate the first risk probability of the sender and receiver information; a matching and acquisition module for matching the user identity of multiple e-commerce platforms based on the sender and receiver information and obtaining the historical total return rate of the user identity in multiple e-commerce platforms; and a fusion and generation module for fusing the first risk probability and the historical total return rate to obtain a second risk probability, wherein when the second risk probability is greater than a preset risk threshold, abnormal order interception information is generated.

[0014] Optionally, in a first implementation of the second aspect of the present invention, the deduplication module includes: a configuration triggering unit, configured to configure a distributed task scheduling system to trigger multiple parallel crawler tasks at a preset time interval, wherein the multiple crawler tasks are respectively connected to the open API interfaces of different e-commerce platforms; a pull unit, configured to pull basic order information through the crawler tasks; and a deduplication construction unit, configured to construct a unique key based on the order ID, the creation time, and the update time, and use a Redis Bloom filter to deduplicate the basic order information pulled from the same e-commerce platform based on the unique key.

[0015] Optionally, in a second implementation of the second aspect of the present invention, the identification and extraction module includes: a collection and labeling unit, used to collect operation log text of historical orders, label the tags representing reverse logistics in the operation log text, and obtain training data; a training unit, used to train a CRF model using the training data; an extraction and output unit, used to extract word features, prefix and suffix features, part-of-speech features, and context window features of the order basic information as input features of the CRF model, and output the model identification result; and an extraction unit, used to extract the waybill number that conforms to the preset regular expression rule as the reverse waybill number from the order basic information when the model identification result is that a waybill reverse tag has been identified.

[0016] Optionally, in a third implementation of the second aspect of the present invention, the query calculation module includes: a calling unit, used to call the corresponding third-party logistics platform API according to the logistics company to which the reverse waybill number belongs; a first query unit, used to use the third-party logistics platform API to query the sender and receiver information corresponding to the reverse waybill number, the sender and receiver information including sender's name, sender's address, sender's phone number, receiver's name, receiver's address, and receiver's phone number; a first construction unit, used to construct the input feature vector of the Naive Bayes classification model, the input feature vector including the anomaly degree of sender and receiver address keywords and the anomaly degree of sender and receiver name keywords; and an input-output unit, used to input the input feature vector into the Naive Bayes classification model and output the first risk probability corresponding to the sender and receiver information.

[0017] Optionally, in a fourth implementation of the second aspect of the present invention, the matching acquisition module includes: a second construction unit, configured to construct a unique user identifier based on the core hash values ​​of the sender's phone number, sender's name, and sender's address in the sender and recipient information; a second query unit, configured to use the unique user identifier to query the user's registration IDs on different e-commerce platforms in a cross-platform user profile database; an aggregation unit, configured to aggregate the total number of orders and total number of returned orders generated by all the registered IDs on all associated platforms within a preset time period; and a calculation unit, configured to calculate the historical total return rate based on the total number of orders and the total number of returned orders.

[0018] Optionally, in a fifth implementation of the second aspect of the present invention, the fusion generation module includes: a fusion unit, configured to perform weighted fusion of the first risk probability and the historical total return rate to obtain a second risk probability; a setting unit, configured to set a risk threshold according to the merchant's risk preference; and a generation unit, configured to generate abnormal order interception information when the second risk probability is greater than the risk threshold, wherein the abnormal order interception information includes an order ID, a reverse tracking number, risk probability details, and suggested actions.

[0019] Optionally, in the sixth implementation of the second aspect of the present invention, it further includes: a sending module, used to send the abnormal order interception information to a management terminal, so that the management terminal generates and displays an interception page based on the interception information; a collection module, used to collect the final processing result of the abnormal order interception information; a merging module, used to merge the basic order information corresponding to the reverse waybill number, the second risk probability, the abnormal order interception information, and the final processing result into a risk control record; an encryption module, used to encrypt the risk control record to obtain an encrypted risk control record; and an uploading module, used to upload the encrypted risk control record to the blockchain.

[0020] A third aspect of the present invention provides a reverse order full-link identification and risk control device, the reverse order full-link identification and risk control device comprising: a memory and at least one processor, the memory storing instructions; at least one processor calling the instructions in the memory to cause the reverse order full-link identification and risk control device to execute each step of the reverse order full-link identification and risk control method described in any of the preceding claims.

[0021] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the steps of the reverse order end-to-end identification and risk control method described in any of the preceding claims.

[0022] In the technical solution of this invention, order basic information of merchants is periodically retrieved from multiple open API interfaces. Duplicate data in the order basic information of the same e-commerce platform is deduplicated. A preset CRF model is used to identify the reverse waybill label in the order basic information. Based on the reverse waybill label, the reverse waybill number is extracted from the corresponding order. This automates the processing of cross-platform reverse order data streams, possessing the ability to integrate multi-source heterogeneous data, improving data processing efficiency, and achieving accurate identification and standardized management of reverse order customer information. Third-party logistics platform APIs are called to query the sending and receiving information corresponding to the reverse waybill number, covering the entire order data processing process and improving data management quality. A preset Naive Bayes classification model is called to calculate the first risk probability of the sending and receiving information, improving the intelligence level of risk identification. The sending and receiving information is matched with the user identity of multiple e-commerce platforms to obtain the historical total return rate of the user identity across multiple e-commerce platforms. The first risk probability and the historical total return rate are fused to obtain a second risk probability. When the second risk probability is greater than a preset risk threshold, abnormal order interception information is generated, improving the efficiency of risk control decisions. Attached Figure Description

[0023] Figure 1 This is a first flowchart of the reverse order full-link identification and risk control method provided in the embodiments of the present invention;

[0024] Figure 2 This is a second flowchart of the reverse order end-to-end identification and risk control method provided in the embodiments of the present invention;

[0025] Figure 3 This is a third flowchart of the reverse order end-to-end identification and risk control method provided in the embodiments of the present invention;

[0026] Figure 4 This is a fourth flowchart of the reverse order end-to-end identification and risk control method provided in this embodiment of the invention;

[0027] Figure 5 A schematic diagram of a reverse order end-to-end identification and risk control device provided in an embodiment of the present invention;

[0028] Figure 6 This is another structural schematic diagram of the reverse order end-to-end identification and risk control device provided in an embodiment of the present invention;

[0029] Figure 7 This is a schematic diagram of the structure of the reverse order end-to-end identification and risk control device provided in an embodiment of the present invention. Detailed Implementation

[0030] This invention provides a method, apparatus, device, and storage medium for full-link identification and risk control of reverse orders. It has the ability to integrate multi-source heterogeneous data, automatically process cross-platform reverse order data streams, improve data processing efficiency, cover the entire process of order data processing, improve data management quality, realize automated and accurate identification and standardized management of reverse order customer information, improve the intelligence level of risk identification, and improve the efficiency of risk control decisions.

[0031] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0032] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the reverse order end-to-end identification and risk control method in this invention includes:

[0033] 101. Periodically retrieve basic order information from merchants from multiple open API interfaces, and deduplicate duplicate data of basic order information from the same e-commerce platform;

[0034] In this embodiment, a pre-configured distributed task scheduling system automatically triggers multiple independent crawler tasks at preset time intervals (e.g., every 5 minutes). These crawler tasks run in parallel, connecting to the open API interfaces of different e-commerce platforms, authenticating using the API keys authorized by the platform, and retrieving the latest basic order information of merchants (including order ID, creation time, update time, product information, buyer information, order status, etc.). Duplicate data of order basic information from the same e-commerce platform is deduplicated.

[0035] 102. Use a pre-set CRF model to identify the waybill reverse label in the basic order information, and extract the reverse waybill number from the corresponding order based on the waybill reverse label;

[0036] In this embodiment, the deduplicated basic order information is input into a pre-trained CRF model. The CRF model learns from the keywords related to reverse logistics, such as "return", "exchange", "rejection", and "refund" marked in the historical order logs, as well as their contextual patterns, to automatically identify whether the current order information contains a reverse tracking label. If a valid reverse tag (such as "return initiated" or "exchange tracking number") is identified, the corresponding reverse tracking number is accurately extracted from the relevant text fields of the order according to the preset regular expression rules.

[0037] 103. Call the API of the third-party logistics platform to query the sending and receiving information corresponding to the reverse waybill number, and call the preset Naive Bayes classification model to calculate the first risk probability of the sending and receiving information;

[0038] In this embodiment, the system determines the logistics company to which the reverse waybill number belongs based on the prefix or encoding rules of the extracted reverse waybill number. Then, it calls the open platform API of the corresponding logistics company, inputs the reverse waybill number, and queries to obtain the detailed sending and receiving information of the reverse waybill, including the sender's name, sender's address, sender's phone number, recipient's name, recipient's address, and recipient's phone number. Next, based on a preset risk keyword library, the system calculates the abnormality of the sender's address keywords (such as the frequency of words containing "unknown" or "test address") and the abnormality of the sender's name keywords (such as the frequency of words containing generic or vague names such as "test" or "Zhang Moumou"), constructs a feature vector, and inputs this feature vector into a pre-trained Naive Bayes classification model. The model outputs the first risk probability based on the probability distribution learned from historical high-risk sending and receiving samples, quantifying the risk level of the reverse order at the sending and receiving information level.

[0039] 104. Match user identities across multiple e-commerce platforms based on shipment and receipt information, and obtain the historical total return rate of each user across multiple e-commerce platforms;

[0040] In this embodiment, the system utilizes core elements in the sender and recipient information (such as the sender's phone number, sender's name, and core parts of the sender's address) to generate a unique, anonymous user identifier using a specific hash algorithm (such as SHA-256). Then, in the cross-platform user profile database maintained by the system, this user identifier is used to perform a correlation query to find all the user's registration IDs (UIDs) on different e-commerce platforms. The system aggregates the total number of orders and total number of returned orders generated by these associated UIDs on all associated e-commerce platforms within a preset time period (such as the past 12 months). Finally, the system calculates the user's historical total return rate by dividing the total number of returned orders by the total number of orders.

[0041] 105. The first risk probability and the historical total return rate are combined to obtain the second risk probability. When the second risk probability is greater than the preset risk threshold, abnormal order interception information is generated.

[0042] In this embodiment, the first risk probability and the historical total return rate are weighted and fused to obtain the second risk probability. The weight ratio can be adjusted according to the merchant's risk preference. When the second risk probability is greater than the preset risk threshold, the reverse order is determined to be a high-risk abnormal order, and abnormal order interception information is generated.

[0043] In this embodiment of the invention, a distributed task scheduling system is used to periodically trigger parallel crawlers across multiple platform APIs, enabling automated collection of order data from e-commerce platforms and solving the problem of integrating multi-source heterogeneous data. A pre-trained CRF model is used to perform semantic-level reverse tag recognition on order logs, overcoming the limitations of traditional regular expression matching and improving the accuracy of waybill extraction. After obtaining the sending and receiving information by calling the logistics API, a Naive Bayes model is used to calculate the keyword anomaly degree of address / name in real time, generating the first risk probability and realizing intelligent risk screening. A unique user identifier is constructed based on the hash value of the sending information, which is associated with historical return data across platforms to accurately calculate the user's historical total return rate. The real-time risk probability and historical return rate are integrated to generate dynamic risk control decisions. When the comprehensive risk value exceeds a preset threshold, abnormal orders are automatically intercepted, reducing the missed detection rate of high-risk orders.

[0044] Please see Figure 2 The second embodiment of the reverse order end-to-end identification and risk control method in this invention includes:

[0045] 201. Configure a distributed task scheduling system to trigger multiple parallel crawler tasks at preset time intervals. The multiple crawler tasks are connected to the open API interfaces of different e-commerce platforms.

[0046] In this embodiment, a scheduled task plan is configured in the distributed task scheduling system. The plan sets a fixed time interval (e.g., 5 minutes) to trigger multiple independent crawler task executors. Each crawler task is pre-configured with the API endpoint URL, authentication information and request parameter template of its target e-commerce platform. These tasks are executed in parallel without blocking each other, ensuring efficient retrieval of data from multiple platforms.

[0047] 202. Retrieve basic order information using a web crawler task;

[0048] In this embodiment, each triggered crawler task executor initiates an HTTP(S) request to the e-commerce platform API it is responsible for, according to its configuration. The request carries the necessary authentication token and query parameters. The crawler task parses the response returned by the API and extracts a structured list of basic order information.

[0049] 203. Construct a unique key based on the order ID, creation time, and update time. Use a Redis Bloom filter to deduplicate the basic order information retrieved from the same e-commerce platform based on the unique key.

[0050] In this embodiment, for a batch of basic order information retrieved from the same e-commerce platform API, the system generates a unique key for each order data. This unique key is generated by concatenating or hashing the order ID, order creation time, and last update time. The system uses the Bloom filter data structure of the Redis database for efficient deduplication. The system first checks whether the unique key of the order already exists in the Bloom filter. If it exists (it is duplicate data), the order record is discarded. If it does not exist, the unique key is added to the Bloom filter, and the order data is retained for subsequent processing.

[0051] 204. Collect historical order operation log texts, label the tags representing reverse logistics in the operation log texts, and obtain training data;

[0052] In this embodiment, a large amount of order operation log text is extracted from the historical order database. Data labelers or semi-automatic rules are used to identify and label fragments or keywords that clearly indicate reverse logistics behavior in these texts and assign them corresponding "waybill reverse labels". The labeled text and its labels constitute the training dataset.

[0053] 205. Use the training data to train and obtain the CRF model;

[0054] In this embodiment, a prepared labeled dataset is used, and the Conditional Random Field (CRF) algorithm is employed for model training. The training process includes extracting features (such as word features, part-of-speech features, context window features, etc.) from the labeled text, learning the conditional probability relationship between features and inverse labels, and finally obtaining a CRF model that can predict whether the input order text contains inverse labels and the label positions based on the input order text information.

[0055] 206. Extract word features, prefix and suffix features, part-of-speech features, and context window features from the basic order information as input features for the CRF model, and output the model recognition results;

[0056] In this embodiment, the basic order information is preprocessed, and then a feature vector is constructed for the model input. This vector mainly includes: the current word, the prefix of the word, the suffix of the word, the part of speech of the current word, and the context window features composed of several words before and after it. These features are input into the trained CRF model, and the model outputs the label prediction results of each position word. Finally, the results are summarized to determine whether the entire text contains a valid "waybill reverse label".

[0057] 207. When the model identifies a reverse waybill label, extract the waybill number that matches the preset regular expression rule from the basic order information as the reverse waybill number.

[0058] In this embodiment, if the CRF model identifies a valid reverse waybill label in the text information of an order, the system will scan all relevant text fields of the order information. The system has a predefined regular expression rule library for common logistics company waybill numbers. Using these rules, the scanned text is matched, and the first string that matches successfully and conforms to the corresponding logistics company's rules is extracted as the "reverse waybill number" of the reverse order.

[0059] In this embodiment of the invention, a distributed scheduling system is configured to trigger parallel crawler tasks at preset intervals, supporting synchronous retrieval of APIs from multiple e-commerce platforms to improve data collection efficiency. A unique key is constructed, and orders from the same platform are deduplicated using a Redis Bloom filter to reduce redundant data processing. A Conditional Random Field (CRF) model is used to extract part-of-speech and contextual features of the order text and identify reverse tags. When a reverse tag is identified, the tracking number is accurately extracted using a regular expression rule base to reduce the false extraction rate.

[0060] Please see Figure 3 The third embodiment of the reverse order end-to-end identification and risk control method in this invention includes:

[0061] 301. Call the corresponding third-party logistics platform API based on the logistics company to which the reverse waybill number belongs;

[0062] In this embodiment, the system maintains a mapping table between logistics company codes and corresponding API service providers. By analyzing the first few characters of the reverse waybill number or specific coding rules, the logistics company to which the waybill belongs is determined. Based on the mapping table, the corresponding third-party logistics query API interface of the logistics company is selected and called.

[0063] 302. Use the API of a third-party logistics platform to query the sender and receiver information corresponding to the reverse waybill number. The sender and receiver information includes the sender's name, sender's address, sender's phone number, receiver's name, receiver's address, and receiver's phone number.

[0064] In this embodiment, a request is sent to the selected third-party logistics platform API, passing in the reverse tracking number to be queried as a parameter. The JSON / XML response data returned by the API is received and parsed. Key logistics detail fields are extracted from the parsed data, especially sender information (sender's name, complete sender's address, sender's contact number) and recipient information (recipient's name (usually merchant's name or warehouse name), complete recipient's address (merchant's warehouse address), recipient's contact number (merchant's customer service or warehouse phone number)). These are combined to form a structured sender and recipient information object.

[0065] 303. Construct the input feature vector of the Naive Bayes classification model. The input feature vector includes the anomaly degree of the sender and recipient address keywords and the anomaly degree of the sender and recipient name keywords.

[0066] In this embodiment, the system predefines and maintains two risk keyword lists: an address risk keyword list (such as "unknown address", "test address", "XX Province XX City XX District XX Street XX No." (too vague or common), "virtual warehouse", etc.) and a name risk keyword list (such as "test", "Mr / Ms", "Zhang Moumou", "Li Si", "administrator", "customer service", etc.). The system constructs the input feature vector of the Naive Bayes classification model, which includes the anomaly degree of the sender and recipient address keywords and the anomaly degree of the sender and recipient name keywords.

[0067] 304. Input the input feature vector into the Naive Bayes classification model and output the first risk probability corresponding to the sender and receiver information;

[0068] In this embodiment, the input feature vector is input into the Naive Bayes classification model. The Naive Bayes classification model is trained based on historical labeled data (high-risk mail samples are labeled as 1, and low-risk samples are labeled as 0). The Naive Bayes classification model calculates the probability that the mail information belongs to the "high-risk" category based on the input feature value and outputs a value between 0 and 1, which is the first risk probability.

[0069] 305. Construct a unique user identifier based on the core hash values ​​of the sender's phone number, sender's name, and sender's address in the mail sending and receiving information;

[0070] In this embodiment, the three core fields that can most stably identify users in the mail sending and receiving information are selected, namely the core parts of the sender's phone number, sender's name, and sender's address. These fields are cleaned, and then the cleaned strings are concatenated and hashed using a strong hash algorithm (such as SHA-256) to generate a fixed-length, unique, and anonymous string as the user's unique identifier.

[0071] 306. Utilize unique user identifiers in a cross-platform user profile database to query the user's registration IDs on different e-commerce platforms;

[0072] In this embodiment, the system maintains a cross-platform user profile database. This database stores user information mapping relationships linked through historical orders, user authorizations, etc. A query is performed in this database using the registration ID. If a matching record is found, the system retrieves the list of registered user IDs (UIDs) of the user on all associated e-commerce platforms stored in that record. If no match is found, it means that the user is appearing for the first time in this system or under this shipment information combination, and cross-platform information cannot be linked at this time (in this case, it can be treated as a new user).

[0073] 307. Aggregate the total number of orders and total number of returned orders generated by all registered IDs across all associated platforms within the past preset time period;

[0074] In this embodiment, the system retrieves the user's registration ID list across all e-commerce platforms, accesses the order center database or the order query interface provided by each platform, and counts the total number of orders and total number of return orders generated by these UIDs on their respective platforms within a specified past time period (e.g., the last 365 days).

[0075] 308. Calculate the historical total return rate based on the total number of orders and the total number of returned orders;

[0076] In this embodiment, the user's historical total return rate is calculated by dividing the total number of returned orders by the total number of orders.

[0077] In this embodiment of the invention, the query response time is optimized by matching the logistics company's API based on the waybill number prefix, a risk keyword library is constructed, a feature vector is generated and input into the Naive Bayes model, and the first risk probability in the range of 0-1 is output. By generating a unique fingerprint for the user, the problem of cross-platform identity association is solved. The total number of orders and the number of returned orders of all associated IDs of the user within a certain period of time are aggregated, the historical return rate is calculated, and the second-level response is supported.

[0078] Please see Figure 4 The fourth embodiment of the reverse order end-to-end identification and risk control method in this invention includes:

[0079] 401. The second risk probability is obtained by weighting and fusing the first risk probability and the historical total return rate;

[0080] In this embodiment, the first risk probability and the historical total return rate are weighted and fused to obtain the second risk probability. The weighting coefficient is set by the merchant according to their own risk strategy.

[0081] 402. Set risk thresholds based on merchants' risk preferences;

[0082] In this embodiment, the merchant sets a global risk threshold in the risk control system backend. This risk threshold is a value between 0 and 1, representing the risk boundary that the merchant is willing to bear. Orders below this threshold are considered acceptable risk or normal orders, while orders above this threshold are considered high-risk abnormal orders that require intervention.

[0083] 403. When the probability of the second risk is greater than the risk threshold, abnormal order interception information is generated. The abnormal order interception information includes the order ID, reverse waybill number, risk probability details and suggested actions.

[0084] In this embodiment, the system compares the calculated second risk probability with the preset risk threshold in real time. When the second risk probability is greater than the risk threshold, it determines that the reverse order has a high risk and generates abnormal order interception information. The abnormal order interception information includes the order ID, reverse waybill number, risk probability details, and suggested actions.

[0085] 404. Send the abnormal order interception information to the management terminal so that the management terminal can generate and display the interception page based on the interception information;

[0086] In this embodiment, the generated abnormal order interception information is pushed in real time to the management terminal (such as a web backend system or customer service workbench APP) used by the merchant or risk control personnel through message queues, internal API calls or WebSockets. After receiving the interception information, the management terminal parses its content and dynamically generates and displays a clear and intuitive interception details page in the corresponding module.

[0087] 405. Collect the final processing results of abnormal order interception information;

[0088] In this embodiment, after the management terminal displays the interception page, manual intervention (or automatic execution in combination with preset rules) is required for final processing. The system records the final processing actions taken for the interception information and their results. For example, the manual review result is "confirmed fraud, refund refused", "information is incorrect, marked as false alarm", "verification is normal, refund allowed", and the operations performed are "add to blacklist", "notify warehouse to intercept", and "release refund".

[0089] 406. Merge the basic order information, second risk probability, abnormal order interception information, and final processing result corresponding to the reverse waybill number into a risk control record;

[0090] In this embodiment, the system associates and integrates all key information related to the high-risk reverse order and packages it into a complete risk control record. This record typically includes the original order basic information that triggered the risk control, the calculated second risk probability, the abnormal order interception information generated by the system, the final processing result given by the human or the system, and the relevant timestamps (retrieval time, calculation time, alarm time, processing time).

[0091] 407. Encrypt the risk control records to obtain encrypted risk control records;

[0092] In this embodiment, in order to protect sensitive information (such as user phone number, address, and order details) and ensure the integrity of records, the system uses encryption algorithms (such as AES-256 symmetric encryption or asymmetric encryption combined with RSA) to encrypt the integrated complete risk control records. The encrypted data is called the encrypted risk control record, and only the system or authorized party with the decryption key can read its original content.

[0093] 408. Upload encrypted risk control records to the blockchain;

[0094] In this embodiment, the encrypted risk control record is used as the data payload to construct a blockchain transaction. The transaction is signed using the system's private key to ensure that the source is trustworthy and tamper-proof. Then, the signed transaction is broadcast to the selected blockchain network. After the nodes in the blockchain network verify the validity of the transaction, they package it into a new block and add it to the distributed ledger. Once successfully uploaded to the chain, the encrypted risk control record obtains the immutable and traceable characteristics provided by the blockchain, providing reliable evidence for subsequent auditing and dispute resolution.

[0095] In this embodiment of the invention, a weighted fusion method is adopted, which supports dynamic adjustment of weights according to the merchant's risk strategy. When the probability of the second risk is greater than the risk threshold, structured interception information is generated, including the order ID, risk probability decomposition value and handling suggestions. The interception information is pushed to the management terminal in real time through the Kafka message queue, and a handling work order page is automatically generated to improve the efficiency of manual decision-making. The risk control records are encrypted with AES-256 and then uploaded to the blockchain to ensure the reliability of audit traceability by utilizing the immutability of the blockchain.

[0096] The above describes the reverse order end-to-end identification and risk control method in the embodiments of the present invention. The following describes the reverse order end-to-end identification and risk control device in the embodiments of the present invention. Please refer to [link / reference]. Figure 5 One embodiment of the reverse order end-to-end identification and risk control device in this invention includes:

[0097] The deduplication module 501 is used to periodically retrieve basic order information from merchants from multiple open API interfaces and deduplicate duplicate data of basic order information from the same e-commerce platform.

[0098] The identification and extraction module 502 is used to identify the waybill reverse label in the basic order information using a preset CRF model, and extract the reverse waybill number from the corresponding order based on the waybill reverse label;

[0099] The query calculation module 503 is used to call the API of a third-party logistics platform to query the sender and receiver information corresponding to the reverse waybill number, and to call the preset Naive Bayes classification model to calculate the first risk probability of the sender and receiver information.

[0100] The matching module 504 is used to match the user identity of multiple e-commerce platforms based on the parcel sending and receiving information, and to obtain the historical total return rate of the user identity in multiple e-commerce platforms.

[0101] The fusion generation module 505 is used to fuse the first risk probability and the historical total return rate to obtain the second risk probability. When the second risk probability is greater than the preset risk threshold, abnormal order interception information is generated.

[0102] In this embodiment, order basic information from merchants is periodically retrieved from multiple open API interfaces. Duplicate data in the order basic information of the same e-commerce platform is deduplicated. A preset CRF model is used to identify reverse waybill tags in the order basic information. Based on the reverse waybill tags, reverse waybill numbers are extracted from the corresponding orders. This automates the processing of cross-platform reverse order data streams, providing multi-source heterogeneous data integration capabilities, improving data processing efficiency, and achieving accurate identification and standardized management of reverse order customer information. Third-party logistics platform APIs are called to query the sending and receiving information corresponding to the reverse waybill numbers, covering the entire order data processing process and improving data management quality. A preset Naive Bayes classification model is used to calculate the first risk probability of the sending and receiving information, improving the intelligence level of risk identification. The sending and receiving information is matched with user identities from multiple e-commerce platforms to obtain the historical total return rate of the user identity across multiple e-commerce platforms. The first risk probability and the historical total return rate are fused to obtain a second risk probability. When the second risk probability is greater than a preset risk threshold, abnormal order interception information is generated, improving the efficiency of risk control decisions.

[0103] Please see Figure 6 Another embodiment of the reverse order end-to-end identification and risk control device in this invention includes:

[0104] The deduplication module 501 is used to periodically retrieve basic order information from merchants from multiple open API interfaces and deduplicate duplicate data of basic order information from the same e-commerce platform.

[0105] The identification and extraction module 502 is used to identify the waybill reverse label in the basic order information using a preset CRF model, and extract the reverse waybill number from the corresponding order based on the waybill reverse label;

[0106] The query calculation module 503 is used to call the API of a third-party logistics platform to query the sender and receiver information corresponding to the reverse waybill number, and to call the preset Naive Bayes classification model to calculate the first risk probability of the sender and receiver information.

[0107] The matching module 504 is used to match the user identity of multiple e-commerce platforms based on the parcel delivery and receipt information, and obtain the historical total return rate of the user identity in multiple e-commerce platforms.

[0108] The fusion generation module 505 is used to fuse the first risk probability and the historical total return rate to obtain the second risk probability. When the second risk probability is greater than the preset risk threshold, abnormal order interception information is generated.

[0109] In this embodiment, the deduplication module 501 includes: a configuration triggering unit 5011, used to configure a distributed task scheduling system to trigger multiple parallel crawler tasks at a preset time interval, wherein the multiple crawler tasks are respectively connected to the open API interfaces of different e-commerce platforms; a pull unit 5012, used to pull basic order information through crawler tasks; and a deduplication construction unit 5013, used to construct a unique key based on the order ID, creation time, and update time, and use a Redis Bloom filter to deduplicate the basic order information pulled from the same e-commerce platform based on the unique key.

[0110] In this embodiment, the identification and extraction module 502 includes: a collection and annotation unit 5021, used to collect operation log text of historical orders, annotate the tags representing reverse logistics in the operation log text, and obtain training data; a training unit 5022, used to train a CRF model using the training data; an extraction and output unit 5023, used to extract word features, prefix and suffix features, part-of-speech features, and context window features of the order basic information as input features of the CRF model, and output the model recognition result; and an extraction unit 5024, used to extract the waybill number that conforms to the preset regular expression rules as the reverse waybill number from the order basic information when the model recognition result is that the waybill reverse tag is identified.

[0111] In this embodiment, the query calculation module 503 includes: a calling unit 5031, used to call the corresponding third-party logistics platform API according to the logistics company to which the reverse waybill number belongs; a first query unit 5032, used to use the third-party logistics platform API to query the sender and recipient information corresponding to the reverse waybill number, the sender and recipient information including sender's name, sender's address, sender's phone number, recipient's name, recipient's address, and recipient's phone number; a first construction unit 5033, used to construct the input feature vector of the Naive Bayes classification model, the input feature vector including the anomaly degree of sender and recipient address keywords and the anomaly degree of sender and recipient name keywords; and an input-output unit 5034, used to input the input feature vector into the Naive Bayes classification model and output the first risk probability corresponding to the sender and recipient information.

[0112] In this embodiment, the matching acquisition module 504 includes: a second construction unit 5041, used to construct a unique user identifier based on the core hash values ​​of the sender's phone number, sender's name, and sender's address in the mailing and receiving information; a second query unit 5042, used to query the user's registration IDs on different e-commerce platforms using the unique user identifier in the cross-platform user profile database; an aggregation unit 5043, used to aggregate the total number of orders and total number of returned orders generated by all registered IDs on all associated platforms within a preset time period; and a calculation unit 5044, used to calculate the historical total return rate based on the total number of orders and total number of returned orders.

[0113] In this embodiment, the fusion generation module 505 includes: a fusion unit 5051, used to perform weighted fusion of the first risk probability and the historical total return rate to obtain a second risk probability; a setting unit 5052, used to set a risk threshold according to the merchant's risk preference; and a generation unit 5053, used to generate abnormal order interception information when the second risk probability is greater than the risk threshold. The abnormal order interception information includes the order ID, reverse waybill number, risk probability details, and suggested actions.

[0114] In this embodiment, the system further includes: a sending module 506, used to send abnormal order interception information to a management terminal, so that the management terminal can generate and display an interception page based on the interception information; a collection module 507, used to collect the final processing results of the abnormal order interception information; a merging module 508, used to merge the basic order information, the second risk probability, the abnormal order interception information, and the final processing results corresponding to the reverse waybill number into a risk control record; an encryption module 509, used to encrypt the risk control record to obtain an encrypted risk control record; and an uploading module 510, used to upload the encrypted risk control record to the blockchain.

[0115] above Figure 5 and Figure 6 The reverse order full-link identification and risk control device in this embodiment of the invention is described in detail from the perspective of modular functional entities. The reverse order full-link identification and risk control device in this embodiment of the invention is described in detail from the perspective of hardware processing.

[0116] Figure 7 This is a schematic diagram of the structure of a reverse order end-to-end identification and risk control device 600 provided in an embodiment of the present invention. The reverse order end-to-end identification and risk control device 600 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 610 (e.g., one or more processors) and a memory 620, and one or more storage media 630 (e.g., one or more mass storage devices) for storing application programs 633 or data 632. The memory 620 and storage media 630 can be temporary or persistent storage. The program stored in the storage media 630 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the reverse order end-to-end identification and risk control device 600. Furthermore, the processor 610 may be configured to communicate with the storage media 630 and execute the series of instruction operations in the storage media 630 on the reverse order end-to-end identification and risk control device 600 to implement the steps of the reverse order end-to-end identification and risk control methods provided in the above-described method embodiments.

[0117] The reverse order end-to-end identification and risk control device 600 may also include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input / output interfaces 660, and / or one or more operating systems 631, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 7 The illustrated structure of the reverse order end-to-end identification and risk control equipment does not constitute a limitation on reverse order end-to-end identification and risk control equipment. It may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0118] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the reverse order full-link identification and risk control method.

[0119] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0120] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0121] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for end-to-end identification and risk control of reverse orders, characterized in that, include: The system periodically retrieves basic order information from merchants through multiple open API interfaces and performs deduplication on duplicate data in the basic order information from the same e-commerce platform. A preset CRF model is used to identify the waybill reverse label in the basic order information, and the reverse waybill number is extracted from the corresponding order based on the waybill reverse label; Call the API of a third-party logistics platform to query the parcel information corresponding to the reverse waybill number, and call the preset Naive Bayes classification model to calculate the first risk probability of the parcel information; Based on the parcel delivery and receipt information, match the user identities of multiple e-commerce platforms and obtain the historical total return rate of the user identity across multiple e-commerce platforms. The first risk probability and the historical total return rate are fused to obtain the second risk probability. When the second risk probability is greater than the preset risk threshold, abnormal order interception information is generated.

2. The method for full-chain identification and risk control of reverse orders according to claim 1, characterized in that, The process of periodically retrieving basic order information from merchants through multiple open API interfaces and deduplicating duplicate data from the same e-commerce platform's basic order information includes: Configure a distributed task scheduling system to trigger multiple parallel crawler tasks at preset time intervals, and the multiple crawler tasks are respectively connected to the open API interfaces of different e-commerce platforms; The web crawler task retrieves basic order information. A unique key is constructed based on the order ID, the creation time, and the update time. Based on the unique key, a Redis Bloom filter is used to deduplicate the basic order information retrieved from the same e-commerce platform.

3. The method for full-chain identification and risk control of reverse orders according to claim 1, characterized in that, The step of using a preset CRF model to identify the waybill reverse label in the order basic information, and extracting the reverse waybill number from the corresponding order based on the waybill reverse label, includes: Collect operation log texts of historical orders, and label the tags representing reverse logistics in the operation log texts to obtain training data; The CRF model is trained using the training data; Extract word features, prefix and suffix features, part-of-speech features, and context window features from the basic order information as input features for the CRF model, and output the model recognition results; When the model identifies a reverse waybill label, the waybill number that conforms to a preset regular expression rule is extracted from the basic order information as the reverse waybill number.

4. The method for full-chain identification and risk control of reverse orders according to claim 1, characterized in that, The process of querying the parcel information corresponding to the reverse waybill number by calling the API of a third-party logistics platform, and calculating the first risk probability of the parcel information by calling a preset Naive Bayes classification model, includes: The API of the corresponding third-party logistics platform is invoked based on the logistics company to which the reverse waybill number belongs; The sender and receiver information corresponding to the reverse waybill number can be queried using the API of the third-party logistics platform. The sender and receiver information includes the sender's name, sender's address, sender's phone number, receiver's name, receiver's address, and receiver's phone number. The input feature vector for constructing the Naive Bayes classification model includes the anomaly degree of sender and recipient address keywords and the anomaly degree of sender and recipient name keywords. The input feature vector is input into the Naive Bayes classification model, which outputs the first risk probability corresponding to the mailing and receiving information.

5. The method for full-chain identification and risk control of reverse orders according to claim 1, characterized in that, The step of matching user identities across multiple e-commerce platforms based on the parcel delivery and receipt information, and obtaining the historical total return rate of the user identity across multiple e-commerce platforms, includes: A unique user identifier is constructed based on the core hash values ​​of the sender's phone number, sender's name, and sender's address in the mail sending and receiving information; The user's unique identifier is used to correlate and query the user's registration ID on different e-commerce platforms in the cross-platform user profile database; Aggregate the total number of orders and total number of returned orders generated by all the aforementioned registered IDs across all associated platforms within the past preset time period; The historical total return rate is calculated based on the total number of orders and the total number of returned orders.

6. The method for full-chain identification and risk control of reverse orders according to claim 1, characterized in that, The first risk probability and the historical total return rate are fused to obtain a second risk probability. When the second risk probability is greater than a preset risk threshold, abnormal order interception information is generated, including: The first risk probability and the historical total return rate are weighted and fused to obtain the second risk probability; Set risk thresholds based on merchants' risk preferences; When the second risk probability is greater than the risk threshold, abnormal order interception information is generated. The abnormal order interception information includes the order ID, reverse waybill number, risk probability details, and suggested actions.

7. The method for full-chain identification and risk control of reverse orders according to claim 1, characterized in that, After fusing the first risk probability and the historical total return rate to obtain a second risk probability, and generating abnormal order interception information when the second risk probability is greater than a preset risk threshold, the method further includes: The abnormal order interception information is sent to the management terminal, so that the management terminal generates and displays an interception page based on the interception information; The final processing result of collecting the abnormal order interception information; The order basic information, the second risk probability, the abnormal order interception information, and the final processing result corresponding to the reverse waybill number are merged into a risk control record; The risk control records are encrypted to obtain encrypted risk control records; The encrypted risk control records are uploaded to the blockchain.

8. A reverse order end-to-end identification and risk control device, characterized in that, include: The deduplication module is used to periodically retrieve basic order information from merchants from multiple open API interfaces and deduplicate duplicate data of the basic order information from the same e-commerce platform. The identification and extraction module is used to identify the waybill reverse label in the basic order information using a preset CRF model, and extract the reverse waybill number from the corresponding order based on the waybill reverse label; The query calculation module is used to call the API of a third-party logistics platform to query the parcel information corresponding to the reverse waybill number, and call the preset Naive Bayes classification model to calculate the first risk probability of the parcel information. The matching and acquisition module is used to match the user identity of multiple e-commerce platforms based on the parcel sending and receiving information, and to obtain the historical total return rate of the user identity in the multiple e-commerce platforms. The fusion generation module is used to fuse the first risk probability and the historical total return rate to obtain a second risk probability. When the second risk probability is greater than a preset risk threshold, abnormal order interception information is generated.

9. A reverse order end-to-end identification and risk control device, characterized in that, The reverse order full-link identification and risk control device includes: a memory and at least one processor, wherein the memory stores instructions; At least one of the processors invokes the instructions in the memory to cause the reverse order end-to-end identification and risk control device to perform the steps of the reverse order end-to-end identification and risk control method as described in any one of claims 1-7.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instruction is executed by the processor, it implements each step of the reverse order full-link identification and risk control method as described in any one of claims 1-7.

Images