Abnormal identification method for logistics package weight information and electronic equipment

Abstract

The embodiment of the invention discloses a logistics package weight information abnormity identification method and electronic equipment. The method comprises the steps of establishing a logistics package weight database of a commodity minimum stock unit SKU dimension according to multiple pieces of forward logistics data; using the data in the database to train a commodity weight prediction model; receiving weight information of a returned goods package uploaded by a logistics service party in the process of providing a returned goods collection service, and determining associated returned goods order information and returned goods SKU information associated with a returned goods order; predicting logistics package reasonable weight interval information of the returned commodity SKU through the commodity weight prediction model; and judging whether the weight information of the returned goods package returned by the logistics service party is within the reasonable weight interval range or not, and performing abnormity judgment. According to the embodiment of the invention, the credibility of the returned goods weight data is improved, and the platform risk is reduced.

Description

Technical Field

[0001] This application relates to the field of information processing technology, and in particular to a method and electronic device for identifying anomalies in logistics parcel weight information. Background Technology

[0002] With the rapid development of product information service systems (also known as e-commerce platforms), the return process has become a crucial factor affecting user experience. Some platforms are improving overall user satisfaction and repurchase rates by increasing return shipping subsidies for core users. For example, if a user purchases a product on the platform and is dissatisfied upon receipt, they need to return it. For ordinary users, they need to pay the return shipping costs themselves, or, if they have purchased shipping insurance, the insurance company can cover the costs. However, for users who are paid members of the platform, the platform can subsidize the return shipping costs. For instance, if the shipping cost is within a certain threshold, the platform can pay the entire cost to the logistics service provider, and so on.

[0003] In the process of platform-subsidized return shipping, the platform first needs to determine the specific shipping costs incurred during the return process. These costs are typically calculated based on the logistics service provider's shipping calculation rules, as well as variables such as the mode of transport (air or land), distance, and weight or volume of the package after the logistics service provider receives the returned package. However, since returns are a weakly controlled last-mile pickup service, the only potential supervisory body is the consumer. But, given substantial subsidies, users often don't need to pay for shipping themselves, and therefore, they typically don't pay attention to the package's weight or volume, let alone verify it. Consequently, logistics service providers may intentionally overreport the package's weight or volume (reporting a weight or volume greater than the actual package weight or volume to the platform), resulting in losses for the platform. Summary of the Invention

[0004] This application provides a method and electronic device for identifying anomalies in logistics parcel weight information, which helps improve the credibility of returned weight data and reduce platform risks.

[0005] This application provides the following solution:

[0006] A method for identifying anomalies in the weight information of logistics parcels, comprising:

[0007] A logistics package weight database based on the minimum inventory unit (SKU) dimension of goods is established based on multiple positive logistics data. The positive logistics data refers to the logistics data generated during the shipment process by the second user after the first user completes the order. The database includes the correspondence between product SKUs and categories, the quantity of goods contained in the logistics package, and the weight of the logistics package.

[0008] The data in the database is used to train the commodity weight prediction model so that the commodity weight prediction model can predict the reasonable weight range of logistics packages based on the input commodity SKU information.

[0009] The system receives the weight information of the returned package uploaded by the logistics service provider during the return pickup service process, determines the associated return order information, and the SKU information of the returned goods associated with the return order;

[0010] The reasonable weight range of the logistics package for the returned product SKU is predicted using the product weight prediction model.

[0011] Anomaly detection is performed by determining whether the weight information of the returned package returned by the logistics service provider is within the reasonable weight range.

[0012] This also includes:

[0013] Obtain the weight information of the original shipping package in the forward logistics chain corresponding to the returned order, so as to make an anomaly judgment on the weight information of the returned package in combination with the weight information of the original shipping package.

[0014] This also includes:

[0015] When establishing the database, the package characteristic categories are labeled according to the product category information to which the product SKU belongs, so that the product weight prediction model can combine the package characteristic category information to predict the reasonable weight range of logistics packages.

[0016] The commodity weight prediction model includes a dynamic adjustment layer. This dynamic adjustment layer is used to determine the tolerance space for the weight of the logistics package corresponding to the current returned commodity SKU based on the attribute information, seasonal factors, logistics route information, and / or user credit rating factors. This allows for dynamic adjustment based on the reasonable weight range information initially predicted by the basic model of the commodity weight prediction model, using the tolerance space to ultimately output the prediction result of the reasonable weight range information.

[0017] The commodity weight prediction model includes a bulky goods strategy prediction layer. This layer determines whether to activate the bulky goods strategy based on the category information of the returned commodity SKU, the logistics service policy, historical feedback data of the returned commodity SKU, and / or evidence data. This allows the base model of the commodity weight prediction model to predict a reasonable weight range based on the prediction results of whether the bulky goods strategy is activated. Bulky goods are defined as goods whose weight calculated by volume is greater than their weighing weight.

[0018] This also includes:

[0019] If it is determined that the weight information of the returned package returned by the logistics service provider exceeds the reasonable weight range predicted by the product weight prediction model, an early warning message is sent to the logistics service provider, and the logistics service provider is prompted to return evidence data. The evidence data includes: image data containing the actual returned package, weighing tools, and weighing readings, so as to verify whether there is indeed an anomaly by comparing the weighing readings in the evidence data with the weight information uploaded by the logistics service provider.

[0020] This also includes:

[0021] The packaging integrity of the returned packages included in the evidence data is checked to determine whether there are any unpackaged or empty boxes. If so, a warning message is sent to the logistics service provider.

[0022] This also includes:

[0023] The system verifies the consistency between the weight information of returned packages returned by the logistics service provider and the weight information returned by other logistics nodes. If any inconsistency is found, a warning message is sent to the logistics service provider.

[0024] This also includes:

[0025] When it is necessary to issue an early warning, a customized prompt message is generated using an artificial intelligence (AI) model.

[0026] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of any of the preceding methods.

[0027] An electronic device, comprising:

[0028] One or more processors; and

[0029] A memory associated with the one or more processors, the memory being used to store program instructions that, when read and executed by the one or more processors, perform the steps of the method described in any of the preceding descriptions.

[0030] A computer program product includes a computer program / computer executable instructions that, when executed by a processor in an electronic device, implement the steps of any of the preceding methods.

[0031] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0032] This application's embodiments allow for the initial establishment of a logistics package weight database at the SKU level based on multiple positive logistics data points. The database data can then be used to train a product weight prediction model, enabling the model to predict a reasonable weight range for logistics packages based on the input product SKU information. Subsequently, upon receiving the weight information of returned packages uploaded by the logistics service provider during the return pickup process, the associated return order information and the associated returned product SKU information can be identified. Then, the product weight prediction model is used to predict the reasonable weight range for the returned product SKU in the logistics package. Finally, anomaly detection can be performed by determining whether the weight information of the returned package returned by the logistics service provider falls within the reasonable weight range. This approach enables reasonable weight range prediction for packages at the SKU level, allowing for anomaly detection of returned packages returned by the logistics service provider, thereby improving the reliability of return weight data and reducing platform risk.

[0033] Of course, any product implementing this application does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a schematic diagram of the system architecture provided in the embodiments of this application;

[0036] Figure 2 This is a flowchart of the method provided in the embodiments of this application;

[0037] Figure 3 This is a schematic diagram of the electronic device provided in the embodiments of this application. Detailed Implementation

[0038] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0039] In this embodiment, to reduce platform risks during the return shipping subsidy process, a method for anomaly identification of return packages is provided, primarily for identifying whether the weight of the packages returned by the logistics service provider is abnormal. Specifically, to achieve the above objective, this embodiment can construct a commodity weight prediction model and train it using data to enable it to predict the reasonable weight range of the logistics package based on the input commodity SKU (Minimum Stock Unit) information. In other words, in this embodiment, an algorithm model can be trained; by inputting commodity SKU and other information into the algorithm model, it can predict the reasonable weight range of the logistics package for that commodity SKU. Thus, when the logistics service provider returns the weight information of the logistics package, the corresponding return order information can be determined first, and the associated returned commodity SKU information can be identified from it. Then, the information of the returned commodity SKU can be input into the algorithm model to predict the reasonable weight range of the logistics package. Finally, by judging whether the weight of the logistics package returned by the logistics service provider is within the aforementioned reasonable weight range, it can be determined whether the weight information of the package returned by the logistics service provider is abnormal. If any anomalies are found, the platform can reduce its risk by requesting the logistics service provider to re-upload the package weight information and provide supporting evidence.

[0040] It's important to note that while traditional logistics service systems may have some abnormal package detection mechanisms, they employ a "one package, one weight" modeling approach. This means they model and predict at the package level, providing users with package weight predictions at best throughout the entire logistics chain. In contrast, this application's embodiment predicts the reasonable weight range for logistics packages at the SKU level based on the entire dataset—a proactive measure before risks occur. This means predictions can be made without waiting for the package to complete its journey, and even if anomalies exist, they won't be disclosed to consumers or merchants. Furthermore, there's no need to wait for shipping subsidies to be issued before taking action, allowing for more effective and direct risk control.

[0041] In this embodiment, the algorithm model used for predicting reasonable weight ranges requires predicting reasonable weight ranges based on the input product SKU information. Therefore, training data is collected at the product SKU level. Specifically, this embodiment can establish a product SKU-level logistics package weight database by collecting multiple positive logistics data points, and then train the algorithm model based on this database. The so-called positive logistics data refers to the logistics data generated during the shipping process by the seller (referred to as the second user) after the buyer (referred to as the first user in this embodiment) completes the order placement. Since the logistics service provider also weighs the logistics package to confirm the shipping cost during the seller's shipping process, and since the seller needs to pay the shipping cost to the logistics service provider (when the seller provides "free shipping" service), the seller usually has strong oversight of the logistics package weight. This makes the logistics package weight information in the positive logistics data relatively accurate. Furthermore, since the forward logistics data also contains product SKU information, a correspondence can be established between product SKUs and the corresponding weight data of the forward logistics packages. Of course, during the shipping process, sellers may package multiple product SKUs into the same logistics package. In this case, the number of product SKUs contained in the logistics package can be recorded in the database, and a correspondence can be established between each product SKU and the weight of the logistics package. That is, assuming a logistics package contains three product SKUs, three records can be generated in the database, each corresponding to one product SKU. Naturally, the weight of the logistics package corresponding to each record is the same. During subsequent model training, the specific model can process these data. In addition, since the system continuously generates new forward logistics data, the data in the database can also be continuously updated, for example, daily, or even hourly in a more real-time manner, etc.

[0042] Alternatively, the data in the aforementioned database can be labeled with packaging types. Packaging types can include standard goods (i.e., goods for which freight costs are calculated based on weight), heavy goods (such as large items like refrigerators and washing machines), fresh produce, and bulky goods (whose volumetric weight is greater than their actual weight, commonly seen in lightweight but large-volume items). Since packaging type also affects freight cost calculation, adding this labeling information and incorporating it into model training can improve the model's prediction accuracy.

[0043] After identifying potential anomalies in the weight information of returned packages from the logistics service provider, a warning message can be sent to the service provider. Simultaneously, the service provider can be requested to re-transmit the package weight information along with supporting evidence. This evidence can include images of the returned package, the weighing instrument, and the weighing reading. The re-transmitted weight information may be the same as or different from the previously transmitted data. However, because the re-transmitted weight information includes supporting evidence, its accuracy is relatively higher, more accurately reflecting the actual weight of the returned package. This leads to more accurate shipping cost calculations and reduces platform risk.

[0044] From a system architecture perspective, see Figure 1 This application embodiment can provide a logistics package anomaly identification module in the return service module of a commodity information service system. It may also include a data collection module, a database establishment and maintenance module, and a model training module. The data collection module is mainly used to collect forward logistics data and save it to a structured database. The data in the database is used to train and test the commodity weight prediction model, enabling the specific commodity weight prediction model to predict the reasonable weight range of logistics packages at the commodity SKU level. After the model training is completed, the logistics package anomaly identification module can determine the associated return order information and further determine the information of the returned commodity SKU after receiving the weight information of the returned package from the specific logistics service provider. Then, by calling the aforementioned trained model, it can predict the reasonable weight range of the logistics package for the specific returned commodity SKU. In this way, by judging whether the package weight information returned by the logistics service provider is within the aforementioned reasonable weight range, it can be determined whether the weight information of the returned package is abnormal. If an anomaly is found, the logistics service provider can be required to re-return the package with supporting evidence, etc.

[0045] The specific implementation schemes provided in the embodiments of this application will be described in detail below.

[0046] First, this application provides a method for identifying anomalies in logistics package information, see [link to relevant documentation]. Figure 2 The method may specifically include:

[0047] S201: Establish a logistics package weight database based on multiple positive logistics data points, with the minimum inventory unit (SKU) as the dimension; wherein, the positive logistics data refers to the logistics data generated during the shipping process by the second user after the first user completes the order placement operation; the database includes the correspondence between product SKUs and categories, the quantity of goods contained in the logistics package, and the weight of the logistics package.

[0048] In this embodiment, multiple positive logistics data points can be collected first, and a logistics package weight database at the SKU level can be established. The positive logistics data refers to the logistics data generated during the process of a first user placing an order and a second user shipping the goods. This positive logistics data is typically monitored by the second user, meaning they are more concerned about the accuracy of the package weight information provided by the logistics service provider. Therefore, the accuracy of the positive logistics data is relatively high. This embodiment can utilize this data as training data to train a product weight prediction model.

[0049] Specifically, the logistics parcel weight database can include product SKU information, category information, the quantity of goods contained in the logistics parcel, and the corresponding relationship between the weights of the logistics parcels. Additionally, in practice, the database can also include fields such as tracking number and shipping time, which can be added as needed.

[0050] It should be noted that the aforementioned positive logistics data can also come from the logistics service provider; that is, the positive logistics data is obtained based on the data returned by the logistics service provider. In practical applications, there may be situations where some logistics service providers do not return data. Therefore, the data collection module in this embodiment can also connect to the parcel data centers of major logistics service providers to obtain standardized weight data of positive logistics parcels, and perform data compensation processing through links such as logistics detail feedback data.

[0051] In addition, outlier filtering can be performed on the collected forward logistics data. For example, records with a weight of 0 or exceeding the reasonable range for the product category can be removed (such as books weighing >5kg). Furthermore, based on the capabilities of AI (artificial intelligence) data analysis models, the weight data of forward logistics packages can be cleaned according to the product structure to store the original return data of the goods.

[0052] Furthermore, in the preferred implementation, a data model can be used to model the product classification characteristics based on product categories (such as electronics, apparel, and fresh produce) and packaging types (standard packaging / customized packaging). This ensures that the overall parcel industry characteristics in different scenarios (standard products / heavy goods / fresh produce / bulky goods) can be reproduced. In other words, parcel characteristic categories can be labeled based on the product category information of the product SKU, indicating whether each product is a standard product, heavy goods, fresh produce, or bulky goods. By attaching specific parcel characteristic category tags to specific products, the parcel industry characteristics of each product are reflected. This labeling process can be automatically completed by the data model based on data returned by the logistics service provider. For example, if the data returned by the logistics service provider for a certain product mainly contains the product's volume information, it indicates that the product may belong to the bulky goods category. Furthermore, the overall weight distribution of the product reflected in multiple positive logistics data returned by the logistics service provider can be used to determine whether the product belongs to the standard product category, and so on.

[0053] The data in the aforementioned database can be updated periodically or in real time. For example, it can be updated daily, hourly, or whenever a new piece of positive logistics data is generated, and so on.

[0054] S202: The product weight prediction model is trained using the data in the database so that the product weight prediction model can predict the reasonable weight range of the logistics package based on the input product SKU information.

[0055] After establishing the aforementioned package weight database, the data in the database can be used to train a product weight prediction model. This model takes the collected and preprocessed data (product SKUs, categories, etc.) as input and outputs a prediction of a reasonable weight range for each product. Specifically, the reasonable weight range can be expressed by calculating the mean (μ), standard deviation (σ), and confidence interval (μ±2σ) based on the historical weight distribution of the product SKUs.

[0056] In practical implementation, the product weight prediction model, in addition to the basic model that predicts a reasonable weight range based on input data such as returned product SKUs, can also include a dynamic adjustment layer. This dynamic adjustment layer can determine the tolerance space for the weight of the corresponding logistics package for a product SKU based on the product SKU's attribute information (including category, specifications, material, size, etc.), seasonal factors (such as the need for additional insulated packaging for fresh produce in winter), logistics route information (differences in packaging filling in different regions, etc.), and / or user credit rating factors. This dynamic adjustment layer can be implemented using machine learning models such as random forests, dynamically determining the size of the tolerance space for different product SKUs. For example, fresh produce can be allowed ±10% deviation (based on a moisture loss model), electronic products ±3% (based on precise packaging data), and so on. In this way, based on the reasonable weight range information initially predicted by the basic model of the product weight prediction model, the tolerance space can be dynamically adjusted to ultimately output the predicted result of the reasonable weight range.

[0057] The aforementioned dynamic adjustment layer can be superimposed on the base model, thus enabling the training of the dynamic adjustment layer during the training of the base model, or the two parts can be trained separately. Specifically, during prediction, the outputs of the two parts are superimposed and calculated to obtain the final prediction result.

[0058] Furthermore, the product weight prediction model may also include a bulky goods strategy prediction layer. This layer is used to determine whether to activate the bulky goods strategy based on the product SKU's category information, logistics service policies, historical feedback data for the product SKU, and / or supporting evidence data. This allows the base model of the product weight prediction model to combine the prediction results of whether the bulky goods strategy is activated to predict reasonable weight range information. The aforementioned bulky goods strategy prediction layer can be a model trained using historical feedback data, supporting evidence data, etc., to predict whether to activate the bulky goods strategy.

[0059] The training process of the model described above can be updated and trained according to a certain period. For example, if the positive logistics database is updated daily, the update of this database can automatically trigger model retraining to ensure the real-time availability of the database.

[0060] S203: Receive the weight information of the returned package uploaded by the logistics service provider during the process of providing return pickup service, determine the associated return order information, and the SKU information of the returned goods associated with the return order.

[0061] After training the product weight prediction model, anomaly detection can be performed on the weight information of returned packages returned by the logistics service provider. Specifically, after a buyer initiates a return request through the product information service system, the system can also provide services such as door-to-door pickup. If the buyer chooses this service, the system can notify the logistics service provider to pick up the package and generate a pickup code for the buyer. When the logistics service provider's courier picks up the package, the buyer can provide the pickup code to the courier, who can then input the code into the logistics service provider's specific logistics system. When the logistics service provider returns package information to the product information service system, it can also include this pickup code. In this way, the product information service system can link the return order information with the logistics information using the pickup code. That is, after the logistics service provider returns the weight information of a returned package, the product information service system can determine the associated return order based on the pickup code and other information, and then obtain the associated returned product SKU information from the return order.

[0062] S204: Predict the reasonable weight range of the logistics package for the returned product SKU using the product weight prediction model.

[0063] After determining the SKU information of the returned goods, this information can be input into the product weight prediction model. The model will then predict the reasonable weight range for the returned product SKU in the logistics package. Specifically, the product SKU title and other information can be used as input to the model, and the model can output the predicted reasonable weight range for the logistics package.

[0064] As mentioned earlier, if package characteristic category information is also used during data collection and model training, the category information of the current returned product SKU can also be input into the model, so that the model can predict the reasonable weight range of logistics packages in combination with the specific package characteristic category.

[0065] In addition, if the specific model also includes a dynamic adjustment layer, the attribute information of the current returned product SKU, the current season information, logistics route and other information can be input into the dynamic adjustment layer to dynamically determine the tolerance space of the current returned product SKU.

[0066] If the specific model also includes a bulky goods strategy prediction layer, then the category information of the current returned product SKU, the historical feedback data and / or evidence data of the logistics service policy for that returned product SKU, etc., can be input into this bulky goods strategy prediction layer to determine whether to activate the bulky goods strategy. Correspondingly, the basic model of the product weight prediction model can combine the prediction results of whether to activate the bulky goods strategy to predict reasonable weight range information.

[0067] Furthermore, in practical applications, besides predicting reasonable weight ranges using models, the weight information of the original shipping package in the forward logistics chain corresponding to the return order can also be obtained. This allows for anomaly detection of the returned package's weight information in conjunction with the weight information of the original shipping package. In other words, for a specific returned item, it actually represents the reverse link from buyer to seller, but also corresponds to the forward link. That is, the buyer previously placed an order for the item, and the seller has already shipped the item and delivered it to the buyer's delivery address via logistics. Subsequently, due to dissatisfaction with the received item or quality issues, the buyer initiates a return, entering the reverse link. Therefore, when specifically detecting anomalies in the weight of returned packages during the return process, the weight information of the original shipping package in the corresponding forward logistics chain can also be used for detection.

[0068] S205: An anomaly is determined by judging whether the weight information of the returned package returned by the logistics service provider is within the reasonable weight range.

[0069] After obtaining the predicted weight range of the logistics package, anomaly detection can be performed by determining whether the weight information of the returned package returned by the logistics service provider falls within the reasonable weight range. Specifically, if the weight information of the returned package returned by the logistics service provider exceeds the reasonable weight range predicted by the product weight prediction model, a warning message can be issued to the logistics service provider, prompting them to submit supporting evidence. This evidence includes image data containing the actual returned package, weighing equipment, and weighing readings. The weight readings in the supporting evidence are then compared with the weight information uploaded by the logistics service provider to verify whether an anomaly has indeed occurred.

[0070] In addition, the packaging integrity of the returned packages included in the evidence data can be checked to determine if there are any unpackaged or empty boxes. If so, they can be marked as "high risk," and a warning message can be sent to the logistics service provider. Furthermore, the weight information of the returned packages returned by the logistics service provider can be checked for consistency with the weight information returned by other logistics nodes. Specifically, this consistency check can be performed using time series anomaly detection algorithms. If any inconsistencies are found, a warning message can be sent to the logistics service provider.

[0071] In an optional implementation, when a warning message needs to be issued, a customized prompt is generated using an AI model. The AI ​​model refers to a deep learning model containing a massive number of parameters. Due to its large parameter scale, such AI models can store and process large amounts of information, thereby achieving higher performance in various tasks, including multimodal content understanding and generation. In this embodiment, the text content generation capability of the AI ​​model can be used to generate the prompt. For example, the generated prompt could be: "Weight deviation detected, please remeasure or take a clear weighing photo," etc. Furthermore, specific image detection in the evidence data can also be accomplished using an AI model.

[0072] After receiving the warning, the pickup clerk can remeasure and resubmit the results, along with supporting evidence. The newly submitted data will then flow back into the core computing layer for anomaly detection. If the re-measured result is within a reasonable weight range, the new data can be incorporated into the product weight prediction model's training set to dynamically update the baseline value for that product SKU. If the resubmitted data is still abnormal, meta-learning algorithms can be used to analyze the re-measured data against historical anomaly patterns to determine if it's a system misjudgment, and so on.

[0073] In addition, a review mechanism can be provided in practical applications. Specifically, risk classification and control can be achieved through multi-layered AI models. This includes review at the distribution center (intelligent verification of transit weight) and platform review. The distribution center review can include data alignment (weight data collected by dynamic weighing sensors at the distribution center is compared in real time with the value entered by the pickup clerk) and anomaly pattern recognition. Anomaly pattern recognition involves using a commodity weight prediction model to simulate a reasonable weight distribution. If the transit weight deviates from the predicted value by more than a threshold, it is marked as "unacceptable." If accepted, the minimum value can be used for settlement; if unacceptable, a risk scoring model can be triggered to assess whether to proceed to platform review.

[0074] Platform verification can be achieved through multi-task learning models, such as a multimodal model based on VisionTransformer (ViT) and BERT, to simultaneously analyze image and text information. Image analysis mainly includes: detecting the compliance of weighing photos (e.g., whether they include complete packages and the weight display interface), and locating abnormal areas (e.g., liquid leaks, concealed items) using attention mechanisms. Text analysis mainly includes: parsing the courier's notes (e.g., "user refused to sign"), and combining historical behavioral data to determine if there are any fraudulent connections.

[0075] In addition, the system can achieve dynamic evolution through reinforcement learning and continuous learning frameworks, which can include dynamic tolerance range adjustment, malicious pickup operator identification mechanism, and continuous model optimization. Dynamic tolerance range adjustment includes: analyzing historical risk control data based on LSTM (Long Short-Term Memory) networks to set personalized thresholds for high-risk pickup operators (e.g., tightening the threshold by 20%); and dynamically adjusting the tolerance range based on factors such as product category return rates and seasonal fluctuations through online learning mechanisms (e.g., expanding the tolerance for fresh produce to ±12% in summer).

[0076] Malicious pickup operator identification mechanisms can include specific methods such as behavioral pattern clustering and risk scoring cards. Behavioral pattern clustering specifically refers to using algorithms like DBSCAN (Density-Based Spatial Clustering of Applications with Noise) to cluster historical operational data of pickup operators (such as anomaly marker frequency and review pass rate) to identify high-risk groups. Risk scoring cards specifically refer to constructing interpretive models based on SHAP (SHapley Additive exPlanations, used to interpret the output of any machine learning model in a unified way) values ​​to quantify the operational risks of pickup operators, generate a blacklist, and potentially trigger manual audits.

[0077] Regarding continuous model optimization, review results (such as manually determined "fraud" cases) can be fed into the model daily (or at other time intervals) to avoid overfitting. Regular comparisons can also be made between the old and new models in terms of accuracy (e.g., anomaly detection rate increased to 98%), processing efficiency (e.g., a 60% reduction in manual review workload), and other metrics to optimize algorithm parameters.

[0078] The above methods can be used to build a multi-level verification system (certification weight comparison → machine review and verification → distribution inspection → manual final review), which can improve the credibility of returned weight data and process efficiency.

[0079] In summary, through the embodiments of this application, a logistics package weight database at the SKU level can be established first based on multiple positive logistics data. The data in this database can then be used to train a product weight prediction model, enabling the model to predict the reasonable weight range of the logistics package based on the input product SKU information. Subsequently, upon receiving the weight information of returned packages uploaded by the logistics service provider during the return pickup process, the associated return order information and the returned product SKU information associated with the return order can be determined first. Then, the reasonable weight range of the returned product SKU in the logistics package can be predicted using the product weight prediction model. Finally, anomaly detection can be performed by determining whether the weight information of the returned package returned by the logistics service provider is within the reasonable weight range. In this way, reasonable weight range prediction for packages at the SKU level can be achieved, thereby enabling anomaly detection of returned packages returned by the logistics service provider, improving the reliability of return weight data and reducing platform risk.

[0080] It should be noted that the embodiments of this application may involve the use of user data. In practical applications, user-specific personal data may be used in the scheme described herein within the scope permitted by applicable laws and regulations, provided that it complies with the applicable laws and regulations of the country (e.g., with the user's explicit consent, with the user being properly notified, etc.).

[0081] Corresponding to the foregoing method embodiments, this application also provides an anomaly identification device for logistics package weight information, which may include:

[0082] The database construction unit is used to establish a logistics package weight database based on multiple positive logistics data points, with the minimum inventory unit (SKU) as the dimension. The positive logistics data refers to the logistics data generated during the shipping process by the second user after the first user completes an order placement. The database includes the correspondence between product SKUs and categories, the quantity of goods contained in the logistics package, and the weight of the logistics package.

[0083] The model training unit is used to train the commodity weight prediction model using the data in the database, so that the commodity weight prediction model can predict the reasonable weight range of the logistics package based on the input commodity SKU information.

[0084] The information determination unit is used to receive the weight information of the returned package uploaded by the logistics service provider during the process of providing return pickup service, determine the associated return order information, and the SKU information of the returned goods associated with the return order;

[0085] The prediction unit is used to predict the reasonable weight range of the logistics package for the returned product SKU using the product weight prediction model.

[0086] The anomaly detection unit is used to detect anomalies by determining whether the weight information of the returned package returned by the logistics service provider is within the reasonable weight range.

[0087] In a specific implementation, the device may further include:

[0088] The original shipment package weight information acquisition unit is used to acquire the weight information of the original shipment package in the forward logistics link corresponding to the return order, so as to combine the weight information of the original shipment package to make an anomaly judgment on the weight information of the return package.

[0089] Additionally, the device may also include:

[0090] The package characteristic category labeling unit is used to label the package characteristic categories according to the product category information to which the product SKU belongs when establishing the database, so that the product weight prediction model can combine the package characteristic category information to predict the reasonable weight range of logistics packages.

[0091] In a specific implementation, the commodity weight prediction model may also include a dynamic adjustment layer. The dynamic adjustment layer is used to determine the tolerance space of the weight of the logistics package corresponding to the current returned commodity SKU based on the attribute information, seasonal factors, logistics route information and / or user credit rating factors. This allows for dynamic adjustment based on the reasonable weight range information initially predicted by the basic model of the commodity weight prediction model, through the tolerance space, so as to finally output the prediction result of the reasonable weight range information.

[0092] In addition, the commodity weight prediction model may also include a bulky goods strategy prediction layer. The bulky goods strategy prediction layer is used to determine whether to activate the bulky goods strategy based on the category information of the returned commodity SKU, the logistics service policy, the historical feedback data and / or evidence data of the returned commodity SKU, so that the basic model of the commodity weight prediction model can predict reasonable weight range information by combining the prediction results of whether the bulky goods strategy is activated; wherein, the bulky goods are: goods whose weight calculated by volume is greater than the weighing weight.

[0093] Furthermore, the device may also include:

[0094] The early warning unit is used to issue an early warning to the logistics service provider if it determines that the weight information of the returned package returned by the logistics service provider exceeds the reasonable weight range predicted by the commodity weight prediction model. The early warning unit also prompts the logistics service provider to return evidence data, which includes image data of the returned package, weighing tools, and weighing readings. This data is used to verify whether there is indeed an anomaly by comparing the weighing readings in the evidence data with the weight information uploaded by the logistics service provider.

[0095] The integrity detection unit is used to perform packaging integrity detection on the physical images of returned packages included in the evidence data to determine whether there are any cases of unpackaged or empty boxes. If so, it further issues a warning message to the logistics service provider.

[0096] The consistency detection unit is used to verify the consistency between the weight information of returned packages returned by the logistics service provider and the weight information returned by other logistics nodes. If there is any inconsistency, an early warning message will be sent to the logistics service provider.

[0097] The prompt generation unit is used to generate customized prompts using an artificial intelligence (AI) model when it is necessary to issue warning information.

[0098] In addition, embodiments of this application also provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method described in any of the foregoing method embodiments.

[0099] And an electronic device, comprising:

[0100] One or more processors; and

[0101] A memory associated with the one or more processors, the memory being used to store program instructions that, when read and executed by the one or more processors, perform the steps of the method described in any of the foregoing method embodiments.

[0102] A computer program product includes a computer program / computer executable instructions that, when executed by a processor in an electronic device, implement the steps of the method described in the foregoing method embodiments.

[0103] in, Figure 3The architecture of an electronic device is illustrated, which may include a processor 310, a video display adapter 311, a disk drive 312, an input / output interface 313, a network interface 314, and a memory 320. The processor 310, video display adapter 311, disk drive 312, input / output interface 313, network interface 314, and memory 320 can communicate with each other via a communication bus 330.

[0104] The processor 310 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to achieve the technical solution provided in this application.

[0105] The memory 320 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 320 can store the operating system 321 for controlling the operation of the electronic device 300, and the basic input / output system (BIOS) for controlling the low-level operations of the electronic device 300. Additionally, it can store a web browser 323, a data storage management system 324, and an anomaly detection and handling system 325, etc. The aforementioned anomaly detection and handling system 325 can be the application program that specifically implements the aforementioned steps in this embodiment. In summary, when the technical solution provided in this application is implemented through software or firmware, the relevant program code is stored in the memory 320 and is called and executed by the processor 310.

[0106] Input / output interface 313 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touch screens, microphones, various sensors, etc., and output devices may include displays, speakers, vibrators, indicator lights, etc.

[0107] Network interface 314 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0108] Bus 330 includes a pathway for transmitting information between various components of the device, such as processor 310, video display adapter 311, disk drive 312, input / output interface 313, network interface 314, and memory 320.

[0109] It should be noted that although the above-described device only shows the processor 310, video display adapter 311, disk drive 312, input / output interface 313, network interface 314, memory 320, bus 330, etc., in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the solution of this application, and does not necessarily include all the components shown in the figures.

[0110] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0111] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0112] The above provides a detailed description of the method and electronic device for identifying anomalies in logistics parcel weight information provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are merely for the purpose of helping to understand the method and its core ideas. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for identifying anomalies in the weight information of logistics parcels, characterized in that, include: A logistics package weight database based on the minimum inventory unit (SKU) dimension of goods is established based on multiple positive logistics data. The positive logistics data refers to the logistics data generated during the shipment process by the second user after the first user completes the order. The database includes the correspondence between product SKUs and categories, the quantity of goods contained in the logistics package, and the weight of the logistics package. The data in the database is used to train the commodity weight prediction model so that the commodity weight prediction model can predict the reasonable weight range of logistics packages based on the input commodity SKU information. The system receives the weight information of the returned package uploaded by the logistics service provider during the return pickup service process, determines the associated return order information, and the SKU information of the returned goods associated with the return order; The reasonable weight range of the logistics package for the returned product SKU is predicted using the product weight prediction model. Anomaly detection is performed by determining whether the weight information of the returned package returned by the logistics service provider is within the reasonable weight range.

2. The method according to claim 1, characterized in that, Also includes: Obtain the weight information of the original shipping package in the forward logistics chain corresponding to the returned order, so as to make an anomaly judgment on the weight information of the returned package in combination with the weight information of the original shipping package.

3. The method according to claim 1, characterized in that, Also includes: When establishing the database, the package characteristic categories are labeled according to the product category information to which the product SKU belongs, so that the product weight prediction model can combine the package characteristic category information to predict the reasonable weight range of logistics packages.

4. The method according to claim 1, characterized in that, The commodity weight prediction model includes a dynamic adjustment layer. The dynamic adjustment layer is used to determine the tolerance space of the weight of the logistics package corresponding to the current returned commodity SKU based on the attribute information, seasonal factors, logistics route information and / or user credit rating factors. This allows for dynamic adjustment based on the reasonable weight range information initially predicted by the basic model of the commodity weight prediction model, through the tolerance space, so as to finally output the prediction result of the reasonable weight range information.

5. The method according to claim 1, characterized in that, The commodity weight prediction model includes a bulky goods strategy prediction layer. This layer is used to determine whether to activate the bulky goods strategy based on the category information of the returned commodity SKU, the logistics service policy, historical feedback data of the returned commodity SKU, and / or evidence data. This allows the base model of the commodity weight prediction model to predict reasonable weight range information by combining the prediction results of whether the bulky goods strategy is activated. Bulky goods are defined as goods whose weight calculated by volume is greater than their weighing weight.

6. The method according to claim 1, characterized in that, Also includes: If it is determined that the weight information of the returned package returned by the logistics service provider exceeds the reasonable weight range predicted by the product weight prediction model, an early warning message is sent to the logistics service provider, and the logistics service provider is prompted to return evidence data. The evidence data includes: image data containing the actual returned package, weighing tools, and weighing readings, so as to verify whether there is indeed an anomaly by comparing the weighing readings in the evidence data with the weight information uploaded by the logistics service provider.

7. The method according to claim 6, characterized in that, Also includes: The packaging integrity of the returned packages included in the evidence data is checked to determine whether there are any unpackaged or empty boxes. If so, a warning message is sent to the logistics service provider.

8. The method according to claim 1, characterized in that, Also includes: The system verifies the consistency between the weight information of returned packages returned by the logistics service provider and the weight information returned by other logistics nodes. If any inconsistency is found, a warning message is sent to the logistics service provider.

9. The method according to any one of claims 6 to 8, characterized in that, Also includes: When it is necessary to issue an early warning, a customized prompt message is generated using an artificial intelligence (AI) model.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method described in any one of claims 1 to 9.

11. An electronic device, characterized in that, include: One or more processors; as well as A memory associated with the one or more processors, the memory being used to store program instructions that, when read and executed by the one or more processors, perform the steps of the method according to any one of claims 1 to 9.

12. A computer program product comprising a computer program / computer-executable instructions, characterized in that, When the computer program / computer executable instructions are executed by a processor in an electronic device, they implement the steps of the method according to any one of claims 1 to 9.

Images