Article delivery method and article delivery system

Abstract

The invention provides an article delivery method and an article delivery system, and solves or improves the technical problems that in the prior art, manual study and judgment of a delivery address consumes a large amount of time, the working efficiency of a deliverer is reduced, and the experience feeling of a user is reduced. According to the method, a user receiving intention is extracted according to text data of an order placing address of a user, meanwhile, an initial delivery address with address classification is determined according to text data segmentation and conversion of the order placing address of the user, the delivery address is determined according to the initial delivery address and the user receiving intention, and the delivery address is brief and clear. And the operation efficiency of delivery personnel is improved. Meanwhile, the type of the receiving intention of the user can assist the delivery personnel to execute proper delivery, so that the user experience is improved. In addition, according to the type of the receiving intention of the user, the receiving intention of the user can be matched, and complaints caused by improper delivery modes can be managed and controlled.

Description

Technical Field

[0001] The present invention relates to the field of logistics technology, and particularly to an article delivery method and an article delivery system. Background Art

[0002] In recent years, with the rapid development of e-commerce, the express delivery business has shown a high growth trend. Among them, receiving and sending express deliveries is the most labor-consuming work in the express delivery business. In order to improve the receiving and sending efficiency of deliverymen, the last-mile delivery has been proposed. In the last-mile delivery network, the emergence of pick-up points (such as Fengchao, Sudiyi, Cainiao Station, etc.) can provide users with package temporary storage services, which have been widely accepted and used in the last mile of logistics delivery. Therefore, currently, through the online order placement channel, the intention of the receiving method is judged based on the user's intention. For example, when guiding the user to place an order, the user can choose door-to-door delivery or delivery to the Fengchao cabinet. The user can write the delivery intention (such as pick-up point, door-to-door delivery, etc.) at the order placement address when placing the order.

[0003] However, in addition to the user-specified intention selection action that can capture the user's intention, a large number of users will describe the receiving intention in the receiving address. For example, "address + do not put in Fengchao", "address + put at the door of the house", "address + must be delivered to the door", etc. Such intentions mainly rely on the deliveryman to manually judge the delivery address during delivery, and then make a manual judgment and disposal. The method of manually judging the delivery address takes a long time and is not conducive to the deliveryman to make advance sorting and disposal in advance, such as batch-disposing the express deliveries to the Fengchao cabinets in the same unit in advance. In addition, manually judging the delivery address is likely to result in inappropriate delivery methods, reducing the user experience. Summary of the Invention

[0004] In view of this, the present invention provides an article delivery method and an article delivery system, which solve or improve the technical problems in the prior art that manually judging the delivery address takes a long time, reduces the work efficiency of the deliveryman, and reduces the user experience.

[0005] According to the first aspect of the present invention, the present invention provides an article delivery method, including: identifying the receiving intention of a user based on the text data of the user's order placement address to determine the category of the user's receiving intention; segmenting and converting the text data of the user's order placement address to determine an initial delivery address with a multi-level address classification; and determining a delivery address based on the category of the user's receiving intention and the initial delivery address, so that the deliveryman delivers the article according to the delivery address.

[0006] In an embodiment of the present invention, the recipient intention of the user is identified based on the text data of the user's order placement address to determine the recipient intention category of the user, including: constructing an intention recognition model, where the intention recognition model includes a BERT Chinese pre-training model, an encoder, a pooling layer, and a classifier; obtaining a training sample set, and annotating each training sample in the training sample set according to the recipient intention category table to determine the classification label of the training sample, where the classification label represents the recipient intention category corresponding to the text data of the order placement address in the training sample; training the intention recognition model based on the training sample set and the classification label of each training sample; and inputting the text data of the user's order placement address into the trained intention recognition model for recognition to determine the recipient intention category of the user.

[0007] In an embodiment of the present invention, based on the text data of the user's order placement address, the text data of the user's order placement address is segmented and converted to determine an initial delivery address with multi-level address classification, including: constructing an address entity recognition model, where the address entity recognition model includes a BERT Chinese pre-training model, an encoder, and a classifier; obtaining a training sample set, and determining the address label of the training sample according to the label definition table and the start character and non-start character corresponding to each label character, where the address label includes the label character, the start character corresponding to the label character, and the non-start character; training the address entity recognition model based on the training sample set and the address label of each training sample; and inputting the text data of the user's order placement address into the trained address entity recognition model for segmentation and conversion to determine an initial delivery address with multi-level address classification.

[0008] In an embodiment of the present invention, the address entity recognition model further includes a CRF model.

[0009] In an embodiment of the present invention, based on the text data of the user's order placement address, the text data of the user's order placement address is segmented and converted to determine an initial delivery address with multi-level address classification, including: when the recipient intention category is door-to-door or Fengchao or self-operated, based on the text data of the user's order placement address, the text data of the user's order placement address is segmented and converted to determine an initial delivery address with multi-level address classification.

[0010] In an embodiment of the present invention, before identifying the recipient intention of the user based on the text data of the user's order placement address to determine the recipient intention category of the user, the item delivery method further includes: obtaining the initial text data of the user's order placement address; and preprocessing the initial text data of the order placement address to determine the text data of the order placement address.

[0011] In an embodiment of the present invention, the preprocessing of the initial text data of the order placement address to determine the text data of the order placement address includes: performing text processing on the initial text data to obtain first text data; obtaining a keyword vocabulary, where the keyword vocabulary includes a plurality of keywords representing the recipient intention of the user; and querying in the first text data according to the keyword vocabulary. When text matching one of the keywords in the keyword vocabulary is found in the first text data, it is determined that the first text data is the text data of the order placement address.

[0012] In an embodiment of the present invention, the article delivery method further includes: when text matching one of the keywords in the keyword vocabulary is not found in the first text data, generating a call information, where the call information is used to indicate that the recipient intention category of the user is not willing to indicate, and prompting the delivery person to call the user to obtain the delivery address.

[0013] In an embodiment of the present invention, the performing text processing on the initial text data to obtain first text data includes: removing punctuation marks and illegal characters from the initial text data of the order placement address; removing words that do not affect the semantics from the initial text data of the order placement address; and trimming the initial text data of the order placement address in a backward manner to determine the first text data of the order placement address.

[0014] In an embodiment of the present invention, before performing text processing on the initial text data to obtain first text data, the preprocessing of the initial text data of the order placement address to determine the text data of the order placement address further includes: identifying the text length of the initial text data; when the text length of the initial text data is greater than or equal to a preset text length, performing text processing on the initial text data to obtain first text data.

[0015] In an embodiment of the present invention, the article delivery method further includes: when the text length of the initial text data is less than or equal to the preset text length, determining the initial text data as abnormal text, and generating a call information, where the call information is used to indicate that the recipient intention category of the user is not willing to indicate, and prompting the delivery person to call the user to obtain the delivery address.

[0016] In an embodiment of the present invention, after determining the delivery address based on the recipient intention category of the user and the initial delivery address, the article delivery method further includes: performing post-processing on the delivery address to determine a delivery address summary of the user, so that the delivery person delivers the article according to the delivery address summary.

[0017] In an embodiment of the present invention, the post-processing of the delivery address data to determine the delivery address summary of the user includes: extracting the last-level address in the delivery address; when a first preset tag character is recognized from the last-level address, determining the last-level address and the text indicating the user's receiving intention category in the delivery address as the delivery address summary; when the first preset tag character is not recognized from the last-level address, extracting, level by level in the delivery address, the first address at the level immediately before the last-level address until a second preset tag character corresponding to the first address is recognized from the first address, and extracting the first address, the address after the first address, and the text indicating the user's receiving intention category in the delivery address as the delivery address summary.

[0018] In an embodiment of the present invention, the receiving intention categories include door-to-door, Fengchao, self-operated, and no intention indicated.

[0019] As a second aspect of the present invention, the present invention provides an article delivery system, including: a receiving intention recognition model for recognizing the receiving intention of a user based on the text data of the user's order address to determine the receiving intention category of the user; an address entity recognition model for segmenting and converting the text data of the user's order address to determine an initial delivery address with multi-level address classification; and a fusion module for determining a delivery address based on the receiving intention category of the user and the initial delivery address, so that a delivery person can deliver an article according to the delivery address.

[0020] The article delivery method provided by the present invention extracts the user's receiving intention according to the text data of the user's order address. At the same time, it segments and converts the text data of the user's order address to determine an initial delivery address with address classification, and determines a delivery address based on the initial delivery address and the user's receiving intention. This delivery address is brief and clear, improving the operation efficiency of the delivery personnel. At the same time, the user's receiving intention category can assist the delivery personnel to perform appropriate deliveries, improving the user experience. In addition, according to the user's receiving intention category, the user's receiving intention can be matched, which is conducive to controlling complaints caused by improper delivery methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] By describing the embodiments of the present invention in more detail in conjunction with the drawings, the above and other objects, features, and advantages of the present invention will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings, the same reference numerals generally represent the same components or steps.

[0022] Figure 1 The following is a schematic flowchart of an article delivery method provided by an embodiment of the present invention.

[0023] Figure 2 The following is a schematic flowchart of an article delivery method provided by another embodiment of the present invention.

[0024] Figure 3 The following is a schematic diagram of the model structure of the bert_Chinese-base model in the present invention.

[0025] Figure 4 The following is a schematic diagram of the model structure of the bert_encoder layer model in the present invention.

[0026] Figure 5 The following is a schematic flowchart of an article delivery method provided by another embodiment of the present invention.

[0027] Figure 6 The following is a schematic flowchart of an article delivery method provided by another embodiment of the present invention.

[0028] Figure 7 The following is a schematic flowchart of an article delivery method provided by another embodiment of the present invention.

[0029] Figure 8 The following is a schematic flowchart of an article delivery method provided by another embodiment of the present invention.

[0030] Figure 9 The following is a schematic flowchart of an article delivery method provided by another embodiment of the present invention.

[0031] Figure 10 The following is a schematic flowchart of an article delivery method provided by another embodiment of the present invention.

[0032] Figure 11 The following is a schematic flowchart of an article delivery method provided by another embodiment of the present invention.

[0033] Figure 12 The following is a schematic diagram of the working principle of an article delivery system provided by an embodiment of the present invention.

[0034] Figure 13 The following is a schematic diagram of the working principle of an electronic device provided by an embodiment of the present invention. Detailed implementation manners

[0035] In the description of the present invention, the meaning of "a plurality" is at least two, such as two, three, etc., unless otherwise specifically defined. In the embodiments of the present invention, all directional indications (such as up, down, left, right, front, back, top, bottom...) are only used to explain the relative position relationship, movement conditions, etc. between components in a specific posture (as shown in the drawings). If the specific posture changes, the directional indications will also change accordingly. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units not listed, or optionally also includes other steps or units inherent to these processes, methods, products or devices.

[0036] In addition, the mention of "embodiment" in this article means that the specific features, structures or characteristics described in connection with the embodiment can be included in at least one embodiment of the present invention. The occurrence of this phrase in various positions in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0037] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0038] Exemplary Method

[0039] As a first aspect of the present invention, the present invention provides an article delivery method. Figure 1 The following shows a schematic flowchart of an article delivery method provided by an embodiment of the present invention, as Figure 1 shown, the article delivery method includes the following steps:

[0040] S1: Identify the recipient intention of the user based on the text data of the user's order address to determine the recipient intention category of the user.

[0041] Specifically, it is possible to manually classify possible recipient intentions to determine a recipient intention category table, which includes recipient intention categories and the recipient intentions included in the recipient intention categories. For example, when the recipient intention category is the Hive type, the recipient intentions included in the Hive type include: Hive, express cabinet, post station, supermarket. In addition, the recipient intention category table can also be manually updated to adapt to more recipient intentions.

[0042] Optionally, the recipient intention categories include door-to-door delivery, Fengchao (a type of smart locker), self-operated, and unstated intention.

[0043] Specifically, the text data of the user's order address refers to the address filled in by the user when placing an order in the system.

[0044] The text data of the order address includes, but is not limited to: the user's address, the specific delivery address of the user, the user's contact information, etc.

[0045] For example, the text data of the order address is: To be received by SF Express, Chengxiang Town, Baise City, Guangxi. Among them, Chengxiang Town, Baise City, Guangxi is the user's address, and SF Express is the user's specific delivery address. Another example: Fengchao, No. 180, Xinsha Road, Shajing Sub-district, Bao'an District, Shenzhen City, Guangdong Province. Among them, No. 180, Xinsha Road, Shajing Sub-district, Bao'an District, Shenzhen City, Guangdong Province is the user's address, and Fengchao is the user's specific delivery address.

[0046] It can be understood that the user's order address does not only include the user's detailed residential address in the traditional sense, but can be the order address as described above (including the address and the specific delivery address).

[0047] Specifically, the user's recipient intention category can be door-to-door delivery, that is, door-to-door distribution. At this time, the user's order address is the user's detailed residential address; Fengchao category, that is, delivering the user's items to Fengchao, express cabinets, supermarkets, stations, etc. for the user to pick up by themselves; self-operated category, that is, delivering the user's items to the designated SF business address, such as a certain SF business point, for the user to pick up by themselves.

[0048] Therefore, when the text data of the user's order address is To be received by SF Express, Chengxiang Town, Baise City, Guangxi, then the user's recipient intention category is self-operated. When the text data of the user's order address is Fengchao, No. 180, Xinsha Road, Shajing Sub-district, Bao'an District, Shenzhen City, Guangdong Province, the user's recipient intention category is Fengchao. When the text data of the user's order address is Room 301, Building 10, Youyi Community, Youyi Road, Hexi District, Tianjin for door-to-door delivery, then the user's recipient intention category is door-to-door delivery. When the text data of the user's order address is Room 301, Building 10, Youyi Community, Youyi Road, Hexi District, Tianjin, and the text data of the user's order address does not clearly state the recipient intention category, then the user's recipient intention category is unstated intention.

[0049] S1 can obtain the user's recipient intention category based on the text data of the order address, that is, determine whether the user is in the door-to-door delivery category, Fengchao category, self-operated category, or unstated intention.

[0050] S2: Split and convert the text data of the user's order address to determine the initial delivery address with multi-level address classification;

[0051] Specifically, the initial delivery address having multiple levels of address classification means that the user's order placement address is divided into multiple levels of addresses according to administrative regions, that is, the provinces, cities, districts, towns, streets, communities, etc. in the user's order placement address are identified. For example, the text data of the user's order placement address is: Shunfeng Express, Chengxiang Town, Baise City, Guangxi. Then, after splitting and converting the text data, the determined initial delivery address is: Shunfeng, Chengxiang Town, Baise City, Guangxi. And this initial delivery address has 3 levels of address classification, namely: Guangxi, Baise City, Chengxiang Town. Another example: The text data of the user's order placement address is Door-to-door service at Room 301, Building 10, Youyi Community, Youyi Road, Hexi District, Tianjin. The user's initial delivery address is then: Door-to-door service at Room 301, Building 10, Youyi Community, Youyi Road, Hexi District, Tianjin, which has 6 levels of address classification, namely: Tianjin City, Hexi District, Youyi Road, Youyi Community, Building 10, Room 301.

[0052] S3: Determine the delivery address based on the user's recipient intention category and the initial delivery address, so that the delivery person can deliver the item according to the delivery address.

[0053] After determining the initial delivery address of the user with multiple levels of address classification, the last-level valid address can be extracted from the initial delivery address;

[0054] After determining the delivery address, the system can display the delivery address on the terminal of the delivery person, so that the delivery person can view the delivery address on the terminal and deliver the item according to the delivery address.

[0055] Then, determine the user's recipient intention according to the type of the user's recipient intention category, and combine it with the last-level valid address of the user as the user's delivery address.

[0056] For example, the text data of the user's order placement address is: Shunfeng, Chengxiang Town, Baise City, Guangxi. S1 can identify that the user's recipient intention is the self-operated category, and S2 can determine that the initial delivery address with three levels of addresses is: Shunfeng, Chengxiang Town, Baise City, Guangxi. And this initial delivery address has 3 levels of addresses, namely: Guangxi, Baise City, Chengxiang Town. Then the user's delivery address is: Chengxiang Town (the last-level valid address in the initial delivery address) Shunfeng (recipient intention category). Then the system sends Chengxiang Town Shunfeng to the delivery personnel within the jurisdiction of Chengxiang Town, and then the delivery personnel deliver the item to the delivery address according to the delivery address for the user to pick up.

[0057] The article delivery method provided by the present invention extracts the user's receiving intention according to the text data of the user's order address. At the same time, it performs segmentation and conversion on the text data of the user's order address to determine the initial delivery address with address classification levels, and determines the delivery address based on the initial delivery address and the user's receiving intention. This delivery address is brief and clear, improving the operation efficiency of the delivery staff. At the same time, the category of the user's receiving intention can assist the delivery staff to perform appropriate deliveries, improving the user experience. In addition, according to the category of the user's receiving intention, the user's receiving intention can be matched, which is beneficial to controlling complaints caused by improper delivery methods.

[0058] In an embodiment of the present invention, as Figure 2 shown, the specific method for determining the category of the user's receiving intention is as follows, that is, S1 (identifying the user's receiving intention based on the text data of the user's order address to determine the category of the user's receiving intention) specifically includes the following steps:

[0059] S11: Construct an intention recognition model, which includes a BERT Chinese pre-training model, an encoder, a pooling layer, and a classifier;

[0060] The specific BERT Chinese pre-training model is the open-source bert_Chinese-base model. The token, embedding, and weight of this model are used to finetune the classification layer, as Figure 3 shown.

[0061] The encoder uses 12 bert_encoderlayers with the same model structure as the bert_Chinese-base model. Among them, the model structure of the bert_encoder layer is the classic bert-base model structure, and the model structure of the bert_encoderlayer is as Figure 4 shown.

[0062] S12: Obtain a training sample set, and label each training sample in the training sample set according to the receiving intention category table to determine the classification label of the training sample. The classification label represents the receiving intention category corresponding to the text data of the order address in the training sample;

[0063] The specific training sample is the text data of historical order addresses.

[0064] The receiving intention category table includes receiving intentions and the corresponding receiving intention categories. For example, when the receiving intention category is the Hive type, the receiving intentions corresponding to the Hive type include: Hive, express delivery, post station, supermarket.

[0065] Each training sample can be labeled according to the recipient intention category table to determine the classification label of the training sample. The classification label refers to the recipient intention category of the training sample. For example, when the recipient intention categories include: door-to-door, self-operated, Fengchao, and no recipient intention indicated. The classification label for the door-to-door category is lable1, the classification label for the Fengchao category is lable2, the classification label for the self-operated category is lable3, and the classification label for the no intention indicated category is lable0.

[0066] S13: Train the intention recognition model based on the training sample set and the classification label of each training sample;

[0067] Specifically, the training sample is input into the bert_Chinese-base model. The vocab.txt and embeddings in the bert_Chinese-base model convert the Chinese characters in the text data of the training sample into embeddings; the converted embeddings are input into 12 bert_encoder layers in the classifier to obtain an output, and the obtained output is input into the pooling layer and the classifier for pooling and softmax to obtain the classification result of the training sample.

[0068] According to the classification label of the training sample and the classification result of the training sample, the intention recognition model performs deep learning to obtain a trained intention recognition model.

[0069] Specifically, the deep learning process of the intention recognition model is divided into: forward inference, loss function calculation, gradient descent, and backpropagation to update weights. Among them, the loss function calculation uses the cross-entropy of the predicted classification and the actual classification.

[0070] S14: Input the text data of the user's order address into the trained intention recognition model for recognition to determine the recipient intention category of the user.

[0071] Input the text data of the user's order address into the intention recognition model trained in S13 for recognition to determine the recipient intention category of the user.

[0072] In an embodiment of the present invention, as Figure 5 shown, the specific method for determining the initial delivery address is as follows, that is, S2 (based on the text data of the user's order address, segment and convert the text data of the user's order address to determine the initial delivery address with multi-level address classification) specifically includes the following steps:

[0073] S21: Construct an address entity recognition model, and the address entity recognition model includes a BERT Chinese pre-trained model, an encoder, and a classifier;

[0074] Specifically, the BERT Chinese pre-trained model is the open-source bert_Chinese-base model. The model structures of the BERT Chinese pre-trained model in the BERT Chinese pre-trained model and the intent recognition model are the same. Both use the tokens, embeddings, and weights of this model and finetune the classification layer, as Figure 3 shown.

[0075] S22: Obtain a training sample set, and determine the address label of the training sample according to the label definition table and the start character and non-start character corresponding to each label character. The address label includes the label character, the start character corresponding to the label character, and the non-start character;

[0076] Specifically, the label definition table includes label characters and the corresponding text data. For example, the label definition table is shown in the following table.

[0077] Label Definition Table

[0078]

[0079] When annotating the training sample, the start character and non-start character of the text data corresponding to each label character are represented by I and B respectively. That is, I represents the start character of the text data corresponding to a label character, and B represents the non-start character of the text data corresponding to a label character. For example: The text data of the user's order address in the training sample is: Received by SF Express, Chengxiang Town, Chengxiang Town, Baise City, Guangxi. Then, after annotating this training sample, the obtained address label of this training sample is:

[0080]

[0081] S23: Train the address entity recognition model based on the training sample set and the address label of each training sample;

[0082] Specifically, the training sample is input into the bert_Chinese-base model, and the vocab.txt and embeddings in the bert_Chinese-base model are used to convert the Chinese characters in the text data of the training sample into embeddings; the converted embeddings are input into 12 bert_encoder layers in the classifier to obtain an output, and the obtained output is input into the classifier for softmax to obtain the classification result of the training sample.

[0083] According to the address label of this training sample and the classification result of the training sample, the intent recognition model is enabled to perform deep learning to obtain the trained intent recognition model.

[0084] Specifically, the deep learning process of the intention recognition model is divided into: forward inference, loss function calculation, gradient descent, and backpropagation to update weights.

[0085] S24: Input the text data of the user's order address into the trained address entity recognition model for segmentation and conversion to determine the initial delivery address with multi-level address classification.

[0086] Input the text data of the user's order address into the address entity recognition model trained in S23 for segmentation and conversion, and determine the initial delivery address with multi-level address classification. This initial delivery address is an address after processing (such as deduplication or removing some irrelevant information), and is not exactly the same as the text data of the user's order address.

[0087] For example, the text data of the user's order address is: Received by SF Express, Chengxiang Town, Chengxiang Town, Baise City, Guangxi. Since there are two Chengxiang Towns, after the entity recognition model recognition, the repeated address will be removed. Therefore, the initial delivery address after the address entity recognition model recognition is: Received by SF Express, Chengxiang Town, Baise City, Guangxi.

[0088] Optionally, the address entity recognition model further includes: a CRF model.

[0089] Specifically, the CRF model combines the characteristics of the maximum entropy model and the hidden Markov model, is an undirected graph model, and is commonly used for annotating or analyzing sequence data.

[0090] During the training process of the address entity recognition model, the training samples are input into the bert_Chinese-base model. The vocab.txt and embeddings in the bert_Chinese-base model convert the Chinese characters in the text data of the training samples into embeddings; the converted embeddings are input into 12 bert_encoder layers in the classifier to obtain an output, and the obtained output is input into the classifier for softmax to obtain an initial classification result, and the initial classification result is input into the CRF model for processing, and finally the classification result of the training samples is output.

[0091] Since in the process of the bert_Chinese-base model prediction, the classification of each character in the sequence result is independent of each other, and the results of the previous and subsequent characters are not considered, but in fact there are many unreasonable combinations. For example, only I-PROV can be before B-PROV, and only B-PROV or I-other entity or O can be after it. The introduction of the CRF model can learn the transition matrix between labels through data, so that the finally output result meets the constraints in the training data and gives an effective and reasonable prediction result.

[0092] Optionally, when the address entity recognition model further includes a CRF model, the loss function of the bert_Chinese-base model is the sum of the cross-entropy of the predicted classification and the actual classification and the loss function of the CRF model.

[0093] Specifically, the loss function of the bert_Chinese-base model is:

[0094]

[0095] In the formula:

[0096] S i = EmissionScore + TransitionScore

[0097] EmissionScore = x 0，START + x 1，B-Person + x 2，I-Person + x 3，O + x 4，B-Organization + x 5，O + x 6，END

[0098] TransitionScore = t START-＞B-Person + t B-Person-＞I-Person + t I-Person-＞O + t O-＞B-Orgnization + t B-Orgnization-＞O + t O-＞END

[0099] Among them, EmissionScore is the classification output result of the BERT model, TransitionScore is the transition score of the CRF model, and the t value in TransitionScore is the learnable parameter value of the CRF transition matrix.

[0100] In an embodiment of the present invention, as Figure 6 shown, the article delivery method further includes:

[0101] S20: Determine whether the recipient intention category is a door-to-door type, a Fengchao type, or a self-operated type;

[0102] When the judgment result of S20 is yes, that is, the recipient intention category is a door-to-door type, a Fengchao type, or a self-operated type, execute S2, that is, based on the text data of the user's order address, segment and convert the text data of the user's order address to determine the initial delivery address with multi-level address classification.

[0103] Only when the recipient intention category of the user recognized in S1 is a door-to-door type, a Fengchao type, or a self-operated type, will S2 be executed.

[0104] In an embodiment of the present invention, as Figure 7 shown, before S1 (identifying the recipient intention of the user based on the text data of the user's order address to determine the recipient intention category of the user), the article delivery method further includes:

[0105] S10: Obtain the initial text data of the user's order address;

[0106] Specifically, the initial text data is the initial text data of the user's order address directly obtained from the order system.

[0107] S11: Preprocess the initial text data of the order address to determine the text data of the order address.

[0108] Specifically, by preprocessing the initial text data, some useless information can be filtered out to obtain more accurate text data.

[0109] Optionally, as Figure 8 shown, the specific processing method for preprocessing the initial text data of the order address can be as follows: that is, S11 (preprocessing the initial text data of the order address to determine the text data of the order address) specifically includes the following steps:

[0110] S110: Perform text processing on the initial text data to obtain the first text data;

[0111] Specifically, performing text processing on the initial text data means cleaning the punctuation marks and words that do not affect the semantics in the initial text data, or cropping the initial text data to obtain more accurate text data.

[0112] Optionally, the specific processing method of the text processing includes the following:

[0113] (1) Text decryption. The user's order address information is top-secret customer information, usually stored in an encrypted form in the production system. In order for the subsequent algorithm inference system to understand the semantics, the encrypted text needs to be decrypted into plaintext first. Therefore, after the user is encrypted, the user's order address information is first decrypted.

[0114] (2) Remove the punctuation marks and illegal characters in the initial text data of the order address, so as to retain the valid text and improve the inference efficiency to a certain extent.

[0115] (3) Remove the words that do not affect the semantics in the initial text data of the order address;

[0116] The words that do not affect the semantics refer to the words that do not affect the address, such as words without specific meanings like "de" "le" "ne".

[0117] To prevent some non-standard waybill address texts from being very long and containing some complete sentences with these words, by removing such words that do not affect the semantics, the text data can be made more accurate. Additionally, when cropping the text later, it will not cut off the keywords due to the length.

[0118] (4) Crop the initial text data of the order placement address in a backward manner to determine the first text data of the order placement address.

[0119] Both the intention recognition model and the address entity recognition model adopted in S1 and S2 use fixed-length tokens. Therefore, for extremely long text data, cropping is required.

[0120] And the user's intention mainly appears in the latter part of the text. Therefore, the method of cropping from the back to the fixed length is adopted.

[0121] S111: Obtain the keyword list, where the keyword list includes multiple keywords representing the user's receiving intention.

[0122] Specifically, the keyword list includes multiple keywords representing the user's receiving intention. For example, self-operated, Fengchao, express cabinet, post station, supermarket, door-to-door, etc.

[0123] The keyword list is set according to experience and will be continuously updated according to new receiving intentions during subsequent application processes.

[0124] S112: Determine whether a text matching one of the keywords in the keyword list can be queried in the first text data.

[0125] That is, query according to the keyword list in the first text data to see if a text matching one of the keywords in the keyword list can be obtained in the first text data.

[0126] When the judgment result in S112 is yes, it means that a text matching one of the keywords in the keyword list can be queried in the first text data, indicating that the first text data includes the user's receiving intention. At this time, S113 can be executed, that is, the first text data can be determined as the text data of the user's order placement address, and the user's receiving intention category can be identified based on this text data.

[0127] When the judgment result in S112 is no, it means that a text matching one of the keywords in the keyword list cannot be queried in the first text data, indicating that the first text data does not include the user's receiving intention. Then S114 is executed, that is, a call information is generated, and the call information is used to indicate that the user's receiving intention category is not expressed, and prompt the dispatcher to call the user to obtain the delivery address.

[0128] S113: Determine that the first text data is the text data of the order placement address.

[0129] S4: Generate a call information, which is used to indicate that the user's recipient intention category is not indicated, so as to prompt the dispatcher to call the user to obtain the delivery address.

[0130] By filtering the text data of the order placement address, subsequent steps are only executed when the text data of the order placement address includes keywords indicating the mail intention, reducing the computational amount of the model and improving the efficiency.

[0131] In an embodiment of the present invention, as Figure 9 shown, before S110 (performing text processing on the initial text data to obtain the first text data), S11 (performing preprocessing on the initial text data of the order placement address to determine the text data of the order placement address), further includes the following steps:

[0132] S101: Identify the text length of the initial text data;

[0133] S102: Determine whether the text length of the initial text data is greater than a preset text length;

[0134] When the judgment result in S102 is yes, it indicates that the text length of the initial text data is greater than the preset text length, that is, execute S111, that is, perform text processing on the initial text data to obtain the first text data.

[0135] When the judgment result in S102 is no, it indicates that the text length of the initial text data is less than or equal to the preset text length, which means the text is too short, then directly classify it as the user not indicating the intention, and execute S4, that is, generate a call information, which is used to indicate that the user's recipient intention category is not indicated, and prompt the dispatcher to call the user to obtain the delivery address.

[0136] Filter out abnormal texts through the text length, mainly for cases where abnormal texts such as blank texts and too short texts can be judged by the text length, classify them as "not indicating the intention", and terminate the process in advance.

[0137] In an embodiment of the present invention, as Figure 10 shown, after S3 (determining the delivery address based on the user's recipient intention category and the initial delivery address), the article delivery method further includes the following steps:

[0138] S5: Perform post-processing on the delivery address to determine the delivery address summary of the user, so that the dispatcher delivers the article according to the delivery address summary.

[0139] Post-process the delivery address obtained in S3 to obtain a delivery address summary, so that the delivery person can intuitively and clearly know the user's delivery address.

[0140] That is, before the express delivery reaches the terminal delivery personnel, the routing of the province, city, and district has been completed. To improve the operation efficiency of the delivery staff, the post-processing can be reserved to the community (COMMU) and more detailed addresses. For example:

[0141] The delivery address obtained in S3 is: Fengchao Pickup Cabinet, Unit 2, Building 2, Zhenyuan Sunshine, between Fangzheng Street, Second Road and Third Road, Textile City Sub-district, Baqiao District, Xi'an City, Shaanxi Province. The delivery address summary is: Fengchao Pickup Cabinet, Unit 2, Building 2, Zhenyuan Sunshine. Another example: The delivery address obtained in S3 is: Express Cabinet on the left side of the shed in the student living area of the East Garden of Fudan University, No. 440 Guoding Road, Wujiaochang Sub-district, Yangpu District, Shanghai. Then the delivery address summary is: Express Cabinet on the left side of the shed in the student living area of the East Garden of Fudan University. Another example: The delivery address obtained in S3 is: SF Express_351P, Room 1013, Building 1, Evergrande Mingdu, No. 161 Jiefang North Road, Xinghualing District, Taiyuan City, Shanxi Province. Then the delivery address is: Room 1013, Building 1, Evergrande Mingdu, SF Express 351P. It should be noted that 351P is the network point code of the SF network point, so the network point code 351P can be retained or not.

[0142] Optionally, as Figure 11 shown, the specific post-processing method, that is, S5 (data post-processing of the delivery address) specifically includes the following steps:

[0143] S51: Extract the last-level address in the delivery address;

[0144] Specifically, the last-level address refers to the smallest address area in the delivery address. For example, the delivery address obtained in S3 is: Express Cabinet on the left side of the shed in the student living area of the East Garden of Fudan University, No. 440 Guoding Road, Wujiaochang Sub-district, Yangpu District, Shanghai. Then the last-level address in this delivery address is: On the left side of the shed in the student living area of the East Garden of Fudan University.

[0145] S52: Determine whether the first preset label character is recognized from the last-level address;

[0146] The first preset label character refers to the label character corresponding to the address that can represent the smallest address level in theory. For example, industrial park, community, apartment, village, neighborhood committee, supermarket, building, campus, phase, lane, alley, building, unit, etc.

[0147] For example, the first preset label character can be COMMU as shown in the above label definition table, representing village / community / industrial park / community name / apartment / neighborhood committee...

[0148] When the judgment result in S52 is yes, it indicates that the first preset label character can be recognized in the delivery address, that is, the smallest address area in the delivery address can be recognized, and then this smallest address area can be used as the delivery address summary. That is, execute S53.

[0149] S53: Determine the last-level address and the text representing the user's receiving intention category in the delivery address as the delivery address summary.

[0150] For example, when the delivery address is: Room 301, Building 10, Youyi Community, Youyi Road, Hexi District, Tianjin City for door-to-door delivery, the last-level address is: Room 301, Building 10, Youyi Community. Then there is a keyword "Community" representing the community in the last-level address, and the delivery address summary of this user is: Room 301, Building 10, Youyi Community for door-to-door delivery.

[0151] When the judgment result in S52 is no, it indicates that the preset keyword is not recognized in the delivery address, that is, the smallest address area cannot be recognized based on the delivery address. For example, the delivery address obtained in S3 is: Shunfeng Express Network, Huancheng Road Junction, Baihe Town, Hengzhou City, Nanning City, Guangxi Zhuang Autonomous Region. The last-level address is: Huancheng Road Junction (label character is P-DES). Huancheng Road Junction is a location word, and the first preset label character (such as COMMU) is not recognized in this last-level address, and the specific location cannot be recognized. At this time, execute S53:

[0152] S54: Extract the first address at the level before the last-level address in the delivery address;

[0153] It should be noted that the first address is one of the levels in the delivery address. For the sake of distinction, it is the first address.

[0154] For example: The delivery address obtained in S3 is: Shunfeng Express Network, Huancheng Road Junction, Baihe Town, Hengzhou City, Nanning City, Guangxi Zhuang Autonomous Region. The last-level address is: Huancheng Road Junction. The first address before "Huancheng Road Junction" is: Baihe Town.

[0155] S55: Judge whether the second preset label character can be recognized in the first address

[0156] The address level represented by the second preset label character is higher than the address level represented by the first preset label character. For example, if the first preset label character COMMU represents village / community / industrial park / residential area / apartment / residents' committee, then the second preset label character is TOWN (representing town);

[0157] When the second preset label character is recognized in the first address, take the first address, the address after the first address, and the text representing the user's receiving intention category in the delivery address as the delivery address summary, that is, execute S56.

[0158] S56: Use the first address, the address after the first address, and the text indicating the user's receiving intention category in the delivery address as the delivery address summary.

[0159] For example, the delivery address obtained in S3 is: Shunfeng Express Outlet, Huancheng Road, Baihe Town, Hengzhou City, Nanning City, Guangxi Zhuang Autonomous Region. The last-level address is: Huancheng Road, and the first address before "Huancheng Road" is: Baihe Town. The first address "Baihe Town" includes the keyword indicating "town". Therefore, the delivery address summary is: Baihe Town (first address) Huancheng Road (last-level address) Shunfeng Express Outlet (user's receiving intention category).

[0160] When the second preset tag character cannot be recognized in the first address, continue to fill in the previous-level address one level forward, and extract the previous-level address again before the first address (which can be the first address), until the preset keyword can be recognized in the extracted first address.

[0161] As the second aspect of the present invention, the present invention also provides an item delivery system, as Figure 12 shown. The item delivery system 200 includes:

[0162] A receiving intention recognition model 210, configured to recognize the user's receiving intention based on the text data of the user's order address to determine the user's receiving intention category;

[0163] Specifically, the receiving intention recognition model 1 is used to execute S1 in the above-mentioned item delivery method (recognize the user's receiving intention based on the text data of the user's order address to determine the user's receiving intention category).

[0164] Optionally, the intention recognition model includes a BERT Chinese pre-training model, an encoder, a pooling layer, and a classifier;

[0165] The BERT Chinese pre-training model is the open-source bert_Chinese-base model. Use the token, embedding, and weight of this model to finetune the classification layer, as Figure 3 shown.

[0166] The encoder uses 12 bert_encoderlayers with the same model structure as the bert_Chinese-base model. Among them, the model structure of the bert_encoder layer is the classic bert-base model structure. The model structure of the bert_encoderlayer is as Figure 4 shown.

[0167] An address entity recognition model 220, which is used to segment and transform the text data of the user's order address based on the text data of the user's order address, so as to determine an initial delivery address with a multi-level address classification;

[0168] Specifically, the address entity recognition model 2 is used to execute S2 in the above-mentioned item delivery method (segment and transform the text data of the user's order address based on the text data of the user's order address, so as to determine an initial delivery address with a multi-level address classification).

[0169] Optionally, the address entity recognition model includes a BERT Chinese pre-trained model, an encoder, and a classifier;

[0170] The BERT Chinese pre-trained model is the open-source bert_Chinese-base model. The model structure of the BERT Chinese pre-trained model in the address entity recognition model is the same as that of the BERT Chinese pre-trained model in the intent recognition model. Both adopt the token, embedding, and weight of this model, and finetune the classification layer, as Figure 3 shown.

[0171] Optionally, the address entity recognition model further includes: a CRF model. The introduction of the CRF model can learn the transition matrix between labels through data, so that the final output result meets the constraints in the training data and gives an effective and reasonable prediction result.

[0172] A fusion module 230, which is used to determine a delivery address based on the user's recipient intent category and the initial delivery address, so that the delivery person can deliver the item to the delivery address according to the delivery address.

[0173] Specifically, the fusion module 3 is used to execute S3 in the above-mentioned item delivery method (determine a delivery address based on the user's recipient intent category and the initial delivery address, so that the delivery person can deliver the item to the delivery address according to the delivery address).

[0174] The item delivery system provided by the present invention extracts the user's recipient intent according to the text data of the user's order address. At the same time, it segments and transforms the text data of the user's order address to determine an initial delivery address with address classification, and determines the delivery address according to the initial delivery address and the user's recipient intent. This delivery address is brief and clear, improving the operation efficiency of the delivery personnel. At the same time, the user's recipient intent category can assist the delivery personnel to perform appropriate deliveries, improving the user experience. In addition, according to the user's recipient intent category, the user's recipient intent can be matched, which is beneficial to controlling complaints caused by improper delivery methods.

[0175] Next, refer to Figure 13Describe an electronic device according to an embodiment of the present invention.

[0176] Figure 13 The structural block diagram of an electronic device according to an embodiment of the present invention is illustrated.

[0177] As Figure 13 shown, the electronic device 100 includes one or more processors 110 and a memory 120.

[0178] The processor 110 may be a central processing unit (CPU) or other forms of processing units with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 100 to perform desired functions.

[0179] The memory 120 may include one or more computer program products, and the computer program products may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory, etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage media, and the processor 11 may run the program instructions to implement the article delivery methods of various embodiments of the present invention described above and / or other desired functions.

[0180] In one example, the electronic device 100 may further include: an input device 130 and an output device 140, and these components are interconnected through a bus system and / or other forms of connection mechanisms (not shown).

[0181] When the electronic device is a stand-alone device, the input device 130 may be a communication network connector for receiving the collected input signals from the first device and the second device.

[0182] In addition, the input device 130 may further include, for example, a keyboard, a mouse, etc.

[0183] The output device 140 may output various information to the outside, including the determined distance information, direction information, etc. The output device 140 may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc.

[0184] Of course, for simplicity, Figure 13 only some of the components related to the present invention in the electronic device 100 are shown, and components such as buses, input / output interfaces, etc. are omitted. In addition, according to specific application scenarios, the electronic device 100 may further include any other appropriate components.

[0185] The present invention provides a computer-readable storage medium storing a computer program. The computer program is used to execute the above-mentioned article delivery method and device. In addition, an embodiment of the present invention may also be a computer program product, which includes computer program information. When the computer program information is run by a processor, the processor is caused to execute the steps in the article delivery method according to various embodiments of the present invention described in this specification.

[0186] The computer program product can be written in any combination of one or more programming languages for executing the program code of the operations of the embodiments of the present invention. The programming languages include object-oriented programming languages such as Java, C++, etc., and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, executed as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0187] In addition, an embodiment of the present invention may also be a computer-readable storage medium storing computer program information. When the computer program information is run by a processor, the processor is caused to execute the steps in the article delivery method according to various embodiments of the present invention described in this specification.

[0188] The computer-readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can, for example, include but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0189] The basic principles of the present invention have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, advantages, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, advantages, effects, etc. are essential for each embodiment of the present invention. In addition, the above-disclosed specific details are only for the purposes of illustration and easy understanding, rather than limitations. The above details do not limit the present invention to necessarily adopt the above specific details to implement.

[0190] The block diagrams of the devices, apparatuses, equipment, and systems involved in the present invention are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "comprising," "including," "having," etc. are open-ended terms meaning "including but not limited to" and can be used interchangeably with each other. The word "or" and "and" used herein refer to the phrase "and / or" and can be used interchangeably with it, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to" and can be used interchangeably with it.

[0191] It should also be noted that in the devices, equipment, and methods of the present invention, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent solutions of the present invention.

Claims

1. An article delivery method, characterized in that, Including: Identifying the recipient intention of the user based on the text data of the user's order address to determine the recipient intention category of the user; Segmenting and converting the text data of the user's order address to determine an initial delivery address with multi-level address classification; And Determining a delivery address based on the recipient intention category of the user and the initial delivery address, so that the delivery person delivers the item according to the delivery address.

2. The article delivery method according to claim 1, wherein Identifying the recipient intention of the user based on the text data of the user's order address to determine the recipient intention category of the user, including: Constructing an intention recognition model, which includes a BERT Chinese pre-trained model, an encoder, a pooling layer, and a classifier; Obtaining a training sample set, and annotating each training sample in the training sample set according to the recipient intention category table to determine the classification label of the training sample, where the classification label represents the recipient intention category corresponding to the text data of the order address in the training sample; Training the intention recognition model based on the training sample set and the classification label of each training sample; Inputting the text data of the user's order address into the trained intention recognition model for recognition to determine the recipient intention category of the user.

3. The article delivery method according to claim 1, wherein Based on the text data of the user's order address, segmenting and converting the text data of the user's order address to determine an initial delivery address with multi-level address classification, including: Constructing an address entity recognition model, which includes a BERT Chinese pre-trained model, an encoder, and a classifier; Obtaining a training sample set, and determining the address label of the training sample according to the label definition table and the start character and non-start character corresponding to each label character, where the address label includes the label character, the start character corresponding to the label character, and the non-start character; Training the address entity recognition model based on the training sample set and the address label of each training sample; Inputting the text data of the user's order address into the trained address entity recognition model for segmentation and conversion to determine an initial delivery address with multi-level address classification.

4. The article delivery method according to claim 3, wherein The address entity recognition model further includes a CRF model.

5. The article delivery method according to claim 1, wherein Based on the text data of the user's order address, segmenting and converting the text data of the user's order address to determine an initial delivery address with multi-level address classification, including: When the recipient intention category is door-to-door or Fengchao or self-operated, segmenting and converting the text data of the user's order address based on the text data of the user's order address to determine an initial delivery address with multi-level address classification.

6. The article delivery method according to claim 1, characterized in that, Before identifying the recipient intention of the user based on the text data of the user's order address to determine the recipient intention category of the user, the item delivery method further includes: Obtaining the initial text data of the user's order address; Preprocessing the initial text data of the order address to determine the text data of the order address.

7. The article delivery method according to claim 6, wherein, Preprocessing the initial text data of the order address to determine the text data of the order address, including: Performing text processing on the initial text data to obtain first text data; Obtaining a keyword vocabulary, wherein the keyword vocabulary includes a plurality of keywords representing the recipient intention of the user; and Querying in the first text data according to the keyword vocabulary, and when text matching one keyword in the keyword vocabulary is found in the first text data, determining the first text data as the text data of the order address.

8. The article delivery method according to claim 7, characterized in that It further includes: When no text matching one keyword in the keyword vocabulary is found in the first text data, generating a call information, which is used to indicate that the recipient intention category of the user is not indicated, and prompting the dispatcher to call the user to obtain the delivery address.

9. The article delivery method according to claim 7, wherein The performing text processing on the initial text data to obtain first text data includes: Removing punctuation marks and illegal characters from the initial text data of the order address; Removing words that do not affect the semantics from the initial text data of the order address; Cropping the initial text data of the order address in a backward manner to determine the first text data of the order address.

10. The article delivery method according to claim 7, characterized in that, Before performing text processing on the initial text data to obtain first text data, the preprocessing the initial text data of the order address to determine the text data of the order address further includes: Identifying the text length of the initial text data; When the text length of the initial text data is greater than or equal to a preset text length, performing text processing on the initial text data to obtain first text data.

11. The article delivery method according to claim 10, characterized in that, It further includes: When the text length of the initial text data is less than or equal to the preset text length, determining the initial text data as abnormal text, and generating a call information, which is used to indicate that the recipient intention category of the user is not indicated, and prompting the dispatcher to call the user to obtain the delivery address.

12. The article delivery method according to claim 1, wherein After determining the delivery address based on the recipient intention category of the user and the initial delivery address, the article delivery method further includes: Performing post-processing on the delivery address to determine the delivery address summary of the user, so that the dispatcher delivers the article according to the delivery address summary.

13. The article delivery method according to claim 12, characterized in that, The performing post-processing on the delivery address to determine the delivery address summary of the user includes: Extracting the last-level address from the delivery address; When a first preset label character is identified from the last-level address, determining the last-level address and the text representing the recipient intention category in the delivery address as the delivery address summary; When the first preset tag character is not recognized from the last-level address, the first address at the level immediately preceding the last-level address is extracted step by step in the delivery address until the second preset tag character corresponding to the first address is recognized from the first address, and the first address, the address after the first address, and the text indicating the user's receiving intention category in the delivery address are extracted as the delivery address summary.

14. The article delivery method according to claim 1, characterized in that, The receiving intention categories include door-to-door, Fengchao, self-operated, and no intention indicated.

15. An article delivery system, characterized in that, It includes: A receiving intention recognition model for recognizing the user's receiving intention based on the text data of the user's order address to determine the user's receiving intention category; An address entity recognition model for segmenting and converting the text data of the user's order address to determine the initial delivery address with multi-level address classification; And A fusion module for determining the delivery address based on the user's receiving intention category and the initial delivery address, so that the delivery person delivers the item according to the delivery address.

Images