Parcel type identification method and device, electronic equipment, storage medium and computer program product

Abstract

The present disclosure relates to a package type identification method and device, electronic equipment, storage medium and computer program product, and relates to the technical field of logistics, which can effectively balance the identification accuracy and efficiency of the package type. The method comprises: determining a to-be-delivered resource object of a to-be-identified package type, and obtaining first resource object description information of the to-be-delivered resource object; determining a similar resource object related to the first resource object description information semantics of the to-be-delivered resource object from a plurality of historical resource objects with pre-determined package type labels; determining first package type identification information of the to-be-delivered resource object according to the package type label of the similar resource object; and determining a package type identification result of the to-be-delivered resource object according to the first package type identification information.

Description

Technical Field

[0001] This disclosure relates to the field of logistics technology, and in particular to a method, apparatus, electronic device, storage medium, and computer program product for identifying package types. Background Technology

[0002] With the development of logistics technology, in order to improve logistics and distribution efficiency, the package type of goods can be identified before they are sent out, so that the appropriate delivery method can be used to send the logistics package.

[0003] In related technologies, when identifying the type of goods package, such as whether multiple goods need to be shipped together, the main methods used are manual judgment, judgment based on the weight and volume of the goods, or judgment through machine learning. However, these methods often fail to effectively balance the accuracy and efficiency of package type identification. Summary of the Invention

[0004] This disclosure provides a method, apparatus, electronic device, storage medium, and computer program product for identifying package types, to at least solve the problem in related technologies of the difficulty in effectively balancing the accuracy and efficiency of package type identification. The technical solution of this disclosure is as follows:

[0005] According to a first aspect of the present disclosure, a package type identification method is provided, comprising:

[0006] Identify the delivery resource object of the package type to be identified, and obtain the first resource object description information of the delivery resource object;

[0007] Among multiple historical resource objects with pre-determined package type labels, identify similar resource objects that are semantically related to the first resource object description information of the resource object to be delivered;

[0008] The first package type identification information of the resource object to be delivered is determined based on the package type label of the similar resource objects;

[0009] The package type identification result of the resource object to be delivered is determined based on the first package type identification information.

[0010] In one embodiment, determining the package type identification result of the resource object to be delivered based on the first package type identification information includes:

[0011] Based on the description information of the first resource object, obtain package type identification prompts;

[0012] The package type identification prompt is input into the language model, and the language model outputs second package type identification information indicating the package type of the resource object to be delivered based on the package type identification prompt.

[0013] Based on the first package type identification information and the second package type identification information, the package type identification result of the resource object to be delivered is determined.

[0014] In one embodiment, obtaining package type identification prompts based on the first resource object description information includes:

[0015] From the first resource object description information, obtain the resource object category to which the resource object to be delivered belongs and the characteristics of the resource object to be delivered;

[0016] Obtain an example of package type identification under the category of the resource object to which the resource object belongs, used to identify whether the resource object belongs to the target package type;

[0017] Based on the package type identification example and the characteristics of the resource object to be delivered, the information of the resource object to be filled in the pre-acquired prompt word construction template is filled in;

[0018] Based on the filling result of the template constructed from the prompt words, a package type identification prompt word is obtained to identify whether the resource object to be delivered belongs to the target package type.

[0019] In one embodiment, determining, among a plurality of historical resource objects with pre-determined package type labels, similar resource objects semantically related to the first resource object description information of the resource object to be delivered includes:

[0020] Obtain the first semantic feature corresponding to the description information of the first resource object;

[0021] Obtain the second semantic features corresponding to each of the multiple historical resource objects with pre-determined package type labels; the second semantic features are determined based on the second resource object description information of the historical resource objects;

[0022] Based on the feature similarity between the first semantic feature and each of the second semantic features, similar resource objects associated with the resource object to be delivered are determined from the plurality of historical resource objects.

[0023] In one embodiment, determining similar resource objects associated with the resource object to be delivered from the plurality of historical resource objects based on the feature similarity between the first semantic feature and each of the second semantic features includes:

[0024] Identify resource objects of the same type as the resource object to be delivered, and obtain third semantic features determined based on the third resource object description information of the resource objects of the same type;

[0025] Based on the third semantic feature and the first semantic feature, a semantic feature to be matched is determined. Based on the feature similarity between the semantic feature to be matched and each of the second semantic features, a similar resource object associated with the resource object to be delivered is determined from the plurality of historical resource objects.

[0026] In one embodiment, the package type label of the historical resource object is determined through the following steps:

[0027] Obtain the package type discrimination rule; the package type discrimination rule is obtained by statistically analyzing the resource object characteristics corresponding to each of multiple target historical resource objects, wherein the target historical resource objects are historical resource objects of the target package type;

[0028] Multiple historical resource objects are classified according to the package type discrimination rule, and package type labels are obtained for each of the multiple historical resource objects based on the classification results; the package type labels are used to indicate whether they belong to the target package type.

[0029] In one embodiment, after determining the package type identification result of the resource object to be delivered based on the first package type identification information, the method further includes:

[0030] If the package type identification result differs from the actual package type of the resource object to be delivered, the package type label of the similar resource object is updated according to the difference.

[0031] According to a second aspect of the present disclosure, a package type identification device is provided, comprising:

[0032] The information acquisition unit is configured to determine the type of package to be identified and the delivery resource object to be acquired, and to acquire the first resource object description information of the delivery resource object.

[0033] The retrieval unit is configured to perform the following operation: among multiple historical resource objects with pre-determined package type labels, determine similar resource objects that are semantically related to the first resource object description information of the resource object to be delivered;

[0034] The identification unit is configured to perform the task of determining first package type identification information of the resource object to be delivered based on the package type label of the similar resource object;

[0035] The result output unit is configured to perform the determination of the package type identification result of the resource object to be delivered based on the first package type identification information.

[0036] According to a third aspect of the present disclosure, an electronic device is provided, comprising:

[0037] processor;

[0038] Memory used to store the processor's executable instructions;

[0039] The processor is configured to execute the instructions to implement the package type identification method as described in any of the preceding claims.

[0040] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided that, when instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform the package type identification method as described in any of the preceding claims.

[0041] According to a fifth aspect of the present disclosure, a computer program product is provided, the computer program product including instructions that, when executed by a processor of an electronic device, enable the electronic device to perform the package type identification method as described in any of the preceding claims.

[0042] The technical solutions provided by the embodiments of this disclosure have at least the following beneficial effects:

[0043] By identifying similar resource objects semantically related to the first resource object description information of the current resource object from multiple historical resource objects with package tags, the package type identification can be performed quickly and accurately by using the package type tags of historical resource objects with similar attributes to the current resource object, even when specific parameters related to the package type identification of the resource object to be delivered are missing. This effectively balances the accuracy and efficiency of package type identification.

[0044] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0045] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0046] Figure 1 This is a flowchart illustrating a package type identification method according to an exemplary embodiment.

[0047] Figure 2 This is a flowchart illustrating a step for determining package type identification results based on prompt words, according to an exemplary embodiment.

[0048] Figure 3 This is a schematic diagram illustrating a package type identification prompt according to an exemplary embodiment.

[0049] Figure 4 This is a flowchart illustrating another package type identification method according to an exemplary embodiment.

[0050] Figure 5 This is a block diagram illustrating a package type identification device according to an exemplary embodiment.

[0051] Figure 6 This is a block diagram illustrating an electronic device according to an exemplary embodiment. Detailed Implementation

[0052] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0053] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0054] It should also be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0055] With the development of logistics technology, in order to improve logistics and distribution efficiency, the package type of goods can be identified before they are sent out, so that the appropriate delivery method can be used to send the logistics package.

[0056] In related technologies, when identifying the type of goods package, such as whether multiple goods need to be shipped together, the main methods used are manual judgment, judgment based on the weight and volume of the goods, or judgment through machine learning.

[0057] Among them, the manual judgment method depends on the staff's familiarity with the goods. Different staff members have different accuracy and efficiency in identifying the package type. For example, if the staff has a low level of understanding of the new goods, the accuracy of the final package type identification will also be very low.

[0058] Judging based on the actual parameters of specific products and preset thresholds also has certain limitations. Taking the combined delivery of multiple products as an example, the relevant technology will determine whether multiple products need to be combined into the same logistics package based on the comparison between the weight and volume of the products and preset thresholds. However, this method requires knowledge of the weight and volume information of the products and the corresponding business rules. However, in many practical scenarios, it is often impossible to obtain parameter information such as the weight and volume of the products in a timely manner. For example, in the case of consolidated shipping, it is necessary to identify the package type before warehousing, but at this time, the weight and volume data of the products are often not available.

[0059] Traditional machine learning methods require an algorithmic foundation and rely on continuous model debugging and training to improve classification accuracy, resulting in a certain learning curve and barrier to entry. Additionally, some methods attempt to directly determine whether goods belong to certain categories and ship them via specified package types, but this approach is overly complex and prone to misjudgment.

[0060] It is evident that the relevant technologies struggle to effectively balance the accuracy and efficiency of package type identification.

[0061] This disclosure provides a method, apparatus, electronic device, storage medium, and computer program product for identifying package types, which can effectively improve the accuracy and efficiency of package type identification.

[0062] In one embodiment, such as Figure 1 As shown, a package type identification method is provided. This embodiment illustrates the application of this method to a server. It is understood that this method can also be applied to a terminal, or to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0063] In step S101, the delivery resource object of the package type to be identified is determined, and the first resource object description information of the delivery resource object is obtained.

[0064] Among them, the resource objects to be delivered can refer to physical resources in the logistics business scenario that need to complete the logistics delivery task. For example, it can be goods ordered through an e-commerce platform, or logistics parcels waiting for subsequent logistics delivery at logistics distribution points or delivery outlets.

[0065] Parcel type can refer to the parcel classification category in the logistics scenario, which is used to distinguish one or more factors such as cargo attributes, delivery management standards, and warehousing rules. For example, it may include, but is not limited to, large items, fragile items, fresh products, bulk shipping materials, general daily necessities, hazardous chemicals, documents and letters.

[0066] Resource object description information can be associated with a resource object and used to characterize the attributes or characteristics of the resource object. The resource object description information can take the form of one or more of text, images, and videos. In some possible embodiments, the resource object description information may include one or more of the resource object's name, purpose, material, cargo parameters, delivery notes, and product category name. For ease of distinction, in this embodiment, the resource object description information of the resource object to be delivered is referred to as the first resource object description information.

[0067] In some optional embodiments, logistics order information can be obtained from one or more data sources such as the logistics business management backend, the business order processing platform, and the warehouse scheduling system, and the resource objects to be delivered can be determined based on the logistics order information. For example, after a user account submits a purchase order for a certain resource object, the resource object indicated by the purchase order can be used as the resource object to be delivered for the identified package type.

[0068] In some exemplary embodiments, the resource object link of the resource object to be delivered in the e-commerce platform can be determined, and the first resource object description information can be obtained based on the resource object introduction page corresponding to the resource object link.

[0069] In some embodiments, the first resource object description information may not include information for directly identifying the package type. For example, to determine whether the resource object to be delivered is a large item, the first resource object description information may not include the volume parameter of the resource object, nor may it include the weight parameter of the resource object. Similarly, to determine whether the resource object to be delivered is a perishable item, the first resource object description information may not include the effective usage date of the resource object.

[0070] In step S102, among multiple historical resource objects with pre-determined package type labels, similar resource objects semantically related to the first resource object description information of the resource object to be delivered are identified.

[0071] Among these, multiple historical resource objects with pre-determined package type labels can refer to resource objects whose package type labeling and classification have been completed by manual review or computer equipment. Package type labels can be package classification identifiers associated with historical resource objects; in some examples, a historical resource object can be associated with one or more package type labels. Similar resource objects can be understood as historical delivery objects that are semantically related to the currently pending delivery resource object.

[0072] In some embodiments, a candidate pool of historical resource objects can be pre-built, which may include multiple historical resource objects associated with package type tags. Then, after obtaining the first resource object description information, multiple historical resource objects can be matched with the first resource object description information to determine one or more historical resource objects semantically related to the first resource object description information of the resource object to be delivered. These one or more matched historical resource objects are then used as similar resource objects to the resource object to be delivered.

[0073] In some possible embodiments, resource object description information of each of multiple historical resource objects can be obtained. For ease of differentiation, the resource object description information corresponding to the historical resource objects can be referred to as the second resource object description information. Subsequently, similar resource objects can be determined from multiple historical resource objects based on the correlation or matching degree between the first resource object description information and multiple second resource object description information.

[0074] In step S103, the first package type identification information of the resource object to be delivered is determined based on the package type label of similar resource objects.

[0075] The first package type identification information is a preliminary classification result indicating the package type of the resource object to be delivered, generated after analysis based on the package type tags carried by one or more similar resource objects obtained through screening.

[0076] In some exemplary embodiments, after filtering out each similar resource object, the package type tags attached to each similar resource object can be analyzed and processed, such as by statistics and summarization, and the package type determination content corresponding to the current resource object to be delivered can be obtained based on the analysis and processing results.

[0077] In an exemplary embodiment, if there are multiple similar resource objects, the frequency of occurrence of various package type labels in all similar resource objects can be counted, and the package type label with the highest frequency can be determined as the first package type identification information corresponding to the resource object to be delivered.

[0078] In some alternative embodiments, the semantic similarity of similar resource objects can be combined to assign corresponding tag weights. The higher the similarity of similar resource objects, the greater the weight of the package type tag they carry. The comprehensive score of various package type tags is calculated by weighted summation, and the tag combination with the highest comprehensive score is selected as the first package type identification information.

[0079] In another embodiment, if the similar resource objects obtained by screening have multiple different types of package type labels, multiple sets of parallel first package type identification information can be generated simultaneously, retaining multiple preliminary judgment tendencies, reserving adjustment space for subsequent final result determination, and avoiding identification errors caused by a single judgment.

[0080] In step S104, the package type identification result of the resource object to be delivered is determined based on the first package type identification information.

[0081] Among them, the package type identification result is the package type of the resource object to be delivered that is ultimately identified by the computer equipment.

[0082] In some possible embodiments, after obtaining the first package type identification information, the first package type identification information can be used as the basis for judgment. Combined with one or more constraints such as logistics industry distribution control rules, cargo transportation control regulations, and actual sorting business specifications of the outlet, the first package type identification information can be verified, corrected, confirmed, or adjusted again, and finally the official package type identification result can be output.

[0083] In the above-described package type identification method, a delivery resource object of the package type to be identified can be determined, and the first resource object description information of the delivery resource object can be obtained. Then, among multiple historical resource objects with pre-determined package type tags, similar resource objects associated with the first resource object description information of the delivery resource object are identified. Next, the first package type identification information of the delivery resource object is determined based on the package type tags of the similar resource objects, and the package type identification result of the delivery resource object is determined based on the first package type identification information. In this embodiment, by identifying similar resource objects semantically related to the first resource object description information of the current delivery resource object from multiple historical resource objects with package tags, when performing package type identification, even in the absence of specific parameters related to the delivery resource object and package type identification, the package type of the current delivery package can be quickly and accurately identified by using the package type tags of historical resource objects with attributes similar to the current delivery resource object, effectively balancing the accuracy and efficiency of package type identification.

[0084] Furthermore, the embodiments disclosed herein are highly reusable and can be applied to the identification of various package types, such as fragile items, fresh produce, and package types for consolidated shipments, which helps to improve the accuracy of package type identification in various logistics scenarios.

[0085] In an exemplary embodiment, step S102, determining similar resource objects associated with the first resource object description information of the resource object to be delivered from among a plurality of historical resource objects with pre-determined package type labels, may include the following steps:

[0086] Obtain the first semantic feature corresponding to the description information of the first resource object; obtain the second semantic feature corresponding to each of the multiple historical resource objects with pre-determined package type labels; determine the second semantic feature based on the description information of the second resource object of the historical resource object; and determine the similar resource object associated with the resource object to be delivered from the multiple historical resource objects based on the feature similarity of the first semantic feature and each second semantic feature.

[0087] The second resource object description information is the resource object description information corresponding to the historical resource object.

[0088] Semantic features can be features that represent semantic content, obtained by parsing, extracting semantics, and vectorizing the descriptive information of resource objects. For ease of distinction, in this embodiment, the semantic features of the first resource object descriptive information are referred to as the first semantic features, and the semantic features of the second resource object descriptive information are referred to as the second semantic features.

[0089] In some embodiments, after obtaining the resource object description information (such as first resource object description information and / or second resource object description information), the resource object description information can be vectorized using a preset feature extraction model or a Large Language Model (LLM) to generate a semantic vector representation of the resource object as a semantic feature. For example, the text information such as the name, specifications, and description of the resource object can be input into an embedding model to generate the corresponding semantic vector representation.

[0090] In a specific implementation, a candidate pool of historical resource objects can be pre-built. The candidate pool of historical resource objects can include multiple historical resource objects associated with package type tags. When building the candidate pool of historical resource objects, for each historical resource object, the second resource object description information of the historical resource object can be obtained, and feature extraction can be performed on the second resource object description information to obtain the second semantic feature corresponding to the second resource object description information.

[0091] After obtaining the first semantic feature of the current resource object to be delivered, the feature similarity between the first semantic feature and each of the second semantic features can be determined. For example, a vector similarity calculation algorithm can be used to determine the similarity between the first semantic feature in vector form and each of the second semantic features in vector form, thus obtaining the feature similarity. The higher the feature similarity, the more similar the attributes or features of the current resource object to be delivered are to the corresponding historical resource objects. Furthermore, based on multiple feature similarities, similar resource objects semantically associated with the resource object to be delivered can be identified from multiple historical resource objects.

[0092] In some possible embodiments, when retrieving similar resource objects based on a first semantic feature and multiple second semantic features, relevant vector retrieval techniques can be used. For the historical resource objects obtained from the initial retrieval, a preset ranking model can be used to reorder each historical resource object initially recalled. Similar resource objects whose ranking order meets preset conditions are then used to form a final list of similar products, further improving the accuracy of the recall results. In semantic retrieval of similar products, combining the powerful semantic recognition capabilities of the language model with a ranking model helps to significantly improve the recognition accuracy of the target package type.

[0093] In some possible embodiments, when reordering the various historical resource objects recalled for the first time, the reordering can be performed by combining the consistency of characteristics between the historical resource objects and the resource objects to be delivered, the historical data of the historical resource objects, and relevant business rules.

[0094] In this embodiment, by extracting the semantic features of the resource object to be delivered and the historical resource object, it is possible to quickly and accurately search for historical resource objects that are more relevant / similar to the current resource object (i.e. the resource object to be delivered) during the package type identification process. The first package type identification information is determined based on the package type labels of the more relevant / similar historical resource objects, which helps to further improve the accuracy and reliability of the first package type identification information.

[0095] In one embodiment, determining similar resource objects associated with the resource object to be delivered from multiple historical resource objects based on the feature similarity of the first semantic feature and each of the second semantic features may include the following steps:

[0096] Identify resource objects of the same type as the resource object to be delivered, and obtain the third semantic features determined based on the third resource object description information of the resource objects of the same type; determine the semantic features to be matched based on the third semantic features and the first semantic features, and determine the similar resource objects associated with the resource object to be delivered from multiple historical resource objects according to the feature similarity between the semantic features to be matched and each second semantic feature.

[0097] Among them, resource objects of the same type are also called similar resource objects, which can be understood as other resource objects that have similar or identical characteristics to the resource object to be delivered. For ease of distinction, the resource object description information of resource objects of the same type is called third resource object description information, and the semantic features of the third resource object description information are called third semantic features.

[0098] In some embodiments, feature consistency judgment can be performed to identify resource objects of the same type by comparing the current resource object with various resource object features of other resource objects.

[0099] For example, the resource object characteristics of the current resource object to be delivered and multiple other resource objects can be obtained. In this example, the resource object characteristics can be Stock Keeping Unit (SKU) dimension characteristics. SKU dimension characteristics can be understood as standardized attribute information bound to each independent product inventory unit (i.e., resource object). For instance, SKU dimension characteristics can include one or more quantifiable and qualitative product attribute data such as the resource object's basic name, size specifications, material, production process, applicable scenarios, color and style, brand and model. Subsequently, the SKU dimension characteristics of the resource object to be delivered and the SKU dimension characteristics of multiple other resource objects can be compared for consistency. During the comparison process, one or more processing methods can be performed on each type of SKU dimension characteristic, such as item-by-item comparison, similarity verification, and overlap determination, ultimately distinguishing the resource objects of the same type as the current resource object to be delivered.

[0100] After identifying resource objects of the same type, we can obtain the third resource object description information of these objects and determine the third semantic feature corresponding to this description information. This third semantic feature, along with the first semantic feature, is then designated as the semantic features to be matched against each second semantic feature. Next, for each semantic feature to be matched, it is matched against each second semantic feature to obtain the feature similarity between the two. Historical resource objects associated with second semantic features whose feature similarity meets the similarity filtering criteria are then considered similar resource objects. Repeating this process for each semantic feature to be matched yields multiple similar resource objects.

[0101] In this embodiment, by identifying resource objects of the same type as the resource object to be delivered, determining the third semantic feature based on the third resource object description information of the resource object of the same type, and determining the semantic features to be matched based on the third semantic feature and the first semantic feature, it is possible to generalize the current resource object to be delivered to the same type, avoid the omission of subsequent similar resource object retrieval due to the current resource object description information being too simple, improve the comprehensiveness and sufficiency of similar resource object retrieval, and provide a foundation for subsequently determining reliable and accurate first package type identification information.

[0102] In one exemplary embodiment, such as Figure 2 As shown, in step S104, determining the package type identification result of the resource object to be delivered based on the first package type identification information may include the following steps:

[0103] In step S201, package type identification prompts are obtained based on the description information of the first resource object.

[0104] Among them, the package type recognition prompts can be text instructions provided to the language model, which instruct the language model to understand and perform the corresponding package type recognition task.

[0105] In some embodiments, semantic extraction can be performed on the description information of the first resource object or information fragments can be extracted from the description information of the first resource object to obtain text extraction results. Then, a template or prompt word generation rule can be constructed based on the text extraction results and pre-acquired prompt words to generate package type identification prompt words.

[0106] In step S202, the package type identification prompt is input into the language model, and the language model outputs second package type identification information indicating the package type of the resource object to be delivered based on the package type identification prompt.

[0107] The language model can refer to a large language model trained with a large number of standard samples. In some possible embodiments, the language model used in step S202 can be an open-source language model, such as a language model developed by other model developers. The language model can also be a model obtained by fine-tuning the pre-trained language model using standard data of package types.

[0108] After obtaining the package type identification prompt, the package type identification prompt can be input into the language model. In response to the input package type identification prompt, the language model can identify the package type of the resource to be delivered based on the information contained in the package type identification prompt. Based on the identification result, the language model outputs package type identification information indicating the package type of the resource to be delivered. For ease of distinction, the package type identification information output by the language model is called the second package type identification information.

[0109] In some embodiments, an agent can be built based on a language model. The agent identifies prompt words based on the package type and outputs second package type identification information indicating the package type of the resource object to be delivered. The agent can dynamically construct prompt words according to different categories to further improve the target package type identification model.

[0110] In step S203, the package type identification result of the resource object to be delivered is determined based on the first package type identification information and the second package type identification information.

[0111] In this step, the package type identification result of the resource object to be delivered can be determined by combining the previously obtained first package type identification information with the second package type identification information output by the language model.

[0112] In one possible embodiment, the first package type identification information and the second package type identification information can be compared for consistency. If they are consistent, the consistency result can be determined as the package type identification result; if the first package type identification information and the second package type identification information conflict, they can be marked for verification or further secondary verification can be performed in combination with other methods.

[0113] In this embodiment, by obtaining package type identification prompts based on the first resource object description information, and then using a language model to output second package type identification information based on these prompts, intelligent secondary identification of the package type of the resource object to be delivered can be performed using the language model, further improving the accuracy of the package type identification result. Furthermore, by fusing the first and second package type identification information to determine the package type identification result, it helps to compensate for potential blind spots in identification based on single semantic retrieval, thereby improving the accuracy of package type identification for the resource object to be delivered.

[0114] In an exemplary embodiment, in step S201, obtaining package type identification prompts based on the first resource object description information may include the following steps:

[0115] From the description information of the first resource object, obtain the resource object category to which the resource object to be delivered belongs and the characteristics of the resource object to be delivered; obtain the package type identification example under the resource object category to identify whether the resource object belongs to the target package type; according to the package type identification example and the characteristics of the resource object to be delivered, fill in the information of the resource object to be filled in the pre-obtained prompt word construction template; according to the filling result of the prompt word construction template, obtain the package type identification prompt word used to identify whether the resource object to be delivered belongs to the target package type.

[0116] In some possible embodiments, the first resource object description information may refer to one or more types of information pre-collected to characterize the basic attributes, appearance attributes, specification attributes, category attributes, and delivery-related attributes of the resource object to be delivered. The first resource object description information may originate from one or more legitimate information acquisition channels, such as merchant-entered information, product listing information, warehouse registration information, online order information, etc.

[0117] In this embodiment, the first resource object description information may include resource object characteristics and the resource object category to which it belongs. The resource object characteristics may include one or more of the following: product name, product image, product specifications, weight, volume, SKU characteristics, etc. The resource object category to which it belongs may also be called the product category. After obtaining the first resource object description information, the resource object category to which the resource object to be delivered and the resource object characteristics to be delivered can be extracted from it.

[0118] Then, relevant package type identification examples can be retrieved based on the resource object category to which the resource object to be delivered belongs. Specifically, package type identification examples can be pre-set for different resource object categories. These examples provide a process for identifying whether a resource object under a given resource object category belongs to the target package type. In some possible embodiments, the package type identification examples can be historical judgment examples of similar category goods that have been manually calibrated or historically verified by computer equipment and have reference validity. The package type identification examples can annotate the resource object characteristics corresponding to the historical resource objects and the final package type result. Furthermore, after obtaining the resource object category to which the resource object to be delivered belongs, package type identification examples under that resource object category can be obtained.

[0119] On the other hand, prompt word construction templates can be obtained. These templates are structured text templates with pre-reserved spaces for filling in fixed information. For example, the template may include preset judgment steps, context information, and may also include the category to which the corresponding resource object belongs, resource object characteristics, and judgment reference examples to be filled in. When prompt words need to be constructed, filling in the corresponding actual business data can generate standardized recognition prompt words. For example... Figure 3 As shown, an example of a prompt word construction template is provided.

[0120] Subsequently, based on the package type identification example and the characteristics of the resource object to be delivered, the information of the resource object to be filled in the pre-acquired prompt word construction template can be filled to obtain the filling result of the prompt word construction template. Then, the filling result can be used as the package type identification prompt word to identify whether the resource object to be delivered belongs to the target package type.

[0121] In this embodiment, by filling in information into a preset prompt word construction template to generate recognition prompt words, it is possible to quickly generate batches of package type recognition prompt words adapted to different resource objects, eliminating the need for manually writing judgment guidance text word by word, effectively reducing the preparation time for package type recognition. On the other hand, the constructed package type recognition prompt words integrate the characteristics of real resource objects to be delivered and related package type recognition examples of similar purposes, making the judgment reference more comprehensive and sufficient. It can not only adapt to the rapid classification of conventional standardized goods, but also enable the language model to combine the content of the prompt words to analyze the package type of special goods, effectively increasing the flexibility of recognition.

[0122] In one exemplary embodiment, the package type label of a historical resource object can be determined through the following steps:

[0123] Obtain the package type discrimination rule; the package type discrimination rule is obtained by statistically analyzing the resource object characteristics corresponding to multiple target historical resource objects, and the target historical resource objects are historical resource objects of the target package type; classify multiple historical resource objects according to the package type discrimination rule, and obtain the package type label of each of the multiple historical resource objects according to the classification results; the package type label is used to indicate whether it belongs to the target package type.

[0124] Among them, the package type discrimination rule refers to the criteria or conditions for determining whether a resource object belongs to the target package type, formed after data statistics and feature summarization. The target historical resource object can refer to each historical resource object that has been clearly marked as the target package type in advance.

[0125] In some embodiments, when obtaining package type discrimination rules, resource object features of target historical resource objects of the target package type can be extracted from historical data. These features may include, but are not limited to, one or more of the following: product name, description, specifications, weight, volume, SKU, etc. Then, the feature distribution of resource object features of multiple target historical resource objects can be statistically analyzed. Based on the statistical results, a confidence threshold can be set to determine the confidence level of a resource object belonging to the target package type when it possesses one or more specific resource object features, thus forming a package type discrimination rule. Furthermore, package type discrimination rules can also be obtained by combining actual business rules. For example, if the target package type is a large package, a package type discrimination rule can be formed by combining actual business rules: "When the weight exceeds X and the volume exceeds Y, it is determined to be a large package."

[0126] After obtaining the package type discrimination rules, multiple historical resource objects in the historical data can be classified into packages according to the package type discrimination rules. Based on the classification results, it can be determined whether each historical resource object belongs to the target package type, and the corresponding package type label can be associated with the historical resource object.

[0127] In this embodiment, by statistically analyzing the resource object characteristics of multiple target historical resource objects, a package type discrimination rule is formed. Based on the package type discrimination rule, multiple historical resource objects are classified and associated with package type tags. On the one hand, the rules can be constructed by making full use of the features of real historical samples, so that the overall target package type judgment logic fits the actual logistics business scenario and improves the overall accuracy of historical package data classification and tag labeling. On the other hand, by classifying and tagging multiple different historical resource objects, a sufficient retrieval basis can be provided for matching and determining the package type tags of subsequent resource objects to be delivered. This makes it easy for resource objects to be delivered to quickly identify whether they belong to the target package type by using the package type tags of historical resource objects when relevant data is lacking.

[0128] In an exemplary embodiment, after determining the package type identification result of the resource object to be delivered based on the first package type identification information, the following steps may also be included:

[0129] If the package type identification result differs from the actual package type of the resource object to be delivered, the package type label of the similar resource objects will be updated according to the difference.

[0130] The actual package type of the resource object to be delivered can be understood as the package type that the resource object to be delivered actually belongs to, after manual verification.

[0131] When the package type identification result differs from the actual package type of the resource object to be delivered, a label correction process can be triggered to promptly correct the product's label information. Specifically, the actual package type of the resource object to be delivered can be obtained. For example, the actual inbound data of the resource object to be delivered can be read, and the actual package type of the resource object to be delivered can be determined based on the package type identified in the actual inbound data.

[0132] If the package type identification result differs from the actual package type of the resource object to be delivered—for example, if the actual package type of the resource object to be delivered is different from the package type identification result—since the package type identification result of the resource object to be delivered is determined based on the package type tags of similar resource objects, the package type tags of the similar resource objects can be further obtained, and updated according to the difference between the package type identification result and the actual package type. For example, the actual package type can be used as the new package type tag for the similar resource objects to update the tags.

[0133] In this embodiment, on the one hand, by pre-configuring package type tags for similar resource objects, the current resource objects to be delivered can be quickly classified and predicted using package type tags; on the other hand, after the automatic identification has a judgment deviation, the tag content of similar resource objects can be corrected in reverse by the difference data formed by the actual package type, so as to continuously optimize the package type labeling database inside the system and gradually reduce the probability of subsequent similar resource objects of the same type having the same type identification error again.

[0134] In some possible embodiments, in the tag correction process, in addition to updating the package type tags of similar resource objects, the corrected tag information can also be fed back to the data analysis, semantic retrieval and similarity generalization modules to form a closed-loop optimization.

[0135] To enable those skilled in the art to better understand the above steps, specific examples are provided below to illustrate the embodiments of this disclosure. However, it should be understood that the embodiments of this disclosure are not limited thereto.

[0136] In this example, we will use a large package as the target package type.

[0137] In the logistics field, under the current consolidation business model, in-warehouse consolidation (i.e., combining multiple parcels into one parcel for delivery) can effectively solve the problem of second-stage transportation costs. However, whether parcels can be truly consolidated after arriving at the warehouse is affected by many factors, such as parcel interception, forced repackaging, failure to be shelved in time, inaccurate delivery time prediction, and excessively large product size. These factors lead to invalid consolidation and waiting times exceeding the allotted time, seriously affecting the overall delivery time of the parcels. Analysis shows that most of the factors causing invalid consolidation are due to excessively large product size. To address this, the parcel type identification method provided in this application embodiment can accurately identify large items.

[0138] First, multiple historical packages can be pre-tagged. For example... Figure 4 As shown, the characteristics of large items can be extracted from historical inbound data stored in a database (such as Hive). These characteristics include, but are not limited to, product name, description, specifications, weight, volume, and SKU. Then, data analysis and statistics are performed to form package type discrimination rules for large items. For example, based on historical inbound data, the feature distribution of large items can be statistically analyzed, and a confidence threshold can be set. This confidence threshold can be used to determine whether a product is large. Additionally, business rules (such as weight exceeding a certain threshold, volume exceeding a certain range, etc.) can be combined to extract a subset of large item characteristics. After obtaining the package type discrimination rules for large items, large package identification and tagging can be performed on various products, resulting in multiple historical resource objects associated with large item tags.

[0139] Subsequently, for orders submitted by user accounts, on the one hand, related products can be retrieved through a similar product search link. Specifically, for the products involved in the order, the first resource object description information of the product can be obtained. For example, the first resource object description information may include, but is not limited to, text information such as the product's name, specifications, and description. Then, the first resource object description information can be vectorized to generate a semantic vector representation of the product, i.e., the first semantic feature. Furthermore, the products involved in the order can be generalized to be similar. For example, by comparing the SKU dimension features of the products (such as name, specifications, materials, etc.), similar products can be identified, and the third resource object description information of the similar products can be obtained. After similar vectorization, the third semantic feature can also be generated. In some examples, the similarity generalization rules can be dynamically adjusted according to business needs and historical data to ensure the accuracy and applicability of the generalization results.

[0140] Subsequently, the first and third semantic features can be used as semantic features to be matched, and similarity calculations can be performed with the second semantic features of each pre-acquired historical resource object. Based on feature similarity, semantically similar products are recalled to form a candidate product set. For the candidate product set, multiple products in the candidate product set can be reordered by combining product feature consistency, historical data, and business rules to generate a final list of similar products. Then, based on the tags of each product in the list of similar products, it is determined whether the products involved in the order are large items, thus obtaining the first package type identification information.

[0141] On the other hand, based on the category of goods and resource characteristics involved in the order, a package type identification prompt can be dynamically constructed. For example, you can refer to... Figure 3 The template example shown constructs a prompt word, which guides a language model-based agent to perform feature recognition, determine whether the current item in the order is a large item, and output a large item score as the second package type identification information. The large item score reflects the probability that the current item in the order is a large item.

[0142] Then, based on the first and second package type identification information, a judgment is made to obtain the product size identification result, i.e., to determine whether the current product is a large item, which serves as the package type identification result. Subsequently, based on the product's shipping data or actual warehousing data, it can be determined whether the current product has been manually verified as a large item, thus obtaining the true package type. When there is a discrepancy between the true package type and the automated output package type identification result, a label correction process is triggered. For example, similar products with pre-labeled package type are corrected, and the corrected label information is fed back to the offline data link module, the similar product vector retrieval link module, the same product generalization module, and the AI ​​link module, forming a closed-loop optimization. For example, in the AI ​​link module, the agent can perform self-learning, dynamically adjusting the discrimination logic of the prompt word template and language model through historical data and feedback, thereby improving recognition accuracy.

[0143] In the above example, a product feature recognition model is constructed by combining historical data analysis, semantic retrieval, similarity generalization, intelligent agent automatic recognition, and feature correction. Among these, the identification of large packages based on multiple methods can more accurately and effectively identify the features of large items. Feature correction achieves a closed loop in the process, enabling dynamic optimization and closed-loop feedback of the system, which helps to achieve efficient recognition.

[0144] In some embodiments, if the products involved in the order have relevant parameters, the third package type identification information can be obtained by identifying whether the product belongs to a large package based on the package type discrimination rules obtained from the offline data link.

[0145] In one embodiment, a large-item feature library can also be constructed. This feature library can contain information about large items and their corresponding feature information, such as one or more feature items like product name, specifications, weight, volume, and SKU. The feature library content can be dynamically updated based on data analysis, semantic retrieval, and the results of similar product generalization. The feature library can provide an API interface, allowing other modules to call the data in the feature library for the identification and management of large items. By continuously maintaining and optimizing the feature library, the construction method of package type identification prompts or the content of extracted resource object description information can be optimized based on the feature items contained in the feature library and the specific resource object characteristics under the feature items.

[0146] Through the above processing, large packages in orders can be identified promptly and effectively in practice. The recall rate and identification accuracy of large items are extremely high, reducing invalid package consolidation caused by the introduction of large items, improving warehouse fulfillment efficiency, and reducing invalid shelving in the warehouse. This avoids additional costs such as package turnover time and warehouse manpower and handling expenses. In addition, because large items can be accurately identified, the delay in the underwriting of large items is avoided, effectively reducing order cancellations due to excessively long delivery times.

[0147] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0148] It is understood that the same / similar parts between the various embodiments of the methods described above in this specification can be referred to each other. Each embodiment focuses on the differences from other embodiments, and relevant parts can be referred to the description of other method embodiments.

[0149] Based on the same inventive concept, this disclosure also provides a package type identification device for implementing the package type identification method described above.

[0150] Figure 5 This is a block diagram illustrating a package type identification device according to an exemplary embodiment. (Refer to...) Figure 5 The device includes:

[0151] The information acquisition unit 501 is configured to determine the type of package to be identified and the delivery resource object to be acquired, and to acquire the first resource object description information of the delivery resource object.

[0152] The retrieval unit 502 is configured to perform the following operation: among a plurality of historical resource objects with pre-determined package type labels, determine similar resource objects that are semantically related to the first resource object description information of the resource object to be delivered;

[0153] The identification unit 503 is configured to perform the task of determining first package type identification information of the resource object to be delivered based on the package type label of the similar resource object;

[0154] The result output unit 504 is configured to perform the determination of the package type identification result of the resource object to be delivered based on the first package type identification information.

[0155] In one embodiment, the result output unit 504 is configured to perform:

[0156] Based on the description information of the first resource object, obtain package type identification prompts;

[0157] The package type identification prompt is input into the language model, and the language model outputs second package type identification information indicating the package type of the resource object to be delivered based on the package type identification prompt.

[0158] Based on the first package type identification information and the second package type identification information, the package type identification result of the resource object to be delivered is determined.

[0159] In one embodiment, the result output unit 504 is configured to perform:

[0160] From the first resource object description information, obtain the resource object category to which the resource object to be delivered belongs and the characteristics of the resource object to be delivered;

[0161] Obtain an example of package type identification under the category of the resource object to which the resource object belongs, used to identify whether the resource object belongs to the target package type;

[0162] Based on the package type identification example and the characteristics of the resource object to be delivered, the information of the resource object to be filled in the pre-acquired prompt word construction template is filled in;

[0163] Based on the filling result of the template constructed from the prompt words, a package type identification prompt word is obtained to identify whether the resource object to be delivered belongs to the target package type.

[0164] In one embodiment, the retrieval unit 502 is configured to perform:

[0165] Obtain the first semantic feature corresponding to the description information of the first resource object;

[0166] Obtain the second semantic features corresponding to each of the multiple historical resource objects with pre-determined package type labels; the second semantic features are determined based on the second resource object description information of the historical resource objects;

[0167] Based on the feature similarity between the first semantic feature and each of the second semantic features, similar resource objects associated with the resource object to be delivered are determined from the plurality of historical resource objects.

[0168] In one embodiment, the retrieval unit 502 is configured to perform:

[0169] Identify resource objects of the same type as the resource object to be delivered, and obtain third semantic features determined based on the third resource object description information of the resource objects of the same type;

[0170] Based on the third semantic feature and the first semantic feature, a semantic feature to be matched is determined. Based on the feature similarity between the semantic feature to be matched and each of the second semantic features, a similar resource object associated with the resource object to be delivered is determined from the plurality of historical resource objects.

[0171] In one embodiment, the apparatus further includes a tag acquisition unit configured to perform:

[0172] Obtain the package type discrimination rule; the package type discrimination rule is obtained by statistically analyzing the resource object characteristics corresponding to each of multiple target historical resource objects, wherein the target historical resource objects are historical resource objects of the target package type;

[0173] Multiple historical resource objects are classified according to the package type discrimination rule, and package type labels are obtained for each of the multiple historical resource objects based on the classification results; the package type labels are used to indicate whether they belong to the target package type.

[0174] In one embodiment, the device is further configured to perform:

[0175] If the package type identification result differs from the actual package type of the resource object to be delivered, the package type label of the similar resource object is updated according to the difference.

[0176] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0177] Each module in the aforementioned package type identification device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0178] Figure 6 This is a block diagram illustrating an electronic device 600 for implementing a package type identification method according to an exemplary embodiment. For example, the electronic device 600 can be a server. (Refer to...) Figure 6 The electronic device 600 includes a processing component 620, which further includes one or more processors, and memory resources represented by memory 622 for storing instructions, such as application programs, that can be executed by the processing component 620. The application programs stored in memory 622 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 620 is configured to execute instructions to perform the methods described above.

[0179] Electronic device 600 may further include: a power supply component 624 configured to perform power management of electronic device 600, a wired or wireless network interface 626 configured to connect electronic device 600 to a network, and an input / output (I / O) interface 628. Electronic device 600 may operate on an operating system stored in memory 622, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, or similar.

[0180] In an exemplary embodiment, a computer-readable storage medium including instructions is also provided, such as a memory 622 including instructions, which can be executed by a processor of an electronic device 600 to perform the above-described method. The storage medium may be a computer-readable storage medium, such as a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device.

[0181] In an exemplary embodiment, a computer program product is also provided, the computer program product including instructions that can be executed by a processor of an electronic device 600 to perform the above-described method.

[0182] It should be noted that the above-mentioned apparatus, electronic equipment, computer-readable storage medium, computer program product, etc., may also include other implementation methods according to the description of the method embodiments. For specific implementation methods, please refer to the description of the relevant method embodiments, which will not be elaborated here.

[0183] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

[0184] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for identifying package type, characterized in that, include: Identify the delivery resource object of the package type to be identified, and obtain the first resource object description information of the delivery resource object; Among multiple historical resource objects with pre-determined package type labels, identify similar resource objects that are semantically related to the first resource object description information of the resource object to be delivered; The first package type identification information of the resource object to be delivered is determined based on the package type label of the similar resource objects; The package type identification result of the resource object to be delivered is determined based on the first package type identification information.

2. The method according to claim 1, characterized in that, The step of determining the package type identification result of the resource object to be delivered based on the first package type identification information includes: Based on the description information of the first resource object, obtain package type identification prompts; The package type identification prompt is input into the language model, and the language model outputs second package type identification information indicating the package type of the resource object to be delivered based on the package type identification prompt. Based on the first package type identification information and the second package type identification information, the package type identification result of the resource object to be delivered is determined.

3. The method according to claim 2, characterized in that, The step of obtaining package type identification prompts based on the first resource object description information includes: From the first resource object description information, obtain the resource object category to which the resource object to be delivered belongs and the characteristics of the resource object to be delivered; Obtain an example of package type identification under the category of the resource object to which the resource object belongs, used to identify whether the resource object belongs to the target package type; Based on the package type identification example and the characteristics of the resource object to be delivered, the information of the resource object to be filled in the pre-acquired prompt word construction template is filled in; Based on the filling result of the template constructed from the prompt words, a package type identification prompt word is obtained to identify whether the resource object to be delivered belongs to the target package type.

4. The method according to claim 1, characterized in that, The step of determining, among multiple historical resource objects with pre-determined package type labels, similar resource objects semantically related to the first resource object description information of the resource object to be delivered includes: Obtain the first semantic feature corresponding to the description information of the first resource object; Obtain the second semantic features corresponding to each of the multiple historical resource objects with pre-determined package type labels; the second semantic features are determined based on the second resource object description information of the historical resource objects; Based on the feature similarity between the first semantic feature and each of the second semantic features, similar resource objects associated with the resource object to be delivered are determined from the plurality of historical resource objects.

5. The method according to claim 4, characterized in that, The step of determining similar resource objects associated with the resource object to be delivered from the plurality of historical resource objects based on the feature similarity between the first semantic feature and each of the second semantic features includes: Identify resource objects of the same type as the resource object to be delivered, and obtain third semantic features determined based on the third resource object description information of the resource objects of the same type; Based on the third semantic feature and the first semantic feature, a semantic feature to be matched is determined. Based on the feature similarity between the semantic feature to be matched and each of the second semantic features, a similar resource object associated with the resource object to be delivered is determined from the plurality of historical resource objects.

6. The method according to any one of claims 1 to 5, characterized in that, The package type label of the historical resource object is determined through the following steps: Obtain the package type discrimination rule; the package type discrimination rule is obtained by statistically analyzing the resource object characteristics corresponding to each of multiple target historical resource objects, wherein the target historical resource objects are historical resource objects of the target package type; Multiple historical resource objects are classified according to the package type discrimination rule, and package type labels are obtained for each of the multiple historical resource objects based on the classification results; the package type labels are used to indicate whether they belong to the target package type.

7. The method according to claim 6, characterized in that, After determining the package type identification result of the resource object to be delivered based on the first package type identification information, the method further includes: If the package type identification result differs from the actual package type of the resource object to be delivered, the package type label of the similar resource object is updated according to the difference.

8. A package type identification device, characterized in that, include: The information acquisition unit is configured to determine the type of package to be identified and the delivery resource object to be acquired, and to acquire the first resource object description information of the delivery resource object. The retrieval unit is configured to perform the following operation: among multiple historical resource objects with pre-determined package type labels, determine similar resource objects that are semantically related to the first resource object description information of the resource object to be delivered; The identification unit is configured to perform the task of determining first package type identification information of the resource object to be delivered based on the package type label of the similar resource object; The result output unit is configured to perform the determination of the package type identification result of the resource object to be delivered based on the first package type identification information.

9. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the package type identification method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is enabled to perform the package type identification method as described in any one of claims 1 to 7.

11. A computer program product, the computer program product comprising instructions, characterized in that, When the instruction is executed by the processor of the electronic device, the electronic device is able to perform the package type identification method as described in any one of claims 1 to 7.

Images