Logistics portrait model generation method and system

Abstract

The invention provides a logistics portrait model generation method and system. The method comprises the steps of obtaining a preset portrait construction scheme; acquiring logistics portrait data from the data source, and storing the logistics portrait data into a storage implementation module of a preset data storage engine; the storage realization module realizes storage of logistics portrait data based on an LSM tree; identifying portrait main body information corresponding to the portrait main body from a storage implementation module; extracting a feature matrix corresponding to the tag type related to the portrait main body under each specified time dimension from the portrait main body information; the specified time dimension is determined based on label types; inputting each feature matrix corresponding to the label type into a portrait label generation model, so that the portrait label generation model outputs a logistics portrait label corresponding to the label type according to a preset business rule and each feature matrix; and constructing a logistics portrait model corresponding to the portrait main body according to the plurality of logistics portrait labels corresponding to the portrait main body.

Description

Technical Field

[0001] The present invention relates to the field of logistics technology, and more specifically, to a logistics portrait model generation method and system. Background Art

[0002] With the rapid development of Internet technology, the logistics industry is also developing faster and faster. However, compared with the Internet industry, the logistics industry is relatively backward in technology development, especially in the construction of logistics image labels. It adopts a non-time-sequenced label construction solution based entirely on business rules. However, the current logistics industry has high requirements for the timeliness of logistics image labels. This solution ignores the sensitivity of logistics image labels in the time dimension, resulting in a decrease in the accuracy of most logistics image labels.

[0003] In addition, in terms of data storage, the existing technology uses HDFS as the storage engine, but the performance of this storage engine is constrained by the traditional data warehouse's scenario of more reads and fewer writes, resulting in low storage performance and unable to meet the continuous updates and application scenarios of the logistics industry. Summary of the invention

[0004] In view of this, the present invention provides a logistics portrait model generation method and system to solve the problem that the prior art adopts a non-time-sequential label construction solution based entirely on business rules in the construction of logistics portrait labels, ignoring the sensitivity of logistics portrait labels in the time dimension, resulting in reduced accuracy of most logistics portrait labels and low data storage performance, which cannot meet the continuous updates and application scenarios of the logistics industry.

[0005] The first aspect of the present invention discloses a method for generating a logistics portrait model, the method comprising:

[0006] Obtaining a preset portrait construction scheme; wherein the portrait construction scheme includes a portrait system, and the portrait system includes multiple portrait subject domains and multiple portrait subjects under each of the portrait subject domains;

[0007] Collecting logistics image data from a data source, and storing the logistics image data in a storage implementation module of a pre-set data storage engine; the storage implementation module implements the storage of logistics image data based on an LSM tree;

[0008] Identifying the portrait subject information corresponding to the portrait subject from the storage implementation module;

[0009] Extracting a feature matrix corresponding to a tag type related to the subject of the portrait at each specified time dimension from the subject information of the portrait; the specified time dimension is determined based on the tag type;

[0010] Input each of the feature matrices corresponding to the label type into a portrait label generation model, so that the portrait label generation model outputs a logistics portrait label corresponding to the label type according to the preset business rules and each of the feature matrices; wherein the portrait label generation model is obtained by training a multi-temporal memory model using historical portrait subject information;

[0011] According to the multiple logistics portrait tags corresponding to the portrait subject, a logistics portrait model corresponding to the portrait subject is constructed.

[0012] Optionally, the method further includes:

[0013] Defining metadata for the logistics portrait model of the portrait subject, and classifying each logistics portrait label in the portrait model after the metadata definition to obtain a corresponding target logistics portrait model;

[0014] According to the target portrait model corresponding to each of the portrait subjects, a logistics portrait label asset library is established, and a corresponding logistics labeling product platform is established based on the logistics labeling asset library, so that business personnel can generate corresponding logistics operation plans through the logistics labeling product platform.

[0015] Optionally, the data source includes a batch data source and a stream data source; the collecting of logistics portrait data from the data source and storing the logistics portrait data in a storage implementation module of a pre-set data storage engine includes:

[0016] Using a batch data collection system, collecting logistics image data from the batch data source through a target data port, and storing the logistics image data in a storage implementation module of a pre-set data storage engine;

[0017] The batch data collection system is used to collect logistics portrait data from the stream data source through a Web server, and the logistics portrait data is stored in a storage implementation module of a pre-set data storage engine.

[0018] Optionally, the batch data collection system includes a linker, a task configurator and a Hadoop cluster, and the batch data collection system is used to collect logistics portrait data from the batch data source through a corresponding data port, and the logistics portrait data is stored in a storage implementation module of a pre-set data storage engine, including:

[0019] Determining a target link configuration template from various pre-set link configuration templates through the linker; wherein the target link configuration template is a link configuration template corresponding to the Hadoop cluster;

[0020] Creating a link corresponding to the target data end according to the target link configuration template by the task configurator, and generating a corresponding task according to the link;

[0021] Synchronizing the task to the Hadoop cluster through the task translator;

[0022] The Hadoop cluster collects the logistics portrait data corresponding to the task from the streaming data source based on the target data end, and migrates the logistics portrait data to a pre-set data storage engine.

[0023] Optionally, the stream data collection system includes a stream receiver, a stream buffer and a stream storage device. The batch data collection system is used to collect logistics image data from the stream data source through a Web server, and the logistics image data is stored in a storage implementation module of a pre-set data storage engine, including:

[0024] Collecting corresponding logistics portrait data from the stream data source through the stream receiver based on the Web server;

[0025] The logistics portrait data is distributed to the stream memory through the stream buffer, and the logistics portrait data is migrated to a pre-set data storage engine through the stream memory.

[0026] Optionally, if the subject of the portrait is a customer, the tag type related to the subject of the portrait is a life cycle, and the specified time dimensions corresponding to the tag type include a monthly time dimension and a daily time dimension, a feature matrix corresponding to the tag type related to the subject of the portrait in each specified time dimension is extracted from the subject information of the portrait, including:

[0027] Counting multiple categories of the portrait subject in the portrait subject information within a preset time to obtain a fixed feature matrix corresponding to the label types related to the portrait subject in the monthly time dimension;

[0028] The number of single amounts of the portrait subject per day is counted from the portrait subject information to obtain a fixed feature matrix corresponding to the label types related to the portrait subject under the daily time dimension.

[0029] Optionally, the step of inputting each of the feature matrices corresponding to the label type into a portrait label generation model so that the portrait label generation model outputs a logistics portrait label corresponding to the label type according to preset business rules and each of the feature matrices, including:

[0030] Input each of the feature matrices corresponding to the label types into a portrait label generation model;

[0031] Through the portrait label generation model, each of the feature matrices is mapped to the corresponding sample label space according to the preset business rules, the features corresponding to each feature matrix are obtained, and each feature is fused into multiple features to obtain the logistics portrait label corresponding to the label type.

[0032] Optionally, the multi-temporal memory model includes an input layer, a recurrent neural network model LSTM, a convolutional layer and an output layer, and the multi-temporal memory model is trained using the historical portrait subject information to obtain a portrait label generation model, including:

[0033] For each of the portrait subjects, historical portrait subject information of the portrait subject can be obtained;

[0034] Extracting a historical feature matrix corresponding to the tag types related to the subject of the portrait at each specified time dimension from the subject information of the historical portrait;

[0035] Input each historical feature matrix under the label type into the recurrent neural network model LSTM through the input layer;

[0036] Through the recurrent neural network model LSTM, each historical feature matrix is ​​mapped to the corresponding sample label space according to the preset business rules to obtain the historical features corresponding to each historical feature matrix, and the historical features are fused through the convolution layer to obtain the historical label of the label type, and the parameters of the recurrent neural network model LSTM are adjusted with the historical label approaching the target logistics portrait label of the label type as the training goal until the recurrent neural network model LSTM converges to obtain the portrait label generation model.

[0037] A second aspect of the present invention discloses a logistics portrait model generation system, the system comprising:

[0038] A portrait construction scheme acquisition unit, used to acquire a preset portrait construction scheme; wherein the portrait construction scheme includes a portrait system, and the portrait system includes multiple portrait subject domains and multiple portrait subjects under each of the portrait subject domains;

[0039] A logistics portrait data collection unit is used to collect logistics portrait data from a data source and store the logistics portrait data in a storage implementation module of a pre-set data storage engine; the storage implementation module implements the storage of logistics portrait data based on an LSM tree;

[0040] A portrait subject information identification unit, used to identify the portrait subject information corresponding to the portrait subject from the storage implementation module;

[0041] A first extraction unit is used to extract a feature matrix corresponding to a tag type related to the portrait subject at each specified time dimension from the portrait subject information; the specified time dimension is determined based on the tag type;

[0042] The logistics portrait label generation unit is used to input each of the feature matrices corresponding to the label type into the portrait label generation model, so that the portrait label generation model outputs the logistics portrait label corresponding to the label type according to the preset business rules and each of the feature matrices; wherein the portrait label generation model is obtained by training the multi-temporal memory model based on the training unit using the historical portrait subject information;

[0043] The logistics portrait model construction unit is used to construct a logistics portrait model corresponding to the portrait subject according to multiple logistics portrait tags corresponding to the portrait subject.

[0044] Optionally, the system further includes:

[0045] A target logistics portrait model determination unit is used to perform metadata definition on the logistics portrait model of the portrait subject, and to classify each of the logistics portrait tags in the portrait model after the metadata definition to obtain a corresponding target logistics portrait model;

[0046] The logistics portrait label asset library establishment unit is used to establish a logistics portrait label asset library according to the target portrait model corresponding to each portrait subject, and establish a corresponding logistics labeling product platform based on the logistics label asset library, so that business personnel can generate corresponding logistics operation plans through the logistics labeling product platform.

[0047] The present invention provides a method and system for generating a logistics portrait model, which obtains a preset portrait construction scheme; wherein the portrait construction scheme includes a portrait system, and the portrait system includes multiple portrait subject domains and multiple portrait subjects under each of the portrait subject domains; logistics portrait data is collected from a data source, and the logistics portrait data is stored in a storage implementation module of a preset data storage engine; the storage implementation module implements the storage of logistics portrait data based on an LSM tree, which can improve storage performance to a certain extent, thereby meeting the continuous update and application scenarios of the logistics industry; the portrait subject information corresponding to the portrait subject is identified from the storage implementation module; the feature matrix corresponding to the label type related to the portrait subject under each specified time dimension is extracted from the portrait subject information; the specified time dimension is determined based on the label type; each of the feature matrices corresponding to the label type is input into a portrait label generation model, so that the portrait label generation model outputs the logistics portrait label corresponding to the label type according to the preset business rules and each of the feature matrices; wherein the portrait label generation model is obtained by training a multi-temporal memory model based on historical portrait subject information; according to the multiple logistics portrait labels corresponding to the portrait subject, a logistics portrait model corresponding to the portrait subject is constructed. The technical solution provided by the present invention fully considers the sensitivity of logistics image labels in the time dimension, thereby improving the accuracy of logistics image labels. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0049] Figure 1 A flow chart of a method for generating a logistics portrait model provided by an embodiment of the present invention;

[0050] Figure 2 A structural diagram of a portrait system provided by an embodiment of the present invention;

[0051] Figure 3 A structural diagram of a data storage engine provided by an embodiment of the present invention;

[0052] Figure 4 A structural diagram of a multi-temporal memory model provided by an embodiment of the present invention;

[0053] Figure 5 A structural schematic diagram of a logistics portrait model generation system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0056] It should be noted that the concepts such as "first" and "second" mentioned in the disclosure of the present invention are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0057] It should be noted that the modifications of "one" and "plurality" mentioned in the disclosure of the present invention are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0058] In order to better understand the technical terms involved in this application, the following is a corresponding explanation of the technical terms involved in this application:

[0059] Portrait: A labeling model abstracted from a group’s attributes, preferences, living habits, behaviors, and other information.

[0060] Hadoop: Hadoop is a distributed system infrastructure developed by the Apache Foundation. Users can develop distributed programs without understanding the underlying details of the distribution.

[0061] Hive: It is a data warehouse processing tool that encapsulates Hadoop at the bottom layer and uses the SQL-like hiveSQL language to implement data query.

[0062] Hdfs: Hadoop Distributed File System (HDFS) is a distributed file system designed to run on commodity hardware.

[0063] JDBC: JDBC stands for Java Database Connectivity, which is a standard Java application programming interface (JAVA API) used to connect the Java programming language and a wide range of databases.

[0064] LSM tree: Log-Structured Merge-Tree. In fact, it is not a specific data structure, but more of a data structure design idea. Most NoSQL database core ideas are based on LSM, but the specific implementations are different.

[0065] B-tree: B-tree is a multi-way balanced tree with ordered in-order traversal results. Unlike binary trees, nodes in B-trees can have multiple child nodes, while binary trees can only have two child nodes.

[0066] Bloom filter: It is a very long binary vector and a series of random mapping functions. Bloom filter can be used to retrieve whether an element is in a set. Its advantages are that the space efficiency and query time are far better than general algorithms, and its disadvantages are that there is a certain misrecognition rate and difficulty in deletion.

[0067] LSTM: Long short-term memory (LSTM) is a special RNN that is mainly used to solve the gradient vanishing and gradient exploding problems in the long sequence training process. In simple terms, compared with ordinary RNN, LSTM can perform better in longer sequences.

[0068] Convolutional layer: It is composed of several convolutional units. The parameters of each convolutional unit are optimized through the back propagation algorithm. The purpose of convolution operation is to extract different features of the input. The first convolutional layer may only extract some low-level features such as edges, lines, and corners. More layers of the network can iteratively extract more complex features from low-level features.

[0069] See also Figure 1 , shows a flow chart of a method for generating a logistics portrait model provided by an embodiment of the present invention, and the method for generating a logistics portrait model specifically includes the following steps:

[0070] S101: Obtain a preset portrait construction scheme; wherein the portrait construction scheme includes a portrait system, and the portrait system includes multiple portrait subject domains and multiple portrait subjects under each portrait subject domain.

[0071] In an embodiment of the present application, the architect may pre-specify a corresponding portrait construction scheme, wherein the portrait construction scheme includes a subject classification scheme, a portrait model design scheme, and a portrait tag management scheme.

[0072] It should be noted that the design process of the subject classification scheme can be as follows: by analyzing the current core business of the enterprise and the business that will be focused on in the next few years, and classifying it according to various logistics business scenarios and logistics industry characteristics, a corresponding portrait system is obtained, such as Figure 2 As shown, the portrait system includes multiple subject domains and multiple portrait subjects under each subject domain. For example, when the subject domain is a customer portrait subject domain, it includes customer subject B and customer subject C. Customer subject B can be a corporate customer, and customer subject C can be an individual customer.

[0073] The portrait tag management solution includes incremental review rules for logistics portrait tags, archiving rules for logistics portrait tags, project space management rules corresponding to logistics portrait tags, development account rules corresponding to logistics portrait tags, and usage rules for logistics portrait tags.

[0074] Among them, the incremental review rules for logistics portrait labels can be: the logistics portrait R&D department is responsible for the overall planning of the logistics portrait label construction work, and the incremental logistics vertical label requirements must undergo relevant reviews before they can be undertaken and subsequent development work can be carried out.

[0075] The archiving rules for logistics portrait labels can be: each department submits the logistics portrait labels to be archived to the portrait department, and integrates them into the logistics portrait label asset library according to a unified dimension.

[0076] The project space management rules corresponding to the logistics portrait tag can be: tag-related scheduling tasks are uniformly stored in the <Logistics Portrait> project space to facilitate unified management of code versions.

[0077] The development account rules corresponding to the logistics portrait tags can be: data development must use a logistics portrait-exclusive account, and sensitive logistics portrait tags must also be applied to the portrait account.

[0078] The usage rules of logistics portrait tags can be as follows: To use portrait assets, you need to apply to the portrait department via email. If you use it without authorization, you will have the right to revoke the relevant permissions and will not be responsible for the data of the application.

[0079] S102: Collect logistics portrait data from the data source, and store the logistics portrait data in a storage implementation module of a pre-set data storage engine; the storage implementation module implements the storage of logistics portrait data based on the LSM tree.

[0080] In the embodiment of the present application, the data source includes a batch data source and a stream data source. The data port corresponding to the batch data source includes a client, a front-end interface and an interface; the data port corresponding to the stream data source may be a Web server.

[0081] It should be noted that due to the wide range of offline data sources, compared with the existing single data end (client) architecture, this application involves three interaction modes: client, front-end page and interface, thereby increasing the versatility of offline data collection.

[0082] In the specific execution of step S102, for each data source, the data collection system corresponding to the data source can be used to collect corresponding logistics portrait data from the data source through the data port corresponding to the data source, and the collected logistics portrait is stored in the storage implementation module of the pre-set data storage engine. Among them, the storage implementation module implements the storage of logistics portrait data based on the LSM tree.

[0083] Optionally, in terms of existing data collection, the existing traditional data collection technology simply uses the sqoop framework to complete data collection, which cannot meet the business scenarios of TB-level ultra-large data volumes in the logistics industry, nor can it guarantee the accuracy and efficiency of data transmission; therefore, when the data source is a batch data source, this application can use a batch data collection system to collect corresponding logistics portrait data from the batch data source through the target data port. Among them, the target data port can be any one of the client, the front-end interface, and the interface.

[0084] When the data source is a streaming data source, a streaming data collection system can be used to collect corresponding logistics portrait data from the streaming data source through a Web server.

[0085] In an embodiment of the present application, a batch data collection system includes a linker, a task configurator, and a Hadoop cluster. The target link configuration template can be determined from various pre-set link configuration templates through the linker; the link corresponding to the target data end is created according to the target link configuration template by the task configurator, and the corresponding task is generated according to the link; the task is synchronized to the Hadoop cluster through the task translator; the logistics portrait data corresponding to the task is collected from the stream data source based on the target data end by the Hadoop cluster, and the logistics portrait data is migrated to the pre-set data storage engine. Among them, the target link configuration template is the link configuration template corresponding to the Hadoop cluster.

[0086] It should be noted that corresponding link configuration templates can be set in the linker for different storage media. The storage medium can be mysql, HDFS file system, Hadoop cluster, etc. For example, the link corresponding to the HDFS file system can be hdfs-configuration-template.

[0087] After detecting that the linker determines the target link configuration template, the task configurator can create a corresponding link according to the target link configuration template and inherit the task corresponding to the link. The task includes corresponding configuration information. For example, if the storage medium corresponding to the target link configuration template is mysql, the configuration information included in the task inherited from the link created according to the target link configuration template can be JDBC link configuration information.

[0088] In an embodiment of the present application, the stream data collection system includes a stream receiver, a stream buffer and a stream storage device, and the corresponding logistics portrait data can be collected from the stream data source based on the Web server through the stream receiver; the logistics portrait data can be distributed to the stream storage device through the stream buffer, and the logistics portrait data can be migrated to a pre-set data storage engine through the stream storage device.

[0089] It should be noted that a log encapsulator is designed in the stream receiver. The function of the log encapsulator is to encapsulate a specified number of data into the smallest storage unit in order to meet the performance requirements during the peak period of real-time data. For example, during the 618 and Double 11 periods, tens of millions of real-time data volumes may be generated per second. This design can greatly reduce the system load and improve the system throughput.

[0090] A memory area for caching data is designed in the flow cache, and the collected logistics portrait data can be stored in this memory area; the data in this area can be persisted to the local disk, which is not only fault-tolerant for fault recovery, but also can meet the business needs for historical real-time data.

[0091] Stream storage can store the logistics portrait data stored in the stream cache. After the logistics portrait data in the stream cache is stored in the stream storage, the logistics portrait data in the stream cache will be deleted. The stream storage defined in this application includes 5 storage medium configurations, including all scenarios using real-time data storage media in the logistics industry, as well as HDFS configuration, HBASE configuration, ES configuration, LOGGER configuration, and FileRoll configuration to support data migration of the stream storage to the downstream data storage module.

[0092] In the embodiment of the present application, a data storage engine is also pre-set, such as Figure 3 As shown; wherein, the data storage engine includes a storage implementation module, an update data module, a write data module, a read data module and a data partition module.

[0093] In some embodiments, the storage implementation module in the embodiments of the present application implements the storage of logistics image data based on LSM input. Among them, the LSM tree can improve the storage capacity of the storage implementation module to a certain extent by exchanging a small part of the read data performance for the write data performance.

[0094] It should be noted that, using the LSM tree, the logistics portrait data can be divided into a data row set and a metadata set; wherein the data row set stores the actual data in the logistics portrait data, and the metadata set is used to store the metadata in the logistics portrait data.

[0095] It should also be noted that the data row set includes cache data row set and disk data row set. The cache data row set can realize real-time data storage, and the disk data row set can realize persistent data storage. The disk data row set can be divided into basic data row set and change data row set. In the change data set, B-tree index can be used to record the change operation of the basic data in the actual data of each logistics portrait data. Among them, the change operation includes update and delete.

[0096] It should also be noted that although LSM trees can improve data writing performance, they reduce data reading performance. In order to improve data reading performance, two types of indexes can be created for the data in the basic data row set to improve data reading performance. The two types of indexes are primary key index and Bloom filter index. Primary key index: uses B-tree to determine the index to achieve efficient search of the primary key; Bloom filter index: uses the primary key to filter invalid data.

[0097] In an embodiment of the present application, the data partitioning module includes a range partitioning rule and a hash partitioning rule. Among them, the range partitioning rule is to divide according to the value range of the partition field, which is conducive to the sequential reading and writing of batch data; the hash partitioning rule can evenly distribute the partition field to each partition according to the hash value, which is conducive to the read and write speed of the distributed data. After the logistics portrait data is stored in the storage implementation module, the logistics portrait data in the storage implementation module can be further stored in the data partitioning module. During the storage process, it can be determined whether the logistics portrait data is a sequential read. If it is a sequential read, the logistics portrait data can be stored in the data partitioning module according to the range partitioning rule; if it is not a sequential read, the logistics portrait data can be evenly distributed to each partition of the data partitioning module according to the hash partitioning rule.

[0098] Correspondingly, when you want to read and process from the data partition module, you can determine whether the data you want to read is a sequential read. If it is a sequential read, you can read the corresponding data from the data partition module according to the range partition rule; if it is not a sequential read, you can use the hash partition rule to read the corresponding data from the partition of the data partition module.

[0099] In an embodiment of the present application, a data write module is used to detect a write request and, based on the primary key and data partitioning strategy, obtain the target partition in the storage implementation module indicated by the write request, and filter out invalid primary keys in the basic data row set of the target partition through the Bloom filter index, and then search the target partition for the primary key indicated by the write request through the primary key index. If so, an error is reported; if not, the data is written to the cache data row set.

[0100] In an embodiment of the present application, the data reading module is used to detect a read request and can obtain the target partition in the storage implementation module indicated in the read request according to the primary key and data partitioning strategy, and filter out invalid primary keys through the Bloom filter index in the basic data row set of the target partition, and then accurately search for the corresponding data through the primary key index, so as to execute the union command to summarize the found data and feed back the summarized data.

[0101] In an embodiment of the present application, the data change module is used to, when a data change request is detected, obtain the target partition in the storage implementation module indicated in the data change request according to the primary key and data partitioning strategy, and in the basic data row set of the target partition, filter the invalid primary key through the Bloom filter index, and then accurately search for data through the primary key index, and update the index change data of the found data according to the data row set.

[0102] S103: Identify the portrait subject information corresponding to the portrait subject from the storage implementation module.

[0103] In the specific process of executing step S103, after the logistics portrait data is stored in the storage implementation module, for each portrait subject, the portrait subject information corresponding to the portrait subject can be identified from the storage implementation module through the data reading module.

[0104] In some embodiments, if the portrait subject domain is the customer portrait subject domain, the corresponding portrait subject may be customer B, and the portrait subject information corresponding to customer B may include customer B's customer revenue information and customer order quantity information.

[0105] S104: extracting a feature matrix corresponding to a tag type related to the portrait subject at each specified time dimension from the portrait subject information; the specified time dimension is determined based on the tag type.

[0106] In the embodiment of the present application, for each portrait subject under each portrait theme domain, multiple tag types related to the portrait subject can be pre-set; for each tag type under the portrait subject, the factors affecting the tag type can be further analyzed, and a corresponding specified time dimension can be set for each factor. For example, the portrait theme domain is the customer portrait subject domain, the corresponding portrait subject can be customer B, and the tag type related to the portrait subject (customer B) can be life cycle. After analysis, the factors affecting the customer's life cycle may be customer income and customer units, and then a specified time dimension corresponding to customer income can be set, and a specified time dimension corresponding to customer unit volume can be set.

[0107] It should be noted that for customer B, if we only look at the number of orders, if the customer has orders but no revenue, then the customer may be labeled as a "lost customer". Therefore, the number of orders and revenue can be combined so that the logistics portrait label obtained is more accurate.

[0108] Most of the income of customer B is usually settled monthly, so the designated time dimension corresponding to customer B's customer income can be set to month; for customer B with frequent transactions, there may be orders every day, so customer orders are far more frequent than customer income. If month is used as the designated time dimension for customer orders, the following situations may occur: if the number of orders of a certain customer B on the 1st is 100, it can be considered that the label of this customer B is "active customer", however, if there are orders only on the 1st of this month, it can be considered that the label of this customer B is "inactive customer". In this way, the customer's label change cannot be perceived, so the designated time dimension corresponding to the customer's order volume can be set to day.

[0109] In the specific process of executing step S104, after obtaining the portrait subject information corresponding to each portrait subject, for each label type related to each portrait subject, the corresponding feature matrix under each specified time dimension corresponding to the label type can be extracted from the portrait subject information of the portrait subject.

[0110] As a preferred method of an embodiment of the present application, if the portrait subject is a customer, and the label type related to the portrait subject is a life cycle, statistics are performed on multiple categories of the portrait subject in the portrait subject information within a preset time to obtain a fixed feature matrix corresponding to the label types related to the portrait subject in the monthly time dimension; the number of single amounts of the portrait subject per day is counted from the portrait subject information to obtain a fixed feature matrix corresponding to the label types related to the portrait subject in the daily time dimension.

[0111] For example, statistics are collected on 100 categories of a customer in the past 12 months to obtain a 12*100 matrix; statistics are collected on the number of orders per day within a year, that is, 365 days, to obtain a 365*daily number of orders matrix.

[0112] S105: Input each feature matrix corresponding to the label type into the portrait label generation model, so that the portrait label generation model outputs the logistics portrait label corresponding to the label type according to the preset business rules and each feature matrix.

[0113] In the embodiment of the present application, a multi-time series memory model can be constructed, and the multi-time series memory model can be trained using the historical portrait subject information to obtain a portrait label generation model. The multi-time series memory model includes an input layer, a recurrent neural network model LSTM, a convolutional layer, and an output layer. Figure 4 shown.

[0114] In some embodiments, the process of using historical portrait subject information to train a multi-time series memory model to obtain a portrait label generation model can be as follows: for each portrait subject, the historical portrait subject information of the portrait subject can be obtained; the historical feature matrix corresponding to the label type related to the portrait subject under each specified time dimension is extracted from the historical portrait subject information; each historical feature matrix under the label type is input into the recurrent neural network model LSTM through the input layer; each historical feature matrix is ​​mapped to the corresponding sample label space through the recurrent neural network model LSTM according to pre-set business rules to obtain the historical features corresponding to each historical feature matrix, and each historical feature is fused through the convolution layer to obtain the historical label of the label type, and the historical label approaches the target logistics portrait label of the label type as the training goal, and the parameters of the recurrent neural network model LSTM are adjusted until the recurrent neural network model LSTM converges to obtain the portrait label generation model.

[0115] In the process of specifically executing step S105, each feature matrix corresponding to the label type can be input into the portrait label generation model through the input layer, so that the portrait label generation model maps each feature matrix to the corresponding sample label space according to the preset business rules, obtains the features corresponding to each feature matrix, and performs multi-feature fusion of each feature through the convolution layer to obtain the logistics portrait label corresponding to the label type, and finally outputs the logistics portrait label of the label type through the output layer.

[0116] S106: Construct a logistics portrait model corresponding to the portrait subject according to multiple logistics portrait tags corresponding to the portrait subject.

[0117] In the specific execution of step S106, for each portrait subject, after obtaining the logistics portrait label corresponding to each label type related to the portrait subject, a logistics portrait model corresponding to the portrait subject can be constructed according to the multiple logistics portrait labels corresponding to the portrait subject.

[0118] S107: Metadata is defined for the logistics portrait model of the portrait subject, and each logistics portrait label in the portrait model after the metadata definition is graded and classified to obtain a corresponding target logistics portrait model.

[0119] In the embodiment of the present application, after constructing the logistics portrait model corresponding to each portrait subject under each portrait subject domain, each logistics portrait model under different portrait subject domains can be divided according to the data granularity, so as to define the metadata of the logistics portrait model under each data granularity, and obtain the group proportion, update frequency and metadata link corresponding to the logistics portrait model. At the same time, in order to distinguish from the model dimensions of the existing data warehouse construction, the logistics portrait labels in the logistics portrait model can be further graded and classified, so as to facilitate the subsequent use of operators and business personnel.

[0120] It should be noted that one data granularity corresponds to one portrait subject. For example, the data granularity corresponding to the aircraft (portrait subject) under the portrait subject domain of capacity portrait can be the flight number, and the data granularity corresponding to the C customer (portrait subject) under the portrait subject domain of customer portrait can be the C code. Metadata definitions are performed for the logistics portrait models under the above two data granularities, and the results are shown in Table 1.

[0121] Table 1:

[0122]

[0123]

[0124] It should be noted that the various logistics image labels under the image subject can be graded and classified according to the caliber, as shown in Table 2:

[0125]

[0126] Among them, caliber refers to: the processing logic of the label.

[0127] S108: Establish a logistics portrait label asset library based on the target portrait model corresponding to each portrait subject, and establish a corresponding logistics labeling product platform based on the logistics labeling asset library, so that business personnel can generate corresponding logistics operation plans through the logistics labeling product platform.

[0128] In the embodiment of the present application, for each portrait subject, after metadata definition and hierarchical classification are performed on the logistics portrait model corresponding to the portrait subject to obtain the target logistics portrait model, a logistics portrait label asset library is established according to the target logistics portrait model corresponding to each portrait subject, and a corresponding logistics labeling product platform is established according to the logistics label asset library. The logistics labeling product platform includes a logistics portrait label query and management module, a logistics group delineation module, and a personalized recommendation.

[0129] It should be noted that business personnel can query all online logistics labels through the logistics image label query and management module. For example, after the business personnel enter the C code of a certain customer B, they can query the corresponding logistics image label from the target logistics image model corresponding to the C code. Among them, the logistics image label can be a logistics image label such as basic attributes, operating conditions, and credit status.

[0130] In the embodiment of the present application, a logistics group corresponding to each tag combination can also be set, and the set logistics group is stored in the logistics group delineation module. A logistics operation plan corresponding to the logistics group can also be formulated, and the set logistics operation plan is stored in the personalized recommendation module. Among them, the logistics group can be "high-value users", "new active users", "high-frequency complaint users", etc., as shown in Table 3. It can be set according to the actual application, and the embodiment of the present application is not limited.

[0131] Table 3:

[0132] crowd ID Creation time Number of people download Push method operate High value users 001 01-01 1000 yes Short message Update / Delete New Active Users 002 01-02 2000 yes mail Update / Delete Frequent complaining users 003 01-03 100 yes Telephone Update / Delete

[0133] In some embodiments, after the corresponding logistics portrait tags are queried, the various logistics portrait tags can be combined into multiple tags, and based on the multiple tag combinations, a matching logistics group can be determined from the logistics group module, and finally a matching logistics operation plan can be determined from the personalized recommendation module based on the determined logistics group, so as to recommend the logistics operation plan to the corresponding user.

[0134] For example, based on Table 3, for "high-value users", the corresponding logistics portrait label combination can be: "active users in the last 7 days", "daily work delivery period is weekends" and "is a fan of the brother's public account"; the corresponding logistics portrait label combination for "new active users" can be: "active users in the last 7 days" and "daily work delivery period is the last 7 days"; the corresponding logistics portrait label combination for "high-frequency complaint users" can be: "frequent complaints about fake delivery by the brother in the last month" and "complaints about the brother's poor service attitude in the last month". The logistics operation plan for "high-value users" can be "special delivery", the logistics operation plan for "new active users" can be "sending corresponding logistics red envelopes", and the logistics operation plan for "high-frequency complaint users" can be that whenever the goods of such users are delivered, the brother is forced to deliver the goods to the door and take photos. Among them, the brother is the deliveryman.

[0135] For logistics customer C, the logistics portrait tags of customer C are found to be "active user in the last 7 days", "weekends for daily work" and "a fan of Xiaoge's official account". By combining various logistics portrait tags, the logistics group corresponding to customer C can be identified as "high-value users", and then a "special delivery" notification can be sent to the user via SMS.

[0136] Due to the surge in daily shipping orders and a large number of complaints from C customers against couriers, the logistics portrait labels corresponding to these C customers are found to be "frequent complaints about false delivery by couriers in the past month" and "complaints about poor service attitude by couriers in the past month". The various logistics portrait labels are combined to identify the corresponding logistics group as "high-frequency complaint users" so that when delivering the C customer's goods, the delivery boy is required to deliver the goods to the door and take photos.

[0137] The present invention provides a method for generating a logistics portrait model, obtaining a preset portrait construction scheme; wherein the portrait construction scheme includes a portrait system, the portrait system includes multiple portrait subject domains and multiple portrait subjects under each portrait subject domain; collecting logistics portrait data from a data source, and storing the logistics portrait data in a storage implementation module of a preset data storage engine; the storage implementation module implements the storage of logistics portrait data based on an LSM tree, which can improve storage performance to a certain extent, thereby meeting the continuous update and application scenarios of the logistics industry; identifying the portrait subject information corresponding to the portrait subject from the storage implementation module; extracting the feature matrix corresponding to the label type related to the portrait subject under each specified time dimension from the portrait subject information; the specified time dimension is determined based on the label type; inputting each feature matrix corresponding to the label type into a portrait label generation model, so that the portrait label generation model outputs a logistics portrait label corresponding to the label type according to the preset business rules and each feature matrix; wherein the portrait label generation model is obtained by training a multi-temporal memory model based on historical portrait subject information; and constructing a logistics portrait model corresponding to the portrait subject according to multiple logistics portrait labels corresponding to the portrait subject. The technical solution provided by the present invention fully considers the sensitivity of the logistics portrait label in the time dimension, thereby improving the accuracy of the logistics portrait label.

[0138] Based on the logistics portrait model generation method disclosed in the embodiment of the present invention, the embodiment of the present invention also discloses a logistics portrait model generation system, such as Figure 5 As shown, the logistics portrait model generation system includes:

[0139] The portrait construction scheme acquisition unit 51 is used to acquire a preset portrait construction scheme; wherein the portrait construction scheme includes a portrait system, and the portrait system includes multiple portrait subject domains and multiple portrait subjects under each portrait subject domain;

[0140] The logistics image data collection unit 52 is used to collect logistics image data from the data source and store the logistics image data in a storage implementation module of a pre-set data storage engine; the storage implementation module implements the storage of logistics image data based on the LSM tree;

[0141] The portrait subject information identification unit 53 is used to identify the portrait subject information corresponding to the portrait subject from the storage implementation module;

[0142] The first extraction unit 54 is used to extract the feature matrix corresponding to the tag type related to the image subject at each specified time dimension from the image subject information; the specified time dimension is determined based on the tag type;

[0143] The logistics image label generation unit 55 is used to input each feature matrix corresponding to the label type into the image label generation model, so that the image label generation model outputs the logistics image label corresponding to the label type according to the preset business rules and each feature matrix; wherein the image label generation model is obtained by the training unit using the historical image subject information to train the multi-time series memory model;

[0144] The logistics portrait model construction unit 56 is used to construct a logistics portrait model corresponding to the portrait subject according to multiple logistics portrait tags corresponding to the portrait subject.

[0145] The specific principles and execution processes of each unit in the training device of the model disclosed in the above embodiment of the present invention are the same as the logistics portrait model generation method disclosed in the above embodiment of the present invention. Please refer to the corresponding parts of the logistics portrait model generation method disclosed in the above embodiment of the present invention, and will not be repeated here.

[0146] The present invention provides a logistics portrait model generation system, obtains a preset portrait construction scheme; wherein the portrait construction scheme includes a portrait system, the portrait system includes multiple portrait subject domains and multiple portrait subjects under each portrait subject domain; collects logistics portrait data from a data source, and stores the logistics portrait data in a storage implementation module of a preset data storage engine; the storage implementation module implements the storage of logistics portrait data based on an LSM tree, which can improve storage performance to a certain extent, thereby meeting the continuous update and application scenarios of the logistics industry; identifies the portrait subject information corresponding to the portrait subject from the storage implementation module; extracts the feature matrix corresponding to the label type related to the portrait subject under each specified time dimension from the portrait subject information; the specified time dimension is determined based on the label type; inputs each feature matrix corresponding to the label type into the portrait label generation model, so that the portrait label generation model outputs the logistics portrait label corresponding to the label type according to the preset business rules and each feature matrix; wherein the portrait label generation model is obtained by training a multi-temporal memory model based on historical portrait subject information; according to the multiple logistics portrait labels corresponding to the portrait subject, a logistics portrait model corresponding to the portrait subject is constructed. The technical solution provided by the present invention fully considers the sensitivity of the logistics portrait label in the time dimension, thereby improving the accuracy of the logistics portrait label.

[0147] Optionally, the logistics portrait model generation system provided by the embodiment of the present invention further includes:

[0148] The target logistics portrait model determination unit is used to perform metadata definition on the logistics portrait model of the portrait subject, and to classify and grade each logistics portrait label in the portrait model after the metadata definition, so as to obtain the corresponding target logistics portrait model;

[0149] The logistics portrait label asset library establishment unit is used to establish a logistics portrait label asset library according to the target portrait model corresponding to each portrait subject, and to establish a corresponding logistics labeling product platform based on the logistics label asset library, so that business personnel can generate corresponding logistics operation plans through the logistics labeling product platform.

[0150] Optionally, the data source includes a batch data source and a stream data source, and the logistics profile data collection unit includes:

[0151] A first logistics portrait data collection unit is used to collect logistics portrait data from a batch data source through a target data port using a batch data collection system, and store the logistics portrait data in a storage implementation module of a pre-set data storage engine;

[0152] The second logistics portrait data collection unit is used to collect logistics portrait data from the stream data source through the Web server using the batch data collection system, and store the logistics portrait data in the storage implementation module of the pre-set data storage engine.

[0153] Optionally, the batch data collection system includes a linker, a task configurator and a Hadoop cluster, and the first logistics image data collection unit includes:

[0154] A target link configuration template determining unit, used to determine a target link configuration template from various pre-set link configuration templates through a linker; wherein the target link configuration template is a link configuration template corresponding to the Hadoop cluster;

[0155] A link creation unit, used to create a link corresponding to the target data end according to the target link configuration template through the task configurator, and generate a corresponding task according to the link;

[0156] The synchronization unit is used to synchronize tasks to the Hadoop cluster through the task translator;

[0157] The first data migration unit is used to collect logistics portrait data corresponding to the task from the stream data source based on the target data end through the Hadoop cluster, and migrate the logistics portrait data to a pre-set data storage engine.

[0158] Optionally, the stream data collection system includes a stream receiver, a stream buffer and a stream memory, and the second stream image data collection unit includes:

[0159] The second logistics portrait data collection subunit is used to collect corresponding logistics portrait data from the stream data source based on the Web server through the stream receiver;

[0160] The second data migration unit is used to distribute the logistics portrait data to the stream storage device through the stream buffer, and migrate the logistics portrait data to a pre-set data storage engine through the stream storage device.

[0161] Optionally, if the subject of the portrait is a customer, the tag type related to the subject of the portrait is a life cycle, and the specified time dimension corresponding to the tag type includes a monthly time dimension and a daily time dimension, the first extraction unit includes:

[0162] The first statistical unit is used to count multiple categories of the portrait subject in the portrait subject information within a preset time to obtain a fixed feature matrix corresponding to the label types related to the portrait subject in a monthly time dimension;

[0163] The second statistical unit is used to count the number of single amounts of the portrait subject every day from the portrait subject information, and obtain a fixed feature matrix corresponding to the label types related to the portrait subject in the daily time dimension.

[0164] Optional, logistics image label generation unit, including:

[0165] A first input unit, used to input each feature matrix corresponding to the label type into the portrait label generation model;

[0166] The logistics portrait label generation subunit is used to map each feature matrix to the corresponding sample label space according to the preset business rules through the portrait label generation model, obtain the features corresponding to each feature matrix, and perform multi-feature fusion on each feature to obtain the logistics portrait label corresponding to the label type.

[0167] Optionally, based on the multi-temporal memory model including the input layer, the LSTM recurrent neural network model, the convolutional layer and the output layer, the training unit includes:

[0168] A historical portrait subject information acquisition unit, used for acquiring historical portrait subject information of each portrait subject;

[0169] The second extraction unit is used to extract the historical feature matrix corresponding to the label types related to the portrait subject at each specified time dimension from the historical portrait subject information;

[0170] The second input unit is used to input each historical feature matrix under the label type into the recurrent neural network model LSTM through the input layer;

[0171] The training subunit is used to map each historical feature matrix to the corresponding sample label space through the recurrent neural network model LSTM according to the preset business rules, obtain the historical features corresponding to each historical feature matrix, and fuse the various historical features through the convolution layer to obtain the historical labels of the label type, and take the historical labels approaching the target logistics portrait labels of the label type as the training goal, adjust the parameters of the recurrent neural network model LSTM until the recurrent neural network model LSTM reaches convergence, and obtain the portrait label generation model.

[0172] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, in which the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0173] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0174] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0175] The above are only preferred embodiments of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for generating a logistics portrait model, characterized in that: The method comprises: Obtaining a preset portrait construction scheme; wherein the portrait construction scheme includes a portrait system, and the portrait system includes multiple portrait subject domains and multiple portrait subjects under each of the portrait subject domains; Collecting logistics image data from a data source, and storing the logistics image data in a storage implementation module of a pre-set data storage engine; the storage implementation module implements the storage of logistics image data based on an LSM tree; Identifying the portrait subject information corresponding to the portrait subject from the storage implementation module; Extracting a feature matrix corresponding to a tag type related to the subject of the portrait at each specified time dimension from the subject information of the portrait; the specified time dimension is determined based on the tag type; Input each of the feature matrices corresponding to the label type into a portrait label generation model, so that the portrait label generation model outputs a logistics portrait label corresponding to the label type according to the preset business rules and each of the feature matrices; wherein the portrait label generation model is obtained by training a multi-temporal memory model using historical portrait subject information; According to the multiple logistics portrait tags corresponding to the portrait subject, a logistics portrait model corresponding to the portrait subject is constructed.

2. The method according to claim 1, characterized in that: The method further comprises: Defining metadata for the logistics portrait model of the portrait subject, and classifying each logistics portrait label in the portrait model after the metadata definition to obtain a corresponding target logistics portrait model; According to the target portrait model corresponding to each of the portrait subjects, a logistics portrait label asset library is established, and a corresponding logistics labeling product platform is established based on the logistics labeling asset library, so that business personnel can generate corresponding logistics operation plans through the logistics labeling product platform.

3. The method according to claim 1, characterized in that The data source includes a batch data source and a stream data source; The step of collecting logistics image data from a data source and storing the logistics image data in a storage implementation module of a pre-set data storage engine includes: Using a batch data collection system, collecting logistics image data from the batch data source through a target data port, and storing the logistics image data in a storage implementation module of a pre-set data storage engine; The batch data collection system is used to collect logistics portrait data from the stream data source through a Web server, and the logistics portrait data is stored in a storage implementation module of a pre-set data storage engine.

4. The method according to claim 3, characterized in that The batch data collection system includes a linker, a task configurator and a Hadoop cluster. The batch data collection system is used to collect logistics image data from the batch data source through the corresponding data port, and the logistics image data is stored in a storage implementation module of a pre-set data storage engine, including: Determining a target link configuration template from various pre-set link configuration templates through the linker; wherein the target link configuration template is a link configuration template corresponding to the Hadoop cluster; Creating a link corresponding to the target data end according to the target link configuration template by the task configurator, and generating a corresponding task according to the link; Synchronizing the task to the Hadoop cluster through the task translator; The Hadoop cluster collects the logistics portrait data corresponding to the task from the streaming data source based on the target data end, and migrates the logistics portrait data to a pre-set data storage engine.

5. The method according to claim 3, characterized in that: The stream data collection system includes a stream receiver, a stream buffer and a stream memory. The batch data collection system is used to collect logistics image data from the stream data source through a Web server, and the logistics image data is stored in a storage implementation module of a pre-set data storage engine, including: Collecting corresponding logistics portrait data from the stream data source through the stream receiver based on the Web server; The logistics portrait data is distributed to the stream memory through the stream buffer, and the logistics portrait data is migrated to a pre-set data storage engine through the stream memory.

6. The method according to claim 1, characterized in that If the subject of the portrait is a customer, the type of label associated with the subject of the portrait is a life cycle, and the specified time dimensions corresponding to the label types include a monthly time dimension and a daily time dimension, a feature matrix corresponding to the label types associated with the subject of the portrait under each specified time dimension is extracted from the subject information of the portrait, including: Counting multiple categories of the portrait subject in the portrait subject information within a preset time to obtain a fixed feature matrix corresponding to the label types related to the portrait subject in the monthly time dimension; The number of single amounts of the portrait subject per day is counted from the portrait subject information to obtain a fixed feature matrix corresponding to the label types related to the portrait subject under the daily time dimension.

7. The method according to claim 1, characterized in that The step of inputting each of the feature matrices corresponding to the label type into a portrait label generation model so that the portrait label generation model outputs a logistics portrait label corresponding to the label type according to a preset business rule and each of the feature matrices, comprises: Input each of the feature matrices corresponding to the label types into a portrait label generation model; Through the portrait label generation model, each of the feature matrices is mapped to the corresponding sample label space according to the preset business rules, the features corresponding to each feature matrix are obtained, and each feature is fused into multiple features to obtain the logistics portrait label corresponding to the label type.

8. The method according to claim 1, characterized in that The multi-time series memory model includes an input layer, a recurrent neural network model LSTM, a convolutional layer and an output layer. The multi-time series memory model is trained using the historical portrait subject information to obtain a portrait label generation model, including: For each of the portrait subjects, historical portrait subject information of the portrait subject can be obtained; Extracting a historical feature matrix corresponding to the tag types related to the subject of the portrait at each specified time dimension from the subject information of the historical portrait; Input each historical feature matrix under the label type into the recurrent neural network model LSTM through the input layer; Through the recurrent neural network model LSTM, each historical feature matrix is ​​mapped to the corresponding sample label space according to the preset business rules to obtain the historical features corresponding to each historical feature matrix, and the historical features are fused through the convolution layer to obtain the historical label of the label type, and the parameters of the recurrent neural network model LSTM are adjusted with the historical label approaching the target logistics portrait label of the label type as the training goal until the recurrent neural network model LSTM converges to obtain the portrait label generation model.

9. A logistics portrait model generation system, characterized in that: The system comprises: A portrait construction scheme acquisition unit, used to acquire a preset portrait construction scheme; wherein the portrait construction scheme includes a portrait system, and the portrait system includes multiple portrait subject domains and multiple portrait subjects under each of the portrait subject domains; A logistics portrait data collection unit is used to collect logistics portrait data from a data source and store the logistics portrait data in a storage implementation module of a pre-set data storage engine; the storage implementation module implements the storage of logistics portrait data based on an LSM tree; A portrait subject information identification unit, used to identify the portrait subject information corresponding to the portrait subject from the storage implementation module; A first extraction unit is used to extract a feature matrix corresponding to a tag type related to the portrait subject at each specified time dimension from the portrait subject information; the specified time dimension is determined based on the tag type; The logistics portrait label generation unit is used to input each of the feature matrices corresponding to the label type into the portrait label generation model, so that the portrait label generation model outputs the logistics portrait label corresponding to the label type according to the preset business rules and each of the feature matrices; wherein the portrait label generation model is obtained by training the multi-temporal memory model based on the training unit using the historical portrait subject information; The logistics portrait model construction unit is used to construct a logistics portrait model corresponding to the portrait subject according to multiple logistics portrait tags corresponding to the portrait subject.

10. The system according to claim 9, characterized in that The system further comprises: A target logistics portrait model determination unit is used to perform metadata definition on the logistics portrait model of the portrait subject, and to classify each of the logistics portrait tags in the portrait model after the metadata definition to obtain a corresponding target logistics portrait model; The logistics portrait label asset library establishment unit is used to establish a logistics portrait label asset library according to the target portrait model corresponding to each portrait subject, and establish a corresponding logistics labeling product platform based on the logistics label asset library, so that business personnel can generate corresponding logistics operation plans through the logistics labeling product platform.

Images