Warehouse-out checking processing method and device, electronic equipment and medium

Abstract

The invention relates to an ex-warehouse checking processing method and device, electronic equipment and a medium. The method comprises the steps that goods information of goods to be ex-warehouse is acquired; determining the similarity between the to-be-delivered goods and similar goods in the warehouse goods according to the goods information; inputting the feature data of the to-be-delivered goods into a pre-trained abnormal risk identification model, and outputting a risk identification result of the to-be-delivered goods; wherein the feature data comprises the similarity; and according to the risk identification result, generating a differentiated target ex-warehouse checking processing strategy. According to the technical scheme of the invention, differentiated ex-warehouse checking processing strategies can be adopted for the to-be-ex-warehouse goods which are easy to confuse and make mistakes and not easy to confuse and make mistakes, so that the rechecking efficiency and the rechecking accuracy of the conditions which are easy to confuse and make mistakes can be accurately improved, and the rechecking manpower and time cost of some goods which are not easy to make mistakes can be reduced or even avoided.

Description

Technical Field

[0001] The present disclosure relates to the fields of warehousing and computers, and in particular to a method, device, electronic equipment and medium for warehouse outbound verification processing. Background Art

[0002] With the widespread development of e-commerce, the types and volume of orders are increasing. How to accurately ship out the goods corresponding to the orders in a short time is a key link in the warehouse management process and an important link in determining the timeliness of logistics. If the shipping time is short but the goods shipped out have quality or quantity problems, the subsequent re-shipment of goods will cost a lot of logistics costs and reduce customer satisfaction. If the goods can be shipped out correctly but it takes a long time, it is easy to cause users to be dissatisfied with the timeliness of logistics.

[0003] In the process of realizing the concept of the present disclosure, the inventors discovered that there are at least the following technical problems in the related art: due to the large order volume and diverse order types, pickers may make mistakes in the picking process. In order to promptly discover and remedy picking errors before shipment, the related art usually adds a review link for all shipped goods before shipment; however, this review link reviews all goods indiscriminately, which not only increases the manpower and time costs required for the review, but also has the problem of poor review results and some review errors. Summary of the invention

[0004] In order to solve the above technical problems or at least partially solve the above technical problems, the embodiments of the present disclosure provide a method, device, electronic device and medium for warehouse outbound verification processing.

[0005] In a first aspect, an embodiment of the present disclosure provides a method for outbound verification processing. The method includes: obtaining product information of the goods to be outbound; determining the similarity between the goods to be outbound and similar goods in the warehouse based on the product information; inputting the feature data of the goods to be outbound into a pre-trained abnormal risk identification model, and outputting the risk identification result of the goods to be outbound; wherein the feature data includes the similarity; and generating a differentiated target outbound verification processing strategy based on the risk identification result.

[0006] According to an embodiment of the present disclosure, the above-mentioned risk identification result includes: the error risk probability of the above-mentioned goods to be shipped. According to the above-mentioned risk identification result, a differentiated target shipping verification and processing strategy is generated, including: determining the relationship between the error risk probability of the above-mentioned goods to be shipped and multiple preset risk intervals; the above-mentioned multiple preset risk intervals correspond to different shipping verification and processing strategies; when the above-mentioned error risk probability falls into the target risk interval, the shipping verification and processing strategy corresponding to the above-mentioned target risk interval is determined as the target shipping verification and processing strategy for the above-mentioned goods to be shipped.

[0007] According to an embodiment of the present disclosure, when the error risk probability of the first goods to be shipped exceeds a set threshold, the target shipping verification processing strategy for the first goods to be shipped includes: marking the first goods to be shipped as a high-risk label; dividing the goods to be shipped marked with the high-risk label into the same target review group; performing centralized picking and centralized verification processing on the goods to be shipped in the target review group, and adopting a multiple verification method during the centralized verification process.

[0008] According to an embodiment of the present disclosure, when the error risk probability of the second goods to be shipped out is less than a set threshold, the target shipping verification processing strategy for the second goods to be shipped out includes: marking the second goods to be shipped out with a low-risk label; performing a single verification on the goods to be shipped out marked with a low-risk label or shipping out after the picking information is confirmed, and the picking information confirmation is used to confirm that the picking information is consistent with the information on the picking task list.

[0009] According to an embodiment of the present disclosure, the training samples of the above-mentioned abnormal risk identification model are constructed based on the first multi-dimensional feature data of the first goods in the historical outbound goods and the second multi-dimensional feature data of the second goods; the above-mentioned first goods include at least one of the following: goods with abnormal outbound verification in the above-mentioned historical outbound goods, and goods with problems reported by users in the above-mentioned historical outbound goods; the above-mentioned second goods are goods with normal outbound verification results. In addition to the feature data of similarity, the above-mentioned first multi-dimensional feature data and the above-mentioned second multi-dimensional feature data also include at least one of the following feature data: multi-batch feature values, multi-specification feature values, and combination package feature values. Among them, the above-mentioned first multi-dimensional feature data and the above-mentioned second multi-dimensional feature data are used as training samples; or, the feature data of each dimension in the above-mentioned first multi-dimensional feature data and the above-mentioned second multi-dimensional feature data are amplified to obtain the amplified feature data in each dimension; the amplified feature data in the corresponding multiple dimensions are used as training samples.

[0010] According to an embodiment of the present disclosure, the above-mentioned abnormal risk identification model is a classification model; in the training stage, the above-mentioned training data are respectively used as the input of the above-mentioned classification model, and the above-mentioned classification model outputs the corresponding error risk prediction probability; the real review results of the above-mentioned first goods and the above-mentioned second goods are used as training labels to train the above-mentioned classification model; wherein, for the real review results of the review abnormality, the probability value corresponding to the assigned training label is 1 or the numerical interval [b, 1], 0<b<1; for the real review results of the review normality, the probability value corresponding to the assigned training label is 0 or the numerical interval [0, a], a<0<b. The above-mentioned classification model includes one of the following models: random forest model, multi-layer perceptron model, logistic regression model, support vector machine model. In the case where the above-mentioned classification model is a random forest model, the feature data or amplified feature data of each dimension in the multi-dimensional feature data in the above-mentioned training samples is used as the input of the corresponding single-dimensional decision tree model, and the single-dimensional decision tree model outputs the error risk prediction probability under the corresponding dimension; the output results of each single-dimensional decision tree model are weighted to obtain the error risk prediction probability corresponding to the multi-dimensional feature; the total number of the above-mentioned single-dimensional decision tree models is the same as the total number after the first multi-dimensional feature data and the second multi-dimensional feature data are dimensionally merged.

[0011] According to an embodiment of the present disclosure, the above-mentioned product information includes at least one of the following: text description information and product image of the above-mentioned product to be shipped. According to the above-mentioned product information, the similarity between the above-mentioned product to be shipped and similar products in the warehouse goods is determined, including: screening based on a pre-constructed text similarity result library to obtain a first similar product matching the above-mentioned text description information and a corresponding text similarity result; determining the maximum value in the above-mentioned text similarity result as the similarity of the above-mentioned product to be shipped; wherein the above-mentioned text similarity result library is generated by calculating the similarity of the text description information of all the goods in the warehouse goods in pairs. Alternatively, according to the above-mentioned product information, the similarity between the above-mentioned product to be shipped and similar products in the warehouse goods is determined, including: screening based on a pre-constructed image similarity result library to obtain a second similar product matching the above-mentioned product image and a corresponding image similarity result; determining the maximum value in the above-mentioned image similarity result as the similarity of the above-mentioned product to be shipped; wherein the above-mentioned image similarity result library is generated by calculating the similarity of the product images of all the goods in the warehouse goods in pairs. Alternatively, based on the above-mentioned product information, determining the similarity between the above-mentioned products to be shipped and similar products in the warehouse includes: screening based on a pre-constructed text similarity result library to obtain a first similar product matching the above-mentioned text description information and a corresponding text similarity result; screening based on a pre-constructed image similarity result library to obtain a second similar product matching the above-mentioned product image and a corresponding image similarity result; generating the similarity of the above-mentioned products to be shipped based on the above-mentioned text similarity result and the above-mentioned image similarity result.

[0012] According to an embodiment of the present disclosure, the above-mentioned product information includes: text description information and product images of the above-mentioned products to be shipped out. Based on the above-mentioned product information, the similarity between the above-mentioned products to be shipped out and similar products in the warehouse is determined, including: screening based on a pre-constructed text-image similarity result library to obtain a third similar product that matches the above-mentioned text description information and the above-mentioned product image and the corresponding text-image similarity result; and determining the maximum value in the above-mentioned text-image similarity result as the similarity of the above-mentioned products to be shipped out. Among them, the above-mentioned text-image similarity result library is generated by similarity calculation of splicing vector groups of all the products in the above-mentioned warehouse goods, and the above-mentioned splicing vector groups are obtained by splicing the text feature vectors corresponding to the text description information of the products and the image feature vectors corresponding to the product images.

[0013] In a second aspect, an embodiment of the present disclosure provides a device for outbound verification processing. The above-mentioned device includes: an information acquisition module, a similarity determination module, a risk identification module and an outbound verification strategy generation module. The above-mentioned information acquisition module is used to obtain the product information of the goods to be shipped out. The above-mentioned similarity determination module is used to determine the similarity between the above-mentioned goods to be shipped out and similar goods in the warehouse according to the above-mentioned product information. The above-mentioned risk identification module is used to input the characteristic data of the above-mentioned goods to be shipped out into a pre-trained abnormal risk identification model, and output the risk identification result of the above-mentioned goods to be shipped out; wherein the above-mentioned characteristic data includes the above-mentioned similarity. The above-mentioned outbound verification strategy generation module is used to generate differentiated target outbound verification processing strategies according to the above-mentioned risk identification results.

[0014] In a third aspect, an embodiment of the present disclosure provides an electronic device. The electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor is used to implement the method for warehouse check processing as described above when executing the program stored in the memory.

[0015] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for warehouse check processing described above is implemented.

[0016] The above technical solution provided by the embodiments of the present disclosure has at least some or all of the following advantages:

[0017] By obtaining the product information of the goods to be shipped; determining the similarity between the goods to be shipped and similar goods in the warehouse according to the above product information; inputting the feature data of the goods to be shipped into a pre-trained abnormal risk identification model, and outputting the risk identification result of the goods to be shipped; since the similarity between the goods to be shipped and other similar goods in the warehouse is used as feature data of one dimension, the feature data under this dimension can indicate the degree to which the goods to be shipped are easily confused with other goods; inputting the feature data of the goods to be shipped into the pre-trained abnormal risk identification model, an efficient and comprehensive evaluation of the possible error risks of the goods to be shipped in the picking process of the shipment can be achieved, and differentiated shipment verification and processing strategies can be adopted for the goods to be shipped that are easy to confuse and easy to confuse, thereby accurately improving the review efficiency and review accuracy of the situations that are easy to confuse and easy to confuse, and can also reduce or even avoid the review manpower and time cost of some goods that are not easy to make mistakes; thereby solving the problem of manpower and time consumption and little effect caused by indiscriminate execution of the same review processing strategy in the related technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or related technical descriptions are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0020] Figure 1 The system architecture of the method for checking the outbound delivery of the present invention is schematically shown;

[0021] Figure 2 The flowchart of the method for checking the outbound delivery according to an embodiment of the present disclosure is schematically shown;

[0022] Figure 3A The detailed implementation flow chart of step S220 according to an embodiment of the present disclosure is schematically shown;

[0023] Figure 3B The detailed implementation flow chart of step S220 according to another embodiment of the present disclosure is schematically shown;

[0024] Figure 3C The detailed implementation flow chart of step S220 according to another embodiment of the present disclosure is schematically shown;

[0025] Figure 3D The detailed implementation flow chart of step S220 according to another embodiment of the present disclosure is schematically shown;

[0026] Figure 4 The structure block diagram of the device for warehouse check processing according to one embodiment of the present disclosure is schematically shown;

[0027] Figure 5 The structural block diagram of the electronic device provided by the embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0028] The purpose of checking the goods is to ensure that the quantity, specifications, item number, color, product grade (good / defective) and other dimensions of the goods shipped are completely consistent with the order. Considering that most of the related technologies rely on manual checking of the goods shipped, due to various complex factors such as similar names of goods in the warehouse, multiple batches of goods, multiple specifications of goods (such as color, size), sets of goods or gifts, it will affect the accuracy of the checker's judgment on whether the goods are shipped normally. In addition, some goods with high error risks and some goods with low error risks are usually mixed together for review without distinction, which will reduce the attention of the checker to the goods with high error risks. Therefore, if you are not careful, you may check some goods that have already had outbound errors (such as missing accessories, missing gifts, color and size do not match the order, etc.) as normal during the verification process, which not only increases the manpower cost and time cost required for the review, but also has the problem of poor review effect and some review errors.

[0029] In view of this, the embodiments of the present disclosure provide a method, device, electronic device and medium for outbound verification processing, the method comprising: obtaining product information of the goods to be shipped; determining the similarity between the goods to be shipped and similar goods in the warehouse based on the product information; inputting the feature data of the goods to be shipped into a pre-trained abnormal risk identification model, and outputting the risk identification result of the goods to be shipped; wherein the feature data includes the similarity; and generating a differentiated target outbound verification processing strategy based on the risk identification result.

[0030] Since the similarity between the goods to be shipped and other similar goods in the warehouse is used as the feature data of one dimension, the feature data under this dimension can indicate the degree to which the goods to be shipped are easily confused with other goods; the feature data of the goods to be shipped are input into the pre-trained abnormal risk identification model to achieve efficient and comprehensive evaluation of the possible error risks of the goods to be shipped during the picking process of the goods to be shipped, and differentiated outbound verification and processing strategies can be adopted for the goods to be shipped that are easy to be confused and easy to be confused, thereby accurately improving the review efficiency and review accuracy of the situations that are easy to be confused and easy to be confused, and can also reduce or even avoid the manpower and time cost of reviewing some goods that are not easy to make mistakes; thereby solving the problem of manpower and time consumption and little effect caused by indiscriminate execution of the same review processing strategy in related technologies.

[0031] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present disclosure.

[0032] Figure 1 The system architecture of the method for warehouse outbound verification processing applicable to the embodiment of the present disclosure is schematically shown.

[0033] Reference Figure 1 As shown, the system architecture 100 of the method for warehouse check processing applicable to the embodiment of the present disclosure includes: terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is a medium that provides a communication link between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0034] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. The terminal devices 101, 102, 103 can be installed with warehousing applications for managing inventory, picking, and outbound processes; other communication client applications can also be installed on the above terminal devices, such as shopping applications, web browser applications, search applications, video playback applications, instant messaging tools, email clients, social platform software, etc. (only for example).

[0035] The terminal devices 101, 102, 103 may be electronic devices with display screens, such as but not limited to: smart phones, tablet computers, laptop computers, picking robots, user intelligent assistant devices (PDAs), etc.

[0036] In some implementation scenarios, the user may be a picker who picks goods, or a checker who performs outbound management or outbound verification. The user may carry a terminal device with him or her to perform interactive operations such as picking goods, outbound management, or outbound verification.

[0037] Alternatively, in other implementation scenarios, the picking and outbound verification process does not require manual execution, and can be performed by a terminal device similar to a picking robot according to a set program corresponding to a preset method.

[0038] The server 105 may be a server for providing services for the above-mentioned warehousing applications. The above-mentioned server may be an application server or a cloud server, etc. In some embodiments, the server 105 may execute the method for outbound verification processing provided by the embodiment of the present disclosure, obtain the target outbound verification processing strategy corresponding to the risk identification result, and feed it back to the terminal device, and the picking robot (an example of a terminal device) or the user (for example, a verification person or a picking person) may execute the corresponding verification process according to the prompt of the target outbound verification processing strategy.

[0039] It should be noted that the method for outbound verification processing provided in the embodiment of the present disclosure can generally be executed by the server 105 or a terminal device with a certain computing capability. Accordingly, the device for outbound verification processing provided in the embodiment of the present disclosure can generally be set in the server 105 or in the above-mentioned terminal device with a certain computing capability. The method for outbound verification processing provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with at least one of the terminal devices 101, 102, 103 and the server 105. Accordingly, the device for outbound verification processing provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with at least one of the terminal devices 101, 102, 103 and the server 105.

[0040] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0041] The first exemplary embodiment of the present disclosure provides a method for checking and processing outbound shipments. The method for checking and processing outbound shipments provided by the embodiment of the present disclosure can generally be executed by the server 105 or a terminal device with certain computing capabilities.

[0042] Figure 2 The flowchart of the method for warehouse outbound verification processing according to an embodiment of the present disclosure is schematically shown.

[0043] Reference Figure 2 As shown, the method for warehouse outbound verification processing provided by the embodiment of the present disclosure includes the following steps: S210, S220, S230 and S240.

[0044] In step S210, the product information of the products to be shipped out is obtained.

[0045] The goods to be shipped can be one or more goods in the shipping task. The shipping task contains one or more orders. An order can correspond to one or more goods. The quantity of the goods is not required and can be one or more.

[0046] According to an embodiment of the present disclosure, the above-mentioned product information includes at least one of the following: text description information of the above-mentioned products to be shipped out, and product images.

[0047] The above-mentioned text description information is information that describes the content, characteristics or functions of the goods to be shipped, such as the name, quantity, size / dimension, color, specification model, grade (qualified products), brand, whether there are gifts, whether there are set products, suitable groups of people (such as young people, the elderly), suitable scenes or venues (such as outdoor, mountaineering), etc. of the goods to be shipped.

[0048] In some embodiments, the text description information can be obtained by the server from the information corresponding to the product details page of the order system (here corresponding to part of the information stored on the background server), or by combining the information obtained from the product details page with the product label information stored in the database.

[0049] The product image refers to the electronic image of the product to be shipped obtained in the server for display in the warehousing application, for example, it can be a display image pulled from the display position (a container) corresponding to the order system. It can also be a product image reconstructed in the warehousing system.

[0050] In step S220, based on the product information, the similarity between the products to be shipped and similar products in the warehouse is determined.

[0051] Figure 3A The detailed implementation flow chart of step S220 according to an embodiment of the present disclosure is schematically shown.

[0052] Reference Figure 3A As shown, in the above step S220, based on the above product information, determining the similarity between the above products to be shipped and similar products in the warehouse includes the following steps: S311 and S312.

[0053] In step S311, a screening is performed based on a pre-built text similarity result library to obtain a first similar product matching the above text description information and a corresponding text similarity result.

[0054] The above-mentioned text similarity result library is generated by calculating the similarity of the text description information of all the goods in the above-mentioned warehouse goods. For example, the text similarity result library stores the text description information of two goods objects and the corresponding similarity results. The matching pair of the goods to be shipped out in the text similarity result library and the corresponding first similar goods can be located through the text description information, and the corresponding text similarity result can be obtained. The text similarity result can be one or more.

[0055] In some embodiments, the above text similarity result library is constructed based on a text matching model (a deep learning model). For example, the text matching model includes: a text embedding layer, a similarity calculation layer, and a deep learning network layer.

[0056] The specific process of building a text similarity result library includes:

[0057] Obtain text description information of two product objects to be compared for similarity among the above warehouse products;

[0058] Based on the text embedding layer, perform word segmentation and word embedding feature extraction on the text description information of the two product objects to obtain the corresponding first word vector group and second word vector group respectively;

[0059] Based on the similarity calculation layer, the similarity between the first word vector group and the second word vector group is calculated to obtain a similarity matrix; the first word vector group and the second word vector group are both two-dimensional arrays, and each dimension is a feature vector corresponding to each morpheme in each product object after word embedding feature processing; for example, the first word vector group contains W morphemes, and the second word vector group contains V morphemes, then the first word vector group contains feature vectors corresponding to W morphemes, and the second word vector group contains feature vectors corresponding to V morphemes; after similarity calculation, a W×V-dimensional similarity matrix is ​​obtained; the specific similarity calculation method can be calculated by dot product similarity or cosine similarity;

[0060] Based on the above deep learning network layer, feature extraction and soft maximization processing are performed on the above similarity matrix to obtain a text similarity result in the interval [0, 1].

[0061] In some embodiments, the first word vector group and the second word vector group are the result of dimension alignment. Alternatively, in other embodiments, there is a dimension difference between the first word vector group and the second word vector group, and the size of the feature map obtained by feature extraction is adjusted based on the dynamic pooling layer in the deep learning network, so that the size of the feature map input to the fully connected layer of the deep learning network is consistent.

[0062] In some embodiments, during the word segmentation process, some functional modules of existing word segmentation systems (such as the jieba word segmentation system) can also be called to perform word segmentation processing; in other embodiments, in order to improve the word segmentation accuracy of professional systems, some industry dictionary data can also be imported into the above-mentioned functional modules as custom dictionary data; then word segmentation processing is performed based on the word segmentation system based on the imported custom dictionary data.

[0063] As an example, the text description information of product A is: "Outdoor sports shoes for men, model 6639, blue, size 42"; the text description information of product B is: "Versatile casual leather shoes for men, model WE005, black, size 43". Take the calculation of the text similarity between two products A and B as an example. For example, the text description information of product A can be segmented to obtain six morphemes (tokens, which are basic units such as individual words or characters): outdoor, sports shoes, men, model 6639, blue, and size 42. After the text description information of product B is segmented, the seven morphemes are obtained: versatile, casual, leather shoes, men, model WE005, black, and size 43.

[0064] In some embodiments, the total dimensions corresponding to the morphemes of the two products A and B are padded and mapped into word embedding vectors.

[0065] Specifically, the length reference is taken from the two products A and B with more morphemes, and the number of morphemes with less is supplemented. Then the length reference number can be determined to be 7 (the total number of morphemes corresponding to product B), and a zero vector is filled for product A with a length less than 7. Then the morpheme sequences corresponding to products A and B are {outdoor, sports shoes, men, 6639 style, blue, size 42, 0 (0 is the data supplemented for alignment)} and {all-match, casual, leather shoes, men, WE005 style, black, size 43} respectively.

[0066] For each morpheme in each morpheme sequence, a pre-trained word2vec model (as an example of a text embedding layer) can be used to generate a word vector of a fixed length (for example, 100 dimensions, 200 dimensions, etc.). The word2vec model is trained and constructed based on relevant corpus. The word2vec model can convert Chinese morphemes into numerical vectors. The training corpus can be determined based on the classified application fields involved in the goods stored in the warehouse. For example, if the warehouse is a clothing warehouse, a fresh food warehouse, or an information appliance (3C, including computers, communications, and consumer electronics) warehouse, the product content in the corresponding field is used as the training corpus.

[0067] After obtaining the word embedding feature vectors corresponding to the morphemes of commodities A and B respectively, similarity calculation can be performed based on the first word vector group and the second word vector group corresponding to each commodity to obtain a similarity matrix.

[0068] Afterwards, feature extraction is performed on the similarity matrix based on the above deep learning network layer to obtain the text similarity result.

[0069] In some embodiments, the above-mentioned deep learning network layers include: convolutional layers, pooling layers, activation layers and fully connected layers. The specific number of layers can be adjusted and optimized according to the training results.

[0070] For example, Table 1 illustrates some text similarity results in the text similarity result library.

[0071] Table 1 Examples of text similarity results in the text similarity result library

[0072] Product Name 1 Product Name 2 Similarity score Men's Slam Dunk T-shirt - Black Size L Men's Slam Dunk T-Shirt - Black XL 0.9 Women's Floral Dress - White Size L Men's Slam Dunk T-Shirt - Black XL 0 Mengniu pure milk 20 boxes Mengniu pure milk 24 boxes 0.8 Dutch potatoes 500g Dutch potatoes 600g 0.95 …… …… ……

[0073] In step S312, the maximum value among the above text similarity results is determined as the similarity of the above goods to be shipped out.

[0074] For example, by determining the maximum value of one or more text similarity results as the similarity of the goods to be shipped, the most likely confusion of the goods to be shipped is used as an influencing factor in evaluating whether abnormalities are likely to occur during the review process, thereby covering risky goods to a greater extent.

[0075] In the embodiment including steps S311 and S312, the first similar goods matching the above text description information and the corresponding text similarity results are obtained by screening based on the pre-built text similarity result library, and the maximum value of one or more text similarity results is determined as the similarity of the goods to be shipped out, and the maximum possible confusion of the goods to be shipped out is used as an influencing factor to evaluate whether it is easy to have abnormalities in the review process, which can cover the goods with abnormal risks in the review to a large extent from the perspective of easy confusion of text descriptions. Since the text similarity result library is pre-generated, when processing the goods to be shipped out, the corresponding time of real-time calculation is saved, and the processing time of direct matching is fast.

[0076] In other embodiments, the similarity may be determined based only on the dimension of the product image; or the similarity may be determined by combining the two dimensions of the product image and the text description information.

[0077] Figure 3B The detailed implementation flow chart of step S220 according to another embodiment of the present disclosure is schematically shown.

[0078] Reference Figure 3B As shown, in the above step S220, based on the above product information, determining the similarity between the above products to be shipped and similar products in the warehouse includes the following steps: S321 and S322.

[0079] In step S321, a screening is performed based on a pre-constructed image similarity result library to obtain a second similar product matching the product image and a corresponding image similarity result.

[0080] The image similarity result library is generated by calculating the similarity of the product images of all the products in the warehouse. For example, the image similarity result library stores product images of two product objects and the corresponding similarity results. The matching pair of the product to be shipped and the corresponding second similar product in the image similarity result library can be located through the product image, and the corresponding image similarity result can be obtained. The image similarity result can be one or more.

[0081] When constructing the image similarity result library, in the process of calculating the image similarity results of the product images, various existing or improved image matching algorithms can be used, such as but not limited to: hash algorithm (for example, determining the similarity between two product images by comparing the Hamming distance between the hash values ​​of the two product images), matching algorithm based on SIFT (scale-invariant feature transform matching algorithm), matching algorithm based on Gist (general search tree) features, and matching algorithm based on ResNet (residual network) full connection in the field of deep learning, etc.

[0082] In step S322, the maximum value among the above image similarity results is determined as the similarity of the above goods to be shipped out.

[0083] By determining the maximum value of one or more image similarity results as the similarity of the goods to be shipped, the most likely confusion of the goods to be shipped is used as an influencing factor in evaluating whether abnormalities are likely to occur during the review process, thereby covering goods with risks to a greater extent.

[0084] In the embodiment including steps S321 and S322, by screening based on the pre-built image similarity result library, a second similar product matching the product image and the corresponding image similarity result are obtained, and the maximum value of one or more image similarity results is determined as the similarity of the product to be shipped, and the maximum possible confusion of the product to be shipped is used as an influencing factor for evaluating whether it is easy to have anomalies in the review process, which can cover the products with abnormal risks in the review to a large extent from the perspective of easy image confusion. Since the image similarity result library is pre-generated, the time corresponding to real-time calculation is saved when processing the product to be shipped, and the processing time of direct matching is fast.

[0085] Figure 3C The detailed implementation flow chart of step S220 according to another embodiment of the present disclosure is schematically shown.

[0086] Reference Figure 3C As shown, in the above step S220, based on the above product information, determining the similarity between the above products to be shipped and similar products in the warehouse includes the following steps: S331, S332 and S333.

[0087] In step S331, a screening is performed based on a pre-built text similarity result library to obtain a first similar product matching the above text description information and a corresponding text similarity result.

[0088] In step S332, a screening is performed based on a pre-constructed image similarity result library to obtain a second similar product matching the product image and a corresponding image similarity result.

[0089] In step S333, the similarity of the goods to be shipped out is generated according to the text similarity result and the image similarity result.

[0090] For example, a maximum value among multiple text similarity results and a maximum value among multiple image similarity results are weighted and calculated to obtain the similarity of the goods to be shipped out.

[0091] In the embodiment including steps S331 to S333, by screening based on a pre-constructed text similarity result library, a first similar product matching the above text description information and a corresponding text similarity result are obtained, by screening based on a pre-constructed image similarity result library, a second similar product matching the above product image and a corresponding image similarity result are obtained, and the maximum value among multiple text similarity results and the maximum value among multiple image similarity results are weighted and calculated to obtain the similarity of the goods to be shipped out. This similarity comprehensively considers the comprehensive similarity of each product in the two dimensions of image and text, and can cover the goods with abnormal risks in the review to a large extent from the perspective of easy confusion of images and easy confusion of texts.

[0092] Figure 3D The detailed implementation flow chart of step S220 according to another embodiment of the present disclosure is schematically shown.

[0093] According to an embodiment of the present disclosure, the above-mentioned product information includes: text description information and product images of the above-mentioned products to be shipped out.

[0094] Reference Figure 3D As shown, in the above step S220, based on the above product information, determining the similarity between the above products to be shipped and similar products in the warehouse includes the following steps: S341 and S342.

[0095] In step S341, a screening is performed based on a pre-constructed text-image similarity result library to obtain a third similar product that matches the above text description information and the above product image and a corresponding text-image similarity result.

[0096] The above-mentioned text-image similarity result library is generated by similarity calculation of the splicing vector groups of all goods in the above-mentioned warehouse goods, and the above-mentioned splicing vector groups are obtained by splicing the text feature vectors corresponding to the text description information of the goods and the image feature vectors corresponding to the goods images. For example, the text-image similarity result library stores the splicing information of the text description information of two goods objects-the splicing information of the goods image and the corresponding similarity results. Through the text description information and the goods image, the matching pair of the goods to be shipped out in the text-image similarity result library and the corresponding third similar goods can be located, and the corresponding text-image similarity results can be obtained. The text-image similarity results can be one or more.

[0097] In step S342, the maximum value among the above text-image similarity results is determined as the similarity of the above goods to be shipped out.

[0098] In the embodiment including steps S341 and S342, by screening based on the pre-built text-image similarity result library, a third similar product matching the above text description information and the above product image and the corresponding text-image similarity result are obtained, and the maximum value of one or more text-image similarity results is determined as the similarity of the product to be shipped out, and the maximum possible confusion of the product to be shipped out is used as an influencing factor for evaluating whether it is easy to have anomalies in the review process, and from the perspective of the two comprehensive dimensions of easy confusion of text and image, the products with abnormal risks in the review can be covered to a large extent. Since the text-image similarity result library is pre-generated, when processing the products to be shipped out, the corresponding time of real-time calculation is saved, and the processing time of direct matching is fast.

[0099] In step S230, the feature data of the goods to be shipped are input into a pre-trained abnormal risk identification model, and the risk identification result of the goods to be shipped is output; wherein the feature data includes the similarity.

[0100] According to an embodiment of the present disclosure, the training samples of the above-mentioned abnormal risk identification model are constructed based on the first multi-dimensional feature data of the first product and the second multi-dimensional feature data of the second product in the historical outbound products.

[0101] The first product includes at least one of the following: products with abnormal delivery verification in the historical delivery products, and products with problems reported by users in the historical delivery products; the second product is a product with normal delivery verification results. The first multi-dimensional feature data and the second multi-dimensional feature data include: the similarity and at least one of the following feature data: multi-batch feature values, multi-specification feature values, and combination package feature values.

[0102] For example, a product S has 5 batches, a total of xs, s, m, L and XL sizes, three colors of red, blue and yellow, and no combination or gift sets, etc., then the corresponding feature data is: the similarity of product S (obtained using step 220 described above), the multi-batch feature value is 5, the multi-specification feature value is 8, and the combination set feature value is 0.

[0103] In some embodiments, the first multi-dimensional feature data and the second multi-dimensional feature data are used as training samples.

[0104] In other embodiments, the feature data of each dimension in the first multi-dimensional feature data and the second multi-dimensional feature data are amplified to obtain amplified feature data in each dimension; and the amplified feature data in the corresponding multiple dimensions are used as training samples.

[0105] For example, in order to amplify the impact of different types and quantities of single-dimensional features on the risk of abnormal review, the feature data of each dimension can be amplified. The processing method can be that when the feature value is not 0, the feature value (including similarity and other feature values) is used as the exponential value of the e index. For example, after the feature value of multiple batches of 5 is amplified, it becomes e 5 , after the multi-specification characteristic value 8 is enlarged, it becomes e 8 .

[0106] In other embodiments, the sample data can be transformed by a certain mathematical method to make the original data dimensionless, that is, all index values ​​are at the same level, to avoid the impact of data dimension on model training, such as converting the number of goods into a compressed value between [0, 1] through min-max standardization or z-score standardization. Normalizing the data can speed up the process of model training and improve the efficiency of model training.

[0107] According to an embodiment of the present disclosure, the above abnormal risk identification model is a classification model.

[0108] During the training stage, the training data are used as inputs of the classification model, and the classification model outputs corresponding error risk prediction probabilities; the actual review results of the first product and the second product are used as training labels to train the classification model; wherein, for the actual review results with abnormal reviews, the probability values ​​corresponding to the training labels are assigned as 1 or in the numerical range of [b, 1], where 0<b<1; and for the actual review results with normal reviews, the probability values ​​corresponding to the training labels are assigned as 0 or in the numerical range of [0, a], where a<0<b.

[0109] In some embodiments, the above classification model includes one of the following models: a random forest model, a multi-layer perceptron model, a logistic regression model, and a support vector machine model.

[0110] In the case where the above-mentioned classification model is a random forest model, the feature data or amplified feature data of each dimension in the multi-dimensional feature data in the above-mentioned training samples is used as the input of the corresponding single-dimensional decision tree model, and the single-dimensional decision tree model outputs the error risk prediction probability under the corresponding dimension; the output results of each single-dimensional decision tree model are weighted to obtain the error risk prediction probability corresponding to the multi-dimensional feature; the total number of the above-mentioned single-dimensional decision tree models is the same as the total number after the first multi-dimensional feature data and the second multi-dimensional feature data are dimensionally merged.

[0111] For example, sample data of multiple dimensions (such as similarity, multi-batch feature values, multi-specification feature values, and combined package feature values) are used as inputs to the random forest model. The random forest model is trained, and the information gain rate or Gini coefficient is used as the loss function to determine the various parameters in the random forest model, such as the number of trees in the forest, the depth of the tree, and the number of leaf nodes, so as to obtain an abnormal risk identification model.

[0112] In some embodiments, weighted processing of the output results of each single-dimensional decision tree model may be performed by calculating the corresponding probability mean under the classification of each dimension (for example, divided into 2 or 3 classifications).

[0113] For example, based on the feature data of each dimension of the four dimensions such as similarity, batch quantity, combination quantity, color and size type in multiple historical outbound goods and the actual outbound abnormality result corresponding to each historical outbound goods as training labels, the decision tree model corresponding to each dimension is trained. The classification of the decision tree model can be binary or more classifications (in the actual situation corresponding to more classifications, there may be outbound verification determination abnormality, determination normality, and outbound risk prediction as neutral state, etc.). Here, a three-classification example is used, as shown in the decision trees 1-4 in the examples of Tables 2-5 below. Then, the decision tree models corresponding to each dimension are combined to obtain a random forest model.

[0114] Table 2 Decision tree model corresponding to similarity dimension 1

[0115] Product similarity abnormal neutral normal ≤0.1 1％ 14％ 85％ 0.1-0.4 30％ 40％ 30％ 0.4-0.8 60％ 15％ 25％ ≥0.8 90％ 5％ 5％

[0116] Table 3 Decision tree model corresponding to batch quantity dimension 2

[0117] Batch Quantity abnormal neutral normal 1 0％ 0％ 100％ 1-5 35％ 20％ 45％ 5-10 50％ 15％ 35％ ≥10 70％ 10％ 20％

[0118] Table 4 Decision tree model 3 corresponding to the combination product dimension

[0119] Combination abnormal neutral normal 1 15％ 10％ 75％ 1-3 15％ 15％ 70％ 3-6 50％ 40％ 10％ ≥6 25％ 5％ 70％

[0120] Table 5 Decision tree model corresponding to color size dimension 4

[0121] Color size abnormal neutral normal 1 0％ 20％ 80％ 2-4 35％ 25％ 40％ >4 60％ 20％ 20％

[0122] The feature data of the four dimensions of similarity, batch quantity, combination quantity and color, size and type of the goods to be shipped correspond to 0.74, 3, 4 and 5 respectively. The feature data of each dimension of the goods to be shipped are matched with the feature data range of the corresponding dimension in the abnormal risk identification model to obtain the decision results corresponding to each dimension, that is, the decision probability of the abnormal risk possibility corresponding to the feature data range of each dimension of the commodity, as shown in Tables 2 to 5. Then, the output results of each single-dimensional decision tree model are weighted to obtain the error risk prediction probability corresponding to the multi-dimensional features. For example, the mean decision probability of the abnormal risk possibility under each dimension can be calculated, and the type corresponding to the maximum value of the mean decision probability (such as the abnormal type in Table 6 below) is used as the error risk prediction probability and the corresponding type (abnormal type) of the goods to be shipped.

[0123] Table 6 Random forest model

[0124] Dimensions Feature data abnormal neutral normal Product similarity score 0.74 60％ 15％ 25％ Batch Quantity 3 35％ 20％ 45％ Number of Combinations 4 50％ 40％ 10％ Color size type 5 60％ 20％ 20％ Selection probability -- 51.3％ 23.7％ 25％

[0125] It should be noted that the training label is normal or abnormal, represented by a value between 0 and 1 (for example, described as a probability value, which can be calculated by the frequency of problems in multiple historical review results). In different value ranges, such as 0 to 0.2, it is considered normal, 0.3 to 0.7 is considered neutral, and greater than 0.7 is considered abnormal. In this way, continuous probability values ​​can be mapped to different categories (normal, abnormal, or neutral risk categories); accordingly, when the random forest model performs classification prediction, it will also predict the probabilities under the corresponding multiple categories.

[0126] In step S240, a differentiated target outbound check processing strategy is generated based on the above risk identification results.

[0127] According to an embodiment of the present disclosure, the above risk identification result includes: the error risk probability of the above goods to be shipped. For example, the above error risk probability is in different probability intervals, which may correspond to different risk categories, and different risk categories correspond to different risk intervals, corresponding to different shipping verification and processing strategies.

[0128] For example, it can be divided into two or more risk categories. If the error risk probability is less than 60% (the specific value can vary), it is considered a low-risk category; if the error risk probability is above 60%, it is considered a high-risk category; the outbound verification and processing strategies corresponding to the low-risk category and the high-risk category are different.

[0129] In other embodiments, it can be divided into more categories and corresponding subdivided warehouse outbound verification processing strategies.

[0130] Among them, in the above step S240, based on the above risk identification results, differentiated target outbound verification and processing strategies are generated, including: determining the relationship between the error risk probability of the above-mentioned goods to be shipped out and multiple preset risk intervals; the above-mentioned multiple preset risk intervals correspond to different outbound verification and processing strategies; when the above-mentioned error risk probability falls into the target risk interval, the outbound verification and processing strategy corresponding to the above-mentioned target risk interval is determined as the target outbound verification and processing strategy for the above-mentioned goods to be shipped out.

[0131] According to an embodiment of the present disclosure, when the error risk probability of the first goods to be shipped exceeds a set threshold (for example, the set threshold is 60%, 70% or 80%), the target shipping verification processing strategy for the first goods to be shipped includes: marking the first goods to be shipped as a high-risk label; dividing the goods to be shipped marked with the high-risk label into the same target review group; performing centralized picking and centralized verification processing on the goods to be shipped in the target review group, and taking multiple verifications during the centralized verification process.

[0132] According to an embodiment of the present disclosure, when the error risk probability of the second goods to be shipped out is less than a set threshold, the target shipping verification processing strategy for the second goods to be shipped out includes: marking the second goods to be shipped out with a low-risk label; performing a single verification on the goods to be shipped out marked with a low-risk label or shipping out after the picking information is confirmed, and the picking information confirmation is used to confirm that the picking information is consistent with the information on the picking task list.

[0133] For example, when the goods in the order are reviewed for shipment, it is determined whether the goods in the order to be reviewed are goods that are easy to be confused and error-prone based on the similarity between the goods information of the goods to be shipped and the various goods in the warehouse; if so, the target shipment verification processing strategy received by the terminal device from the server or processed by the terminal device itself is: prompting the reviewer to conduct multiple checks; and multiple orders that are prone to errors are gathered together and assigned to reviewers or pickers for centralized picking and centralized review. In this way, in the process of centralized picking or centralized verification of goods that are prone to errors, attention can be focused on verifying the parts that may be wrong. Compared with the process of mixing various types of goods together and checking them equally in the related art, the review efficiency and accuracy of situations that are easy to be confused and error-prone can be accurately improved, and the manpower and time costs of reviewing some goods that are not prone to errors can be reduced or even avoided; thereby solving the problem of manpower and time consumption and little effect caused by indiscriminate execution of the same review processing strategy in the related art.

[0134] In summary, in the embodiment including steps S210 to S240, the product information of the goods to be shipped is obtained; based on the product information, the similarity between the goods to be shipped and similar goods in the warehouse is determined; the feature data of the goods to be shipped is input into a pre-trained abnormal risk identification model, and the risk identification result of the goods to be shipped is obtained as an output; since the similarity between the goods to be shipped and other similar goods in the warehouse is used as feature data of one dimension, the feature data under this dimension can indicate the degree to which the goods to be shipped are easily confused with other goods; the feature data of the goods to be shipped is input into a pre-trained abnormal risk identification model, and the risk identification result of the goods to be shipped is obtained as an output; The collected data is input into a pre-trained abnormal risk identification model to achieve an efficient and comprehensive assessment of the possible error risks of goods to be shipped during the picking process. Differentiated outbound verification and processing strategies can be adopted for goods to be shipped that are easy to be confused or prone to errors, thereby accurately improving the review efficiency and review accuracy for situations that are easy to be confused or prone to errors, and can also reduce or even avoid the manpower and time costs of reviewing some goods that are not prone to errors; thereby solving the problem of indiscriminate execution of the same review processing strategy in related technologies, which leads to manpower and time consumption with little effect.

[0135] A second exemplary embodiment of the present disclosure provides an apparatus for warehouse-out verification processing.

[0136] Figure 4 The structure block diagram of the device for warehouse outbound verification processing according to an embodiment of the present disclosure is schematically shown.

[0137] Reference Figure 4 As shown, the device 400 for warehouse-out verification processing provided by the embodiment of the present disclosure includes: an information acquisition module 401, a similarity determination module 402, a risk identification module 403 and a warehouse-out verification strategy generation module 404.

[0138] The information acquisition module 401 is used to acquire the product information of the products to be shipped out.

[0139] The similarity determination module 402 is used to determine the similarity between the goods to be shipped and similar goods in the warehouse according to the goods information.

[0140] The risk identification module 403 is used to input the multi-dimensional feature data of the goods to be shipped into a pre-trained abnormal risk identification model, and output the risk identification results of the goods to be shipped; wherein the similarity of the goods to be shipped is used as the feature data of one dimension in the multi-dimensional feature data.

[0141] The above-mentioned outbound check strategy generation module 404 is used to generate differentiated target outbound check processing strategies according to the above-mentioned risk identification results.

[0142] According to an embodiment of the present disclosure, the above-mentioned risk identification result includes: the error risk probability of the above-mentioned goods to be shipped. According to the above-mentioned risk identification result, a differentiated target shipping verification and processing strategy is generated, including: determining the relationship between the error risk probability of the above-mentioned goods to be shipped and multiple preset risk intervals; the above-mentioned multiple preset risk intervals correspond to different shipping verification and processing strategies; when the above-mentioned error risk probability falls into the target risk interval, the shipping verification and processing strategy corresponding to the above-mentioned target risk interval is determined as the target shipping verification and processing strategy for the above-mentioned goods to be shipped.

[0143] According to an embodiment of the present disclosure, when the error risk probability of the first goods to be shipped exceeds a set threshold, the target shipping verification processing strategy for the first goods to be shipped includes: marking the first goods to be shipped as a high-risk label; dividing the goods to be shipped marked with the high-risk label into the same target review group; performing centralized picking and centralized verification processing on the goods to be shipped in the target review group, and adopting a multiple verification method during the centralized verification process.

[0144] According to an embodiment of the present disclosure, when the error risk probability of the second goods to be shipped out is less than a set threshold, the target shipping verification processing strategy for the second goods to be shipped out includes: marking the second goods to be shipped out with a low-risk label; performing a single verification on the goods to be shipped out marked with a low-risk label or shipping out after the picking information is confirmed, and the picking information confirmation is used to confirm that the picking information is consistent with the information on the picking task list.

[0145] According to an embodiment of the present disclosure, the training samples of the above-mentioned abnormal risk identification model are constructed based on the first multi-dimensional feature data of the first goods in the historical outbound goods and the second multi-dimensional feature data of the second goods; the above-mentioned first goods include at least one of the following: goods with abnormal outbound verification in the above-mentioned historical outbound goods, and goods with problems reported by users in the above-mentioned historical outbound goods; the above-mentioned second goods are goods with normal outbound verification results. In addition to the feature data of similarity, the above-mentioned first multi-dimensional feature data and the above-mentioned second multi-dimensional feature data also include at least one of the following feature data: multi-batch feature values, multi-specification feature values, and combination package feature values. Among them, the above-mentioned first multi-dimensional feature data and the above-mentioned second multi-dimensional feature data are used as training samples; or, the feature data of each dimension in the above-mentioned first multi-dimensional feature data and the above-mentioned second multi-dimensional feature data are amplified to obtain the amplified feature data in each dimension; the amplified feature data in the corresponding multiple dimensions are used as training samples.

[0146] According to an embodiment of the present disclosure, the above-mentioned abnormal risk identification model is a classification model; in the training stage, the above-mentioned training data are respectively used as the input of the above-mentioned classification model, and the above-mentioned classification model outputs the corresponding error risk prediction probability; the real review results of the above-mentioned first goods and the above-mentioned second goods are used as training labels to train the above-mentioned classification model; wherein, for the real review results of the review abnormality, the probability value corresponding to the assigned training label is 1 or the numerical interval [b, 1], 0<b<1; for the real review results of the review normality, the probability value corresponding to the assigned training label is 0 or the numerical interval [0, a], a<0<b. The above-mentioned classification model includes one of the following models: random forest model, multi-layer perceptron model, logistic regression model, support vector machine model. In the case where the above-mentioned classification model is a random forest model, the feature data or amplified feature data of each dimension in the multi-dimensional feature data in the above-mentioned training samples is used as the input of the corresponding single-dimensional decision tree model, and the single-dimensional decision tree model outputs the error risk prediction probability under the corresponding dimension; the output results of each single-dimensional decision tree model are weighted to obtain the error risk prediction probability corresponding to the multi-dimensional feature; the total number of the above-mentioned single-dimensional decision tree models is the same as the total number after the first multi-dimensional feature data and the second multi-dimensional feature data are dimensionally merged.

[0147] According to an embodiment of the present disclosure, the above-mentioned product information includes at least one of the following: text description information and product image of the above-mentioned product to be shipped. According to the above-mentioned product information, the similarity between the above-mentioned product to be shipped and similar products in the warehouse goods is determined, including: screening based on a pre-constructed text similarity result library to obtain a first similar product matching the above-mentioned text description information and a corresponding text similarity result; determining the maximum value in the above-mentioned text similarity result as the similarity of the above-mentioned product to be shipped; wherein the above-mentioned text similarity result library is generated by calculating the similarity of the text description information of all the goods in the warehouse goods in pairs. Alternatively, according to the above-mentioned product information, the similarity between the above-mentioned product to be shipped and similar products in the warehouse goods is determined, including: screening based on a pre-constructed image similarity result library to obtain a second similar product matching the above-mentioned product image and a corresponding image similarity result; determining the maximum value in the above-mentioned image similarity result as the similarity of the above-mentioned product to be shipped; wherein the above-mentioned image similarity result library is generated by calculating the similarity of the product images of all the goods in the warehouse goods in pairs. Alternatively, based on the above-mentioned product information, determining the similarity between the above-mentioned products to be shipped and similar products in the warehouse includes: screening based on a pre-constructed text similarity result library to obtain a first similar product matching the above-mentioned text description information and a corresponding text similarity result; screening based on a pre-constructed image similarity result library to obtain a second similar product matching the above-mentioned product image and a corresponding image similarity result; generating the similarity of the above-mentioned products to be shipped based on the above-mentioned text similarity result and the above-mentioned image similarity result.

[0148] According to an embodiment of the present disclosure, the above-mentioned product information includes: text description information and product images of the above-mentioned products to be shipped out. Based on the above-mentioned product information, the similarity between the above-mentioned products to be shipped out and similar products in the warehouse is determined, including: screening based on a pre-constructed text-image similarity result library to obtain a third similar product that matches the above-mentioned text description information and the above-mentioned product image and the corresponding text-image similarity result; and determining the maximum value in the above-mentioned text-image similarity result as the similarity of the above-mentioned products to be shipped out. Among them, the above-mentioned text-image similarity result library is generated by similarity calculation of splicing vector groups of all the products in the above-mentioned warehouse goods, and the above-mentioned splicing vector groups are obtained by splicing the text feature vectors corresponding to the text description information of the products and the image feature vectors corresponding to the product images.

[0149] For more details and specific implementation methods of this embodiment, reference can be made to the detailed description of the first embodiment, which will not be repeated here.

[0150] Any number of the functional modules included in the above-mentioned device 500 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. At least one of the functional modules included in the device 500 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware and firmware or in any appropriate combination of any of them. Alternatively, at least one of the functional modules included in the device 500 can be at least partially implemented as a computer program module, and when the computer program module is run, the corresponding function can be executed.

[0151] A fifth exemplary embodiment of the present disclosure provides an electronic device.

[0152] Figure 5 The structural block diagram of the electronic device provided by the embodiment of the present disclosure is schematically shown.

[0153] Reference Figure 5 As shown, the electronic device 500 provided by the embodiment of the present disclosure includes a processor 501, a communication interface 502, a memory 503 and a communication bus 504, wherein the processor 501, the communication interface 502 and the memory 503 communicate with each other through the communication bus 504; the memory 503 is used to store computer programs; the processor 501 is used to implement the method of warehouse verification processing as described above when executing the program stored in the memory.

[0154] The sixth exemplary embodiment of the present disclosure further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for warehouse-out verification processing as described above is implemented.

[0155] The computer-readable storage medium may be included in the device or apparatus described in the above embodiment; or it may exist independently without being assembled into the device or apparatus. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0156] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0157] It should be noted that the collection, collection, update, analysis, processing, use, transmission, storage and other aspects of user personal information involved in the technical solution provided by the embodiments of the present disclosure are in compliance with the provisions of relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken for user personal information to prevent illegal access to user personal information data and maintain the security of user personal information, network security and national security.

[0158] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0159] The foregoing is merely a specific embodiment of the present disclosure, which enables those skilled in the art to understand or implement the present disclosure. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features applied herein.

Claims

1. A method for checking and processing outbound goods, characterized in that: include: Get the product information of the goods to be shipped out; Determine the similarity between the goods to be shipped and similar goods in the warehouse according to the goods information; Inputting the characteristic data of the goods to be shipped into a pre-trained abnormal risk identification model, and outputting the risk identification result of the goods to be shipped; the characteristic data includes the similarity; Based on the risk identification results, differentiated target outbound verification and processing strategies are generated.

2. The method according to claim 1, characterized in that The risk identification result includes: the error risk probability of the goods to be shipped out; According to the risk identification results, a differentiated target outbound verification and processing strategy is generated, including: Determine the relationship between the error risk probability of the goods to be shipped and a plurality of preset risk intervals; the plurality of preset risk intervals correspond to different shipping verification and processing strategies; When the error risk probability falls within the target risk interval, the outbound check processing strategy corresponding to the target risk interval is determined as the target outbound check processing strategy for the goods to be outbound.

3. The method according to claim 2, characterized in that When the error risk probability of the first goods to be shipped exceeds a set threshold, the target shipping verification processing strategy for the first goods to be shipped includes: marking the first goods to be shipped with a high-risk label; dividing the goods to be shipped marked with the high-risk label into the same target review group; performing centralized picking and centralized verification processing on the goods to be shipped in the target review group, and adopting a multiple verification method during the centralized verification process.

4. The method according to claim 3, characterized in that When the error risk probability of the second goods to be shipped out is less than a set threshold, the target shipping verification processing strategy for the second goods to be shipped out includes: marking the second goods to be shipped out with a low-risk label; performing a single verification on the goods to be shipped out marked with a low-risk label or shipping out after the picking information is confirmed, and the picking information confirmation is used to confirm that the picking information is consistent with the information on the picking task list.

5. The method according to any one of claims 1 to 4, characterized in that The training samples of the abnormal risk identification model are constructed based on the first multi-dimensional feature data of the first product and the second multi-dimensional feature data of the second product in the historical outbound goods; the first product includes at least one of the following: the products with abnormal outbound verification in the historical outbound goods, and the products with problems reported by users in the historical outbound goods; the second product is the products with normal outbound verification results; In addition to the feature data of similarity, the first multi-dimensional feature data and the second multi-dimensional feature data further include at least one of the following feature data: multi-batch feature values, multi-specification feature values, and combination package feature values; wherein the first multi-dimensional feature data and the second multi-dimensional feature data are used as training samples; or, Amplify the feature data of each dimension in the first multi-dimensional feature data and the second multi-dimensional feature data to obtain amplified feature data in each dimension; and use the amplified feature data in the corresponding multiple dimensions as training samples.

6. The method according to claim 5, characterized in that The abnormal risk identification model is a classification model; in the training stage, the training data are respectively used as inputs of the classification model, and the classification model outputs the corresponding error risk prediction probability; The classification model is trained by using the real review results of the first product and the second product as training labels; wherein, for the real review results of abnormal review, the probability value corresponding to the assigned training label is 1 or the numerical range of [b, 1], 0<b<1; and for the real review results of normal review, the probability value corresponding to the assigned training label is 0 or the numerical range of [0, a], a<0<b; The classification model includes one of the following models: random forest model, multi-layer perceptron model, logistic regression model, support vector machine model; In the case where the classification model is a random forest model, the feature data or amplified feature data of each dimension in the multidimensional feature data in the training sample is used as the input of the corresponding single-dimensional decision tree model, and the single-dimensional decision tree model outputs the error risk prediction probability under the corresponding dimension; the output results of each single-dimensional decision tree model are weighted to obtain the error risk prediction probability corresponding to the multi-dimensional feature; the total number of the single-dimensional decision tree models is the same as the total number after the first multi-dimensional feature data and the second multi-dimensional feature data are dimensionally merged.

7. The method according to claim 5, characterized in that The product information includes at least one of the following: text description information of the product to be shipped and product images; Determining the similarity between the goods to be shipped and similar goods in the warehouse according to the goods information includes: Screening is performed based on a pre-built text similarity result library to obtain a first similar product matching the text description information and a corresponding text similarity result; the maximum value in the text similarity result is determined as the similarity of the product to be shipped out; wherein the text similarity result library is generated by calculating the similarity of the text description information of all the products in the warehouse; or, Screening is performed based on a pre-built image similarity result library to obtain a second similar product matching the product image and a corresponding image similarity result; the maximum value in the image similarity result is determined as the similarity of the product to be shipped out; wherein the image similarity result library is generated by calculating the similarity of product images of all products in the warehouse in pairs; or, Screening is performed based on a pre-constructed text similarity result library to obtain a first similar product matching the text description information and a corresponding text similarity result; screening is performed based on a pre-constructed image similarity result library to obtain a second similar product matching the product image and a corresponding image similarity result; and similarity of the product to be shipped is generated based on the text similarity result and the image similarity result.

8. The method according to claim 5, characterized in that The product information includes: text description information and product images of the product to be shipped; Determining the similarity between the goods to be shipped and similar goods in the warehouse according to the goods information includes: Screening based on a pre-built text-image similarity result library to obtain a third similar product that matches the text description information and the product image and a corresponding text-image similarity result; Determine the maximum value among the text-image similarity results as the similarity of the goods to be shipped; Among them, the text-image similarity result library is generated by calculating the similarity of splicing vector groups of all goods in the warehouse in pairs, and the splicing vector groups are obtained by splicing text feature vectors corresponding to the text description information of the goods and image feature vectors corresponding to the goods images.

9. A device for checking and processing outbound delivery, characterized in that: include: An information acquisition module is used to obtain product information of products to be shipped out; A similarity determination module, used to determine the similarity between the goods to be shipped and similar goods in the warehouse according to the goods information; A risk identification module, used for inputting the characteristic data of the goods to be shipped into a pre-trained abnormal risk identification model, and outputting a risk identification result of the goods to be shipped; wherein the characteristic data includes the similarity; The outbound check strategy generation module is used to generate differentiated target outbound check processing strategies according to the risk identification results.

10. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory, used to store computer programs; A processor, for implementing the method according to any one of claims 1 to 8 when executing a program stored in a memory.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images