Method, device, storage medium and electronic device for recognizing object

Abstract

The application discloses a kind of methods for identifying object, device, storage medium and electronic equipment, it is related to artificial intelligence technical field.Therein, the method includes: obtaining the transaction behavior data of at least one object;The transaction behavior data is time-sequenced, and first time sequence is obtained;First time sequence is processed by data dimension reduction, and first feature vector is obtained;The similarity of first feature vector and second feature vector is calculated, and the calculation result is obtained;According to the calculation result, at least one target object is determined from at least one object, and the target object group formed by at least one target object is obtained, wherein the target object group is the potential customer group participating in target marketing activity.The application solves the technical problem that the customer group identification model in the prior art is constructed for single product and the construction time is too long, which leads to the inability to consider the effectiveness and timeliness of multi-product marketing activities.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a method and device for identifying an object, a storage medium and an electronic device. BACKGROUND

[0002] The channel department of a financial institution is responsible for using existing active customers to undertake product marketing tasks from cross-business lines. Generally, in actual operation, there is a contradiction between the effectiveness and timeliness of marketing activities, that is, on the one hand, the superior unit has certain index requirements for the conversion rate of marketing activities, and on the other hand, due to factors such as channel traffic sharing and changes in business objectives, such marketing activities often show the characteristics of short cycle and frequent switching, resulting in that the channel department cannot balance the effectiveness and timeliness of marketing activities.

[0003] At present, the channel department often uses a method based on subjective expert experience or a method using a machine learning model designed for specific products to screen target customers. The former method is simple, but the conversion rate is often difficult to meet the requirements. Especially in the marketing of some unfamiliar scenarios or niche products, business personnel often lack marketing experience and cannot use rigid rules to define potential target customers according to explicit conditions, resulting in inaccurate identification of potential target customers. The latter method can improve the conversion rate, but the design and implementation cycle of months cannot meet the requirements of the channel department for the timeliness of marketing activities. In addition, in the marketing of some niche products, due to the serious imbalance of positive and negative samples and the contradiction with the mathematical assumptions of most models, classical discriminant models often cannot be directly used, and more complex and refined customized solutions need to be introduced to cope with them, which will inevitably further worsen the timeliness of the corresponding technical solutions.

[0004] At present, no effective solution has been proposed to solve the above problems. SUMMARY

[0005] The embodiments of the present application provide a method and device for identifying an object, a storage medium and an electronic device to at least solve the technical problem that the customer identification model in the prior art is constructed for a single product and the construction time is too long, which cannot balance the effectiveness and timeliness of multi-category product marketing activities.

[0006] According to an aspect of some embodiments of the present application, a method for identifying an object is provided, comprising: obtaining transaction behavior data of at least one object; performing time series processing on the transaction behavior data to obtain a first time series, wherein the first time series is used to represent a behavior feature of the at least one object; performing data dimension reduction processing on the first time series to obtain a first feature vector; calculating a similarity between the first feature vector and a second feature vector to obtain a calculation result, wherein the second feature vector is a feature vector of a reference object, and the feature vector of the reference object represents a data feature corresponding to a target marketing activity; and determining at least one target object from the at least one object according to the calculation result to obtain a target object group composed of the at least one target object, wherein the target object group is a potential customer group participating in the target marketing activity.

[0007] Further, the method for identifying an object further comprises: performing format conversion processing on the transaction behavior data to obtain an initial time series; and performing normalization processing on the initial time series to obtain the first time series.

[0008] Further, the method for identifying an object further comprises: performing normalization processing on discrete features in the initial time series to obtain normalized values of the discrete features; performing normalization processing on continuous features in the initial time series to obtain normalized values of the continuous features; and performing splicing processing on the normalized values of the discrete features and the normalized values of the continuous features to obtain the first time series.

[0009] Further, the method for identifying an object further comprises: performing encoding processing on the first time series based on a first residual network to obtain an initial first feature vector; and generating the first feature vector based on the initial first feature vector.

[0010] Further, the method for identifying an object further comprises: receiving an object list of the reference object; obtaining a reference time series based on the object list; and determining the second feature vector based on the reference time series.

[0011] Further, the method for identifying an object further comprises: respectively calculating mathematical expectation values of the first feature vector and the second feature vector to obtain a first expectation value and a second expectation value; and calculating a similarity between the first expectation value and the second expectation value to obtain the calculation result.

[0012] Further, the method for identifying an object further comprises: inputting the calculation result into a first clustering model to perform clustering calculation to obtain mathematical expectation values of a plurality of feature vectors; performing sorting processing on the mathematical expectation values of the plurality of feature vectors, and determining a feature vector corresponding to a maximum mathematical expectation value as a target feature vector; and determining at least one target object corresponding to the target feature vector from the at least one object to obtain the target object group.

[0013] Further, the method for identifying the object further comprises: obtaining object numbers corresponding to at least one target object in the target object group, and generating a target object list; and sending the target object list to a preset platform, where the preset platform is configured to recommend the target marketing activity to the object corresponding to the number based on the target object list.

[0014] According to another aspect of the embodiments of the present application, a device for identifying an object is further provided, which comprises: an obtaining module configured to obtain transaction behavior data of at least one object; a first processing module configured to perform time-series processing on the transaction behavior data to obtain a first time series, where the first time series is used to represent behavior characteristics of the at least one object; a second processing module configured to perform data dimension reduction processing on the first time series to obtain a first feature vector; a calculation module configured to calculate a similarity between the first feature vector and a second feature vector to obtain a calculation result, where the second feature vector is a feature vector of a reference object, and the feature vector of the reference object represents data characteristics corresponding to a target marketing activity; and a determination module configured to determine at least one target object from the at least one object according to the calculation result to obtain a target object group composed of the at least one target object, where the target object group is a potential customer group participating in the target marketing activity.

[0015] According to another aspect of the embodiments of the present application, a computer readable storage medium is further provided, which stores a computer program, where the computer program is configured to execute the above-mentioned method for identifying an object when running.

[0016] According to another aspect of the embodiments of the present application, an electronic device is further provided, which comprises one or more processors; and a memory configured to store one or more programs, which make the one or more processors implement a program for running when the one or more programs are executed by the one or more processors, where the program is configured to execute the above-mentioned method for identifying an object when running.

[0017] In the embodiment of the present application, the object behavior data rule is represented by a feature vector, and the target object group similar to the reference object in behavior is identified by taking the feature vector of the reference object as a reference. First, transaction behavior data of at least one object is obtained, then the transaction behavior data is processed in time sequence to obtain a first time sequence, then the first time sequence is processed in data dimension reduction to obtain a first feature vector, then the similarity between the first feature vector and a second feature vector is calculated to obtain a calculation result, and then at least one target object is determined from the at least one object according to the calculation result to obtain a target object group composed of the at least one target object. The target object group is a potential customer group participating in a target marketing activity, the first time sequence is used to represent the behavior characteristics of the at least one object, and the second feature vector is the feature vector of the reference object, which represents the data characteristics corresponding to the target marketing activity.

[0018] In the above process, the transaction behavior data is processed in time sequence to obtain a first time sequence, and the first time sequence is processed in data dimension reduction to obtain a first feature vector, which realizes a low-dimensional feature vector representation of customers in the behavior level with high discrimination, so that in the subsequent process of determining the target object group, the target customer group and the non-target customer group can be quickly distinguished based on the low-dimensional feature vector representation by calculating the similarity with the historical behavior of the reference object. And the low-dimensional feature vector representation has universality in the scenario of using customer behavior rules to distinguish potential target customer groups, greatly improving the efficiency of multi-product marketing decision support.

[0019] Therefore, by the technical scheme of the present application, the purpose of identifying the target object group is achieved, thereby realizing the technical effect of improving the efficiency of identifying the target object group, and further solving the technical problem in the prior art that the customer group identification model is constructed for a single product and the construction time is too long, which leads to the inability to balance the effectiveness and timeliness of multi-product marketing activities. BRIEF DESCRIPTION OF DRAWINGS

[0020] The drawings described herein are used to provide further understanding of the present application, and form a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation on the present application. In the drawings:

[0021] Figure 1 is a flowchart of an optional method for identifying an object according to an embodiment of the present application;

[0022] Figure 2 is a structural diagram of an optional system for identifying an object according to an embodiment of the present application;

[0023] Figure 3is a structural schematic diagram of an optional VAE network according to an embodiment of the present application;

[0024] Figure 4 is a structural schematic diagram of an optional ResNet Encoder 1D network according to an embodiment of the present application;

[0025] Figure 5 is a flowchart of an optional feature time series data generation process according to an embodiment of the present application;

[0026] Figure 6 is a flowchart of an optional time series data preprocessing process according to an embodiment of the present application;

[0027] Figure 7 is a flowchart of an optional dimension-reduced feature representation mapping generation process according to an embodiment of the present application;

[0028] Figure 8 is a flowchart of an optional dimension-reduced feature representation mapping storage process according to an embodiment of the present application;

[0029] Figure 9 is a flowchart of an optional customer dimension-reduced feature representation vector regular generation process according to an embodiment of the present application;

[0030] Figure 10 is a flowchart of an optional target customer group extraction request process according to an embodiment of the present application;

[0031] Figure 11 is a flowchart of an optional reference benchmark generation process according to an embodiment of the present application;

[0032] Figure 12 is a flowchart of an optional target customer group generation process according to an embodiment of the present application;

[0033] Figure 13 is a schematic diagram of an optional object recognition device according to an embodiment of the present application;

[0034] Figure 14 is a schematic diagram of an optional electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0035] In order to enable persons skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by persons skilled in the art without creative labor should fall within the scope of protection of the present application.

[0036] It should be noted that the terms "first", "second", and the like in the description and in the claims of the present application and in the above-described drawings are used to distinguish similar objects and are not necessarily used to describe a particular chronological or sequential order. It should be understood that the data thus used can be interchanged, where appropriate, so that the embodiments of the application described herein can be carried out in other than the order shown or described herein. Furthermore, the terms "comprising" and "having" and any variations thereof are intended to cover a non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a list of steps or units can not necessarily be limited to those steps or units that are clearly listed, but can include other steps or units that are not clearly listed or inherent to such processes, methods, products, or apparatuses.

[0037] It should be noted that the relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, analyzed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties. For example, an interface is provided between the system and the relevant user or institution. Before obtaining the relevant information, the interface needs to send a request to the aforementioned user or institution, and after receiving the consent information feedback from the aforementioned user or institution, the relevant information is obtained.

[0038] Embodiment 1

[0039] According to an embodiment of the present application, an embodiment of a method for identifying an object is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0040] Figure 1 is a flowchart of an optional method for identifying an object according to an embodiment of the present application, as shown in Figure 1 The method comprises the following steps:

[0041] Step S101, obtaining transaction behavior data of at least one object.

[0042] In the above steps, the at least one object can be a customer of a financial institution, and the transaction behavior data can be data related to transaction behavior accumulated by the customer at the financial institution (including but not limited to the number of times the customer logs in to the mobile bank, the amount of money transferred, the customer's asset balance, etc.), customer attribute data (such as marital status, work experience, residence, etc.).

[0043] In step S102, the transaction behavior data is subjected to time series processing to obtain a first time series, wherein the first time series is used to represent the behavior characteristics of the at least one object.

[0044] In step S103, the first time series is subjected to data dimension reduction processing to obtain a first feature vector.

[0045] Specifically, in the embodiment, the identification of the target object group mainly involves three processes, i.e., time series processing of feature variables, generation of a dimension reduction feature representation vector of customer behavior, and customer group identification for specific marketing applications. Among them, the time series processing of feature variables realizes two main functions, i.e., time series generation of index dimension and preprocessing function of time series. First, the time series generation function of index dimension extracts the state value of each index dimension within a certain time range at the minimum time granularity according to the preset requirements from the full active customers, and processes the state value of each index dimension into a time series arranged in time sequence according to the preset time granularity requirement. Further, according to the attribute of the feature (divided into discrete feature and continuous feature), different methods are used for data normalization preprocessing operation to obtain the first time series.

[0046] Optionally, the generation of the dimension reduction feature representation vector of customer behavior is to input the time series that has completed the normalization processing into a VAE (Variational Auto-Encoder) network model for training according to a preset feature arrangement order, so as to form a mapping relationship between the input time series and the dimension reduction feature representation. When the mapping relationship in the form of network parameters is persistently stored, the system can periodically use the time series generated by the latest behavior data of the full active customers to generate the dimension reduction feature representation of the latest behavior data of the full active customers, i.e., the first feature vector, by loading the mapping relationship into the network structure, and save it in the data warehouse. Among them, the VAE network model adopts a structure similar to the ResNet50 network on both sides for input and regeneration of normalized time series data, and the middle part is used to form two groups of vectors representing the mathematical expectation value and standard deviation value of the dimension reduction feature representation vector of customer behavior.

[0047] In step S104, the similarity between the first feature vector and a second feature vector is calculated to obtain a calculation result, wherein the second feature vector is a feature vector of a reference object, and the feature vector of the reference object represents the data characteristics corresponding to the target marketing activity.

[0048] In step S105, at least one target object is determined from the at least one object according to the calculation result to obtain a target object group composed of the at least one target object, wherein the target object group is a potential customer group participating in the target marketing activity.

[0049] Optionally, the customer group identification process for specific marketing applications mainly extracts the corresponding dimension reduction feature representation results from the data warehouse through the customers participating in the target marketing activities, i.e., reference objects, and calculates the mathematical expectation value of the dimension reduction feature representation of each group in the form of N groups to form M reference benchmark vectors, i.e., second feature vectors.

[0050] Further, the similarity between the first feature vector and the second feature vector is calculated to form the targeted feature for the target customer group identification of this marketing activity. Further, according to different values of the component quantity parameter K within a certain range, the fitting effect of the targeted feature probability distribution is obtained through the GMM (Gaussian Mixture Model) model. Further, the component Gaussian function with the maximum mathematical expectation value in the optimal effect is selected as the cluster where the target customer group is located. Optionally, using the trained GMM model, all customers belonging to the cluster can be extracted to obtain the target object list.

[0051] Optionally, Figure 2 is a structural diagram of an optional object identification system according to an embodiment of the present application, as Figure 2 shown, the system includes 9 devices, which are feature time sequence generation device 1, time sequence data preprocessing device 2, dimension reduction feature representation mapping generation device 3, dimension reduction feature representation mapping storage device 4, customer dimension reduction representation vector periodic generation device 5, target customer group extraction request processing device 6, reference benchmark generation device 7, target customer group generation device 8 and data warehouse 9. Optionally, the subsequent “time sequence processing of feature variables”, “dimension reduction feature representation vector generation processing of customer behavior” and “customer group identification processing for specific marketing applications” will be introduced in detail in combination with the present system.

[0052] Based on the schemes defined in the above steps S101 to S105, it can be known that in the embodiment of the present application, the way of identifying the target object customer group similar to the reference object in behavior by using the feature vector to represent the object behavior data rule and taking the feature vector of the reference object as the benchmark, first acquires the transaction behavior data of at least one object, then performs time sequence processing on the transaction behavior data to obtain a first time sequence, then performs data dimension reduction processing on the first time sequence to obtain a first feature vector, then calculates the similarity between the first feature vector and a second feature vector to obtain a calculation result, and then determines at least one target object from the at least one object according to the calculation result to obtain a target object group composed of the at least one target object. The target object group is a potential customer group participating in the target marketing activity, the first time sequence is used to represent the behavior characteristics of the at least one object, the second feature vector is the feature vector of the reference object, and the feature vector of the reference object represents the data characteristics corresponding to the target marketing activity.

[0053] It is easy to note that in the above process, by time-series processing of transaction behavior data, a first time series can be obtained, and then by data dimension reduction processing of the first time series, a first feature vector can be obtained, realizing a low-dimensional feature vector representation of customers with high distinction at the behavior level, so that in the subsequent process of determining the target object group, based on the low-dimensional feature vector representation, the similarity with the reference object historical behavior is calculated, and the target customer group and the non-target customer group are quickly distinguished. And the low-dimensional feature vector representation has universality in the scene of using customer behavior rules to distinguish potential target customer groups, greatly improving the efficiency of multi-product marketing decision support.

[0054] Therefore, through the technical scheme of the present application, the purpose of identifying the target object group is achieved, thereby realizing the technical effect of improving the efficiency of identifying the target object group, and further solving the technical problem in the prior art that the customer group identification model is constructed for a single product and the construction time is too long, which leads to the inability to balance the effectiveness and timeliness of multi-product marketing activities.

[0055] In an optional embodiment, in the process of time-series processing of transaction behavior data to obtain a first time series, first, the transaction behavior data is subjected to format conversion processing to obtain an initial time series, and then the initial time series is subjected to normalization processing to obtain the first time series.

[0056] Optionally, the time-series processing of the feature variable is mainly realized by three modules of the feature time-series data generation device 1, the time-series data preprocessing device 2 and the data warehouse 9. Specifically, the feature time-series data generation device 1 receives a request for generating feature time-series data sent by the customer dimension-reduced feature representation vector periodic generation device 5 or the dimension-reduced feature representation mapping generation device 3, and then extracts a basic customer list CL from the data warehouse 9 with a preset SQL condition D1 (for example, mobile bank activity) as a customer screening condition. Further, based on a feature index list FL marked with a preset use transaction code (for example, the code of financial product A), and with a preset SQL condition D2 (for example, whether to purchase financial product A, purchase times, etc.) as an index feature screening condition, and with the basic customer list CL as a limited range, the feature time-series data generation device 1 extracts N f index feature data for each customer from the data warehouse 9. Wherein, the extraction result of the feature index k of the customer j can be represented as a sequence df j,k covering the last 52 weeks, arranged in ascending order of time, and retaining the order sequence of each element date label therein. Further, by subtracting the adjacent element date labels in reverse order from the sequence, the time difference between adjacent elements is calculated to form a time difference sequence dd j,k . The elements in the time difference sequence dd j,k with a value greater than 1 are calculated and corresponding positions Search in df and element with index and Construct a date tag from Starting from the day after the date label The date tag of the previous day with df j,k Sequential sequences ds with the same structure j,k And make all data values ​​in the sequence equal to The data value. In Traverse all The corresponding generated Insert into and This process ultimately forms a complete, ascending time series ff. j,k Furthermore, based on the system settings, the sum of weekly indicator data is calculated on a weekly basis, and then sorted in ascending order by time to form a complete time series containing only the data. j,k Optionally, all customers and metrics characteristics can be iterated through using the same method, ultimately forming a vertical length of N. f The horizontal length is 52, the depth is the same as the length of the basic customer list CL, the vertical direction is the same as the feature order of the feature index list FL, and the depth direction is the same as the customer order of the basic customer list CL.

[0057] Furthermore, using the customer behavior tensor T and the basic customer list CL as inputs to the time-series data preprocessing device 2, the customer behavior tensor T is sliced ​​vertically from top to bottom, and each tensor slice is normalized. Based on a preset index feature attribute lookup table FM, which is ordered identically to the feature index list FL, the feature type corresponding to each slice is matched.

[0058] Optionally, if the index feature corresponding to the slice is continuous, then the kernel density estimation algorithm is used to fit the probability density function of the continuous index feature slice TC. Since the kernel density estimation algorithm is a non-parametric fitting method, it cannot obtain a parametric representation of the probability density function; instead, it uses an equally spaced sequence as the numerical set of the slice TC. Reference sequence TC ref And generate the probability density values ​​p corresponding to the elements in this reference sequence. ref This is used as the probability density function fitting curve for the continuous index feature c. Optionally, for each element TC in the slice TC... k Find the closest point in the reference sequence. and their corresponding probability density values Calculate the cumulative distribution function value of the point Output the normalized value of the feature element.

[0059] Optionally, if the slice corresponds to a discrete type of index feature, calculate the cumulative distribution function value of the point k Number of occurrences n k Percentage of slice length n, take the logarithm of the percentage as the mapping of each discrete variable value Optionally, estimate the probability density function P of the mapping sequence by exponential distribution parameter estimation map .

[0060] Further, for each element TD k in the slice TD , calculate the cumulative distribution function value of the point Output the normalized value of the feature element. Further, the time series data preprocessing device 2 concatenates each normalized slice in the horizontal direction to output the normalized customer behavior tensor TS.

[0061] Optionally, when the system receives a request from the customer dimension reduction feature representation vector periodic generation device 5 or the dimension reduction feature representation mapping generation device 3, it will complete the time series processing of the feature variable through the feature time series data generation processing and the time series data preprocessing. The feature time series data generation processing and the time series data normalization processing operation steps are described in detail below.

[0062] Optionally, Figure 5 is an optional feature time series data generation processing flowchart according to an embodiment of the present application, as Figure 5 shown, the feature time series data generation processing is implemented through the following steps:

[0063] Step 100: After receiving a request from the customer dimension reduction feature representation vector periodic generation device 5 or the dimension reduction feature representation mapping generation device 3, the pre-set SQL condition D1 is used as the customer screening condition to extract the basic customer list CL from the data warehouse 9. Based on the pre-set feature index list FL marked with the use transaction code, the pre-set SQL condition D2 is used as the index feature screening condition, and the basic customer list CL is used as the limited range to extract N f types of index feature data for each customer from the data warehouse 9.

[0064] Step 101: The extraction result of the feature index k of the customer j can be represented as a sequence df j,k covering the last 52 weeks, arranged in ascending order of time, and retaining the order sequence of each element date label. By subtracting the adjacent element date labels in reverse order from the sequence, the time difference between adjacent elements is calculated to form a time difference sequence ddj,k .

[0065] Step 102: Calculate the time difference sequence dd j,k Elements with median greater than 1 And the corresponding position Find the element indexed by And In df And Construct a date label from The date label of the day after The date label of the day before and df j,k The sequence ds with the same structure as j,k , and make all data values in the sequence equal to Data values. In , iterate through all Corresponding Insert into Between Finally form a complete time series ff with ascending order of dates j,k .

[0066] Step 103: According to the system settings, calculate the sum of the index data of each week in natural weeks, arrange in ascending order of time, and form a complete time series fw containing only data j,k .

[0067] Step 104: Traverse all customers and index characteristics according to the same method, and finally form a customer behavior tensor T with vertical length Nf, horizontal length 52, depth same as the length of the basic customer list CL, and the same characteristics as the index list FL in the vertical direction. The order of the customers in the depth direction is the same as the customer order in the basic customer list CL.

[0068] In an optional embodiment, in the process of normalizing the initial time series to obtain the first time series, first, the discrete features in the initial time series are normalized to obtain the normalized values of the discrete features, then the continuous features in the initial time series are normalized to obtain the normalized values of the continuous features, and then the normalized values of the discrete features and the normalized values of the continuous features are spliced to obtain the first time series.

[0069] Optionally, Figure 6 is an optional flowchart for preprocessing time series data according to an embodiment of the application, as Figure 6 shown, the time series data normalization process is implemented by the following steps:

[0070] Step 200: Upon receiving a time series data preprocessing request, the time series data preprocessing device 2 takes the customer behavior tensor T and the basic customer list CL as inputs, slices the customer behavior tensor T from top to bottom along the vertical direction, and matches the feature type corresponding to each slice according to the preset indicator feature attribute comparison table FM, which is ordered the same as the feature indicator list FL.

[0071] Step 201: If the index feature corresponding to the slice is continuous, then use the kernel density estimation algorithm to fit the probability density function of the continuous index feature slice TC. For the numerical set of slice TC... Using a kernel density estimation algorithm, the range is... TC isochronous reference sequence ref Generate the corresponding probability density value sequence p ref ,in, Representative take The largest element in TC. ref The spacing between elements is determined by the precision w and is preset within the system. The corresponding operations can be expressed by the following formula:

[0072]

[0073] in, and Representing TC ref The value of the r-th element in the sequence and its corresponding probability density. yes The u-th element in the array, where n represents The number of elements in the kernel, h represents the bandwidth parameter in the kernel density estimation algorithm, which is preset by the system. This represents the kernel density function expressed as a standard normal distribution.

[0074] Step 202: Following the order of the basic customer list (CL), ... Each value TC k From the reference sequence TC ref Find the element with the smallest difference. calculate Cumulative distribution function TC k Mapped to The output is the result of normalization.

[0075] Step 203: If the index feature corresponding to the slice is discrete, then for the continuous index feature slice TD, the numerical set... Each discrete value TD k The number of times n k The percentage of the slice length n, taking the logarithm of the percentage yields the mapping set of discrete variable values. That is

[0076] Step 204: mapping set Exponential distribution parameter estimation is performed to obtain its probability density function P map , for each element in the mapping set Calculate its cumulative distribution function value about P map According to the order of the basic customer list CL, map TD k to as the normalized output.

[0077] Step 205: The time series data preprocessing device 2 splices each slice after normalization processing according to the order of the basic customer list CL along the horizontal direction to output the normalized customer behavior tensor TS and the basic customer list CL.

[0078] It should be noted that in the above process, the normalization conversion of different attribute data is realized, which meets the basic requirements of the deep learning model for the specifications of the input data, so that data of different attributes can be processed in the same way under a unified model framework. In addition, the application adopts one-dimensional convolution processing as a feature extraction unit, so that the system can extract more general customer behavior rules while effectively avoiding crosstalk caused by simultaneous processing of different types of behavior data.

[0079] In an optional embodiment, in the process of performing data dimension reduction processing on the first time series to obtain the first feature vector, the first time series is encoded based on a first residual network to obtain an initial first feature vector; and the first feature vector is generated based on the initial first feature vector.

[0080] In an optional embodiment, before calculating the similarity between the first feature vector and the second feature vector to obtain the calculation result, an object list of a reference object is received, and then the reference time series is obtained based on the object list, and the second feature vector is determined based on the reference time series.

[0081] Optionally, in an optional embodiment, in the process of calculating the similarity between the first feature vector and the second feature vector to obtain the calculation result, the mathematical expectation values of the first feature vector and the second feature vector are calculated respectively to obtain first and second expectation values, and then the similarity between the first and second expectation values is calculated to obtain the calculation result.

[0082] ​​Optionally, the customer behavior dimension reduction feature representation vector generation process is mainly implemented through six modules, namely, the feature time series data generation device 1, the time series data preprocessing device 2, the dimension reduction feature representation mapping generation device 3, the dimension reduction feature representation mapping storage device 4, the customer dimension reduction feature representation vector regular generation device 5 and the data warehouse 9.

[0083] Specifically, the dimension reduction feature representation mapping generation device 3 triggers the feature time series data generation device 1 at a specified time period and obtains the normalized customer behavior tensor TS from the time series data preprocessing device 2 at a certain preset time frequency (for example, 3 months).

[0084] Further, the normalized customer behavior tensor TS is divided into a plurality of micro-batch tensors BTS along the depth direction from near to far according to a preset micro-batch size BS, and each size is [BS, N f , 52], wherein the last dimension of the tensor represents the channel to be processed by the subsequent one-dimensional convolution operation. Optionally, if the depth size of the normalized customer behavior tensor TS cannot be divided by the micro-batch size BS, the size of the last micro-batch tensor BTS is [DS%BS, N f , 52], wherein DS is the depth size of the normalized customer behavior tensor TS, and % is the remainder operation. Optionally, the micro-batch tensors BTS are input into the VAE network as shown in Figure 3 , and the restored micro-batch tensor BTS' is generated to restore the input micro-batch tensor BTS, and the network is trained.

[0085] Specifically, as shown in Figure 3 , the Linear_mu and Linear_var modules in the VAE network represent standard fully connected network layers, C out represents the number of channels of the output data after processing. exp((·) / 2) represents a tensor with the same size as the input data, which is calculated by taking half of each element of the input tensor as the exponent with e as the base. RG represents a module for generating random data in the range [0, 1).

[0086] In addition, as shown in Figure 3 , the VAE network includes two processing modules on both sides, which are the first residual network, namely ResNet Encoder 1D and ResNet Decoder 1D, and have the same structure but opposite data flow. Specifically, the two processing modules are respectively used to extract rules from the input time series and restore the original state of the input time series from the extracted rules.

[0087] Specifically, Figure 4is a structural schematic diagram of an optional ResNet Encoder 1D network according to an embodiment of the present application, as shown in Figure 4 Conv1d represents a module for performing a standard convolution operation along the channel dimension direction of a data tensor, where K is the size of the convolution kernel, S is the convolution step, P is the number of boundary padding, C out represents the dimension of the output after processing. BN represents a module for adjusting the input value distribution of neurons in the neural network to a normal distribution by a normalization method, so that the input of each layer of the neural network maintains the same distribution; ReLu is a ReLu activation function in the neural network, and tanh is a tanh activation function in the neural network. MaxPooling represents a MaxPool1d maximum value pooling module, where K is the size of the pooling window, S is the sliding pooling window step, P is the number of boundary padding, and D is the step between elements inside the pooling window. The Adaptive_Avg_Pool1d module is a standard 1D adaptive average pooling module in the neural network, and the Linear module represents a standard fully connected network layer. Block1, Block2, and Block3 represent custom reuse modules, which are combinations of the modules in the dashed boxes.

[0088] Optionally, based on the foregoing VAE network structure, using the micro-batch tensor BTS as the training data, the network is trained according to the preset training round number and the learning rate parameter. Further, the training result with the minimum loss value in the training process is obtained as the final confirmed network training result. Optionally, the network parameters in the final confirmed network training result are output to the dimension reduction feature representation mapping storage device 4 as the customer behavior dimension reduction feature representation mapping relationship.

[0089] Further, after receiving the customer behavior dimension reduction feature representation mapping relationship in the form of network parameters, the dimension reduction feature representation mapping storage device 4 saves it in the storage medium with the current date as the index. Further, when receiving a request from the customer dimension reduction feature representation vector generation device 5, the dimension reduction feature representation mapping device 4 sends the customer behavior dimension reduction feature representation mapping relationship corresponding to the latest date in the current storage as the result to the customer dimension reduction feature representation vector generation device 5.

[0090] Further, the customer dimension reduction feature representation vector generation device 5 triggers the feature time series data generation device 1 at a certain preset time frequency (for example, 3 months), and obtains the basic customer list CL from the time series data preprocessing device 2, as well as the normalized customer behavior tensor TS qMeanwhile, the customer reduced-dimension feature representation vector generation device 5 also acquires the latest customer behavior reduced-dimension feature representation mapping relationship from the reduced-dimension feature representation mapping device 4 in the form of a service request. Optionally, by loading the network parameters in this mapping relationship into the VAE network shown in FIG. 6, the mapping relationship from customer behavior to the low-dimension feature space obtained by previous training can be restored. Figure 3

[0091] Further, by traversing the normalized customer behavior tensor TS q along the depth direction and taking each traversal result as the input of the network, the mathematical expectation value of the customer behavior reduced-dimension feature representation vector output by the Linear_mu module can be obtained. Further, according to the traversal order, the correspondence between the mathematical expectation value of the customer behavior reduced-dimension feature representation vector and the customer number in the basic customer list CL is determined, and a data item composed of the customer number, the mathematical expectation value of the customer behavior reduced-dimension feature representation vector and the generation date is formed. Finally, all the data items are merged into a data table and loaded into the data warehouse 9.

[0092] Optionally, when a preset time period node is reached, a trigger starts the customer behavior reduced-dimension feature representation vector generation processing operation, and then the customer behavior reduced-dimension feature representation vector generation work is completed through the three operation steps of the reduced-dimension feature representation mapping generation processing, the reduced-dimension feature representation mapping storage processing and the customer reduced-dimension feature representation vector regular generation processing. The reduced-dimension feature representation mapping generation processing, the reduced-dimension feature representation mapping storage processing and the customer reduced-dimension feature representation vector regular generation processing operation steps are described in detail below.

[0093] Optionally, Figure 7 is a flowchart of an optional reduced-dimension feature representation mapping generation processing according to an embodiment of the present application, as shown in FIG. 5, the reduced-dimension feature representation mapping generation processing is implemented through the following steps: Figure 7

[0094] Step 300: When the reduced-dimension feature representation mapping generation device 3 reaches a preset time period node, it sends a time series processing request of the feature variable to the feature time series data generation device 1, and obtains the normalized customer behavior tensor TS from the time series data preprocessing device 2.

[0095] ​​Step 301: Divide the normalized customer behavior tensor TS into multiple micro-batch tensors BTS, each of which has a size of [BS, Nf, 52], along the depth direction from near to far, with the preset micro-batch size BS as the standard, where the last dimension of the tensor represents the channel to be processed by the subsequent one-dimensional convolution operation. If the depth size of the normalized customer behavior tensor TS cannot be divided by the micro-batch size BS, the size of the last micro-batch tensor BTS is [DS%BS, Nf, 52], where DS is the depth size of the normalized customer behavior tensor TS, and % is the modulo operation.

[0096] Step 302: Use the VAE network structure defined above to train the network with the micro-batch tensor BTS as the training data, according to the preset training epoch and learning rate parameter lr. The micro-batch tensor BTS with a size of [BS, Nf, 52] will generate a transformed tensor BTS' of the same size after passing through the VAE network structure, as well as two sets of vectors of the customer behavior dimension reduction feature representation mathematical expectation μ (generated by the Linear_mu module of Figure 3 Figure 3

[0097]

[0098] where σ is the learning fluctuation factor in the VAE network, and is the introduced transformation amount in each training round of the VAE network. To avoid excessive fluctuation, σ is initialized to 1, and each training round is assigned σ = log(1+e σ-6 )-6. When the loss value loss gradually converges in each training round, the training result with the minimum loss value in the training process is obtained as the final confirmed network training result.

[0099] Step 303: The network parameters in the final confirmed network training result are regarded as the customer behavior dimension reduction feature representation mapping relationship M VAE output to the dimension reduction feature representation mapping storage device 4.

[0100] Optionally, Figure 8 is an optional dimension reduction feature representation mapping storage process according to an embodiment of the application, as shown in Figure 8 The dimension reduction feature representation mapping storage process is implemented by the following steps:

[0101] Step 400: After receiving the customer behavior dimension reduction feature representation mapping relationship M VAE in the form of network parameters, the dimension reduction feature representation mapping storage device 4 will save it in the storage medium with the current date dt as the index.

[0102] ​​Step 401: When receiving a request from the customer reduced-dimension feature representation vector periodic generation device 5, the reduced-dimension feature representation mapping device 4 maps the customer behavior reduced-dimension feature representation mapping relationship corresponding to the most recent date dt in the current storage to the customer behavior reduced-dimension feature representation mapping relationship corresponding to the current date. As a result, the customer behavior reduced-dimension feature representation vector periodic generation device 5 is sent.

[0103] Optionally, Figure 9 is a flowchart of an optional customer reduced-dimension feature representation vector periodic generation process according to an embodiment of the present application, as shown in Figure 9 The customer reduced-dimension feature representation vector periodic generation process is implemented through the following steps:

[0104] Step 500: The customer reduced-dimension feature representation vector periodic generation device 5 triggers the feature time series data generation device 1 at a preset time frequency, obtains the basic customer list CL from the time series data preprocessing device 2, and obtains the normalized customer behavior tensor TS in the depth direction in the same order as the customers in the basic customer list CL q .

[0105] Step 501: The customer reduced-dimension feature representation vector periodic generation device 5 also obtains the customer behavior reduced-dimension feature representation mapping relationship M from the reduced-dimension feature representation mapping device 4 in the form of a service request, which is the most recent customer behavior reduced-dimension feature representation mapping relationship from the current date VAE .

[0106] Step 502: Load the network parameters in this mapping relationship into the VAE network as shown in Figure 3 , which loads the mapping relationship M from the customer behavior to the low-dimensional feature space obtained by previous training VAE .

[0107] Step 503: Traverse the normalized customer behavior tensor TS in the depth direction q , and take each traversal result as the input of the network, to obtain the mathematical expectation value Em of the customer behavior reduced-dimension feature representation vector output from the Linear_mu module.

[0108] Step 504: Determine the correspondence between the mathematical expectation value Em of the customer behavior reduced-dimension feature representation vector and the customer number in the basic customer list CL according to the traversal order, to form a data item composed of the customer number Pid, the mathematical expectation value Em of the customer behavior reduced-dimension feature representation vector, and the generation date dt. Finally, all data items are merged into a data table and loaded into the data warehouse 9.

[0109] In an alternative embodiment, in the process of determining at least one target object from the at least one object according to the calculation result, obtaining a target object group composed of the at least one target object, the calculation result is input into a first clustering model for clustering calculation, to obtain mathematical expectation values of the plurality of feature vectors, then the mathematical expectation values of the plurality of feature vectors are sorted, and the feature vector corresponding to the maximum mathematical expectation value is determined as a target feature vector, then at least one target object corresponding to the target feature vector is determined from the at least one object, to obtain the target object group.

[0110] Optionally, in an alternative embodiment, after determining at least one target object from the at least one object according to the calculation result, obtaining a target object group composed of the at least one target object, the object number corresponding to the at least one target object in the target object group is obtained, a target object list is generated, and then the target object list is sent to a preset platform, wherein the preset platform is used to recommend target marketing activities to the object corresponding to the number based on the target object list.

[0111] Specifically, the customer group identification process for specific marketing applications is mainly implemented through four modules, namely a target customer group extraction request device 6, a reference benchmark generation device 7, a target customer group generation device 8, and a data warehouse 9. Optionally, the marketing department user can trigger the target customer group extraction request device 6 by uploading a reference customer group list RCL composed of customer numbers. First, the customer group extraction request device 6 sends the reference customer group list RCL to the reference benchmark generation device 7, then obtains a target customer list TCL composed of customer numbers from the target customer group generation device 8, and feeds back the target customer list TCL as a processing result to the marketing department user, to complete all operations.

[0112] Further, the reference benchmark generating device 7, after receiving the reference customer list RCL sent by the customer group extracting request device 6, will randomly select a part of the customers according to the preset reference customer number RCC to form a reference customer seed list RCSL, and the rest of the customers will form a reference customer test list RCTL. Then, using the customer numbers in the reference customer seed list RCSL and the reference customer test list RCTL, the mathematical expectation values of the corresponding reduced dimension feature representation vectors are extracted from the data warehouse 9 respectively according to the principle of the closest data generation date to the current date, to form the seed feature vector mathematical expectation value SFVA and the test feature vector mathematical expectation value TFVA. Thereafter, according to the preset number of reference customers per group RGS, the RGS seed feature vector mathematical expectation values SFVA are divided into a group to form RGN seed feature vector groups in a random extraction manner. Further, all the seed feature vector groups are traversed, and the mathematical expectation value of each dimension is calculated in each group to form a reference feature vector for each group. Finally, the RGN reference feature vectors, the reference customer list RCL and the test feature vector mathematical expectation value TFVA are output as results to the target customer group generating device 8.

[0113] Further, the target customer group generating device 8, after receiving the output of the reference benchmark generating device 7, will first extract a potential customer list QCL composed of the remaining customer numbers and the corresponding reduced dimension feature representation vector mathematical expectation values QVA from the data warehouse 9 by limiting the customer numbers not in the reference customer list RCL according to the principle of the closest data generation date to the current date. Further, all the reduced dimension feature representation vector mathematical expectation values QVA are traversed, and each reduced dimension feature representation vector mathematical expectation value QVA is converted into a reference target similarity vector RSV with RGN dimensions by calculating the cosine similarity of each QVA and each reference feature vector. Each test feature vector mathematical expectation value TFVA is converted into a test reference target similarity vector TRSV with RGN dimensions in the same way. Finally, the reference target similarity vector RSV is input, a standard RGN-dimensional GMM model is trained under each clustering number parameter in the preset clustering number variation range of [2, 5), and the cluster with the largest modulus of the mathematical expectation vector obtained by each training is taken as the target customer cluster.

[0114] Optionally, in order to select the best clustering result as the output, the hit rate of the test reference target similarity vector TRSV in the target customer cluster is calculated according to different clustering number parameter values, the F1 score is calculated, and finally the model corresponding to the highest F1 value is selected as the target customer identification model TRM. Finally, the target customer identification model TRM is used to calculate the mathematical expectation value QVA of the full-dimensional reduced feature representation vector, and according to the corresponding relationship between the reduced feature representation vector and the potential customer list QCL, the customer number records belonging to the target customer cluster are filtered out, and the result is returned to the target customer cluster extraction request processing device 6.

[0115] Optionally, when the marketing department user uploads the reference customer list composed of customer numbers, the customer cluster identification processing operation for specific marketing applications is triggered, and the customer cluster identification processing operation for specific marketing applications is completed through the three operation steps of target customer cluster extraction request processing, reference benchmark generation processing, and target customer cluster generation processing. The target customer cluster extraction request processing, reference benchmark generation processing, and target customer cluster generation processing operation steps are described in detail below.

[0116] Optionally, Figure 10 is a flowchart of an optional target customer cluster extraction request processing according to an embodiment of the present application, as shown in Figure 10 The target customer cluster extraction request processing is implemented through the following steps:

[0117] Step 600: When the marketing department user uploads the reference customer list RCL composed of customer numbers, the target customer cluster extraction request processing is triggered, and the customer cluster extraction request device 6 sends the reference customer list RCL to the reference benchmark generation device 7.

[0118] Step 601: After the customer cluster extraction request device 6 receives the target customer list TCL composed of customer numbers from the target customer cluster generation device 8, the target customer list TCL is fed back to the marketing department user as the processing result.

[0119] Optionally, Figure 11 is a flowchart of an optional reference benchmark generation processing according to an embodiment of the present application, as shown in Figure 11 The reference benchmark generation processing is implemented through the following steps:

[0120] Step 700: After the reference benchmark generation device 7 receives the reference customer list RCL sent by the customer cluster extraction request device 6, a part of the customers are randomly selected as the reference customer seed list RCSL according to the preset reference customer number RCC, and the remaining customers form the reference customer test list RCTL.

[0121] Step 701: Using the customer numbers in the reference customer seed list RCSL and the reference customer test list RCTL, the mathematical expectation values of the corresponding reduced dimension feature representation vectors are extracted from the data warehouse 9 respectively, with the principle of the data generation date being closest to the current date, to form the seed feature vector mathematical expectation value SFVA and the test feature vector mathematical expectation value TFVA.

[0122] Step 702: According to the preset number of reference customers per group RGS, the RGS seed feature vector mathematical expectation values SFVA are divided into a group to form RGN seed feature vector groups in a random extraction manner. The mathematical expectation values of each dimension are calculated in each group, and a reference feature vector is formed for each group.

[0123] Step 703: The RGN reference feature vectors, the reference customer list RCL, and the test feature vector mathematical expectation value TFVA are output as results to the target customer generation device 8.

[0124] Optionally, Figure 12 is an optional target customer generation process flowchart according to an embodiment of the present application, as Figure 12 shown, the target customer generation process is implemented by the following steps:

[0125] Step 800: After receiving the output of the reference benchmark generation device 7, the target customer generation device 8 will first extract the potential customer list QCL composed of the remaining customer numbers and the mathematical expectation values QVA of the corresponding reduced dimension feature representation vectors from the data warehouse 9, with the principle of the data generation date being closest to the current date and the condition of limiting the customer numbers not being in the reference customer list RCL.

[0126] Step 801: Traverse all the mathematical expectation values QVA of the reduced dimension feature representation vectors, and convert each mathematical expectation value QVA of the reduced dimension feature representation vector into a reference target similarity vector RSV with RGN dimensions by calculating the cosine similarity of each QVA and each reference feature vector. Each test feature vector mathematical expectation value TFVA is converted into a test reference target similarity vector TRSV with RGN dimensions using the same method.

[0127] Step 802: Taking the reference target similarity vector RSV as input, a standard RGN-dimensional GMM model is trained under each clustering number parameter in the preset clustering number variation range of [2, 5), and the cluster with the largest modulus mathematical expectation vector obtained by each training is taken as the target customer cluster.

[0128] Step 803: In order to select the result with the best clustering effect as the output, the hit situation of the test reference target similarity vector TRSV in the target customer cluster needs to be calculated according to different clustering quantity parameter values, and the F1 score corresponding to each parameter value is calculated, and finally the model corresponding to the highest F1 value is selected as the target customer recognition model TRM.

[0129] Step 804: The target customer recognition model TRM is used to calculate the mathematical expectation value QVA of the full-dimensional reduced feature representation vector, and according to the correspondence between the mathematical expectation value QVA of the reduced feature representation vector and the potential customer list QCL, the customer number record belonging to the target customer cluster is filtered out, and the result is returned to the target customer cluster extraction request processing device 6.

[0130] It should be noted that in the embodiment, a general customer recognition model for product marketing is introduced, which can not only make up for the low target customer conversion rate in manual customer recognition and the inability to define the customer group in the marketing of niche products due to insufficient prior knowledge, but also effectively solve the problems of low research and development timeliness and inability to be directly applied under the condition of serious sample skew when using a classical machine learning model for customer recognition.

[0131] In addition, the general customer recognition model in the embodiment has general feature processing capability, which greatly improves the implementation efficiency of the customer recognition scheme in specific product marketing. Taking customer behavior data as the starting point, a low-dimensional feature vector representation with high discrimination of customer behavior is realized by using a deep learning-based generative model. In subsequent specific product marketing, by calculating the similarity with the reference customer historical behavior, the target customer group and the non-target customer group can be quickly distinguished. By saving the low-dimensional feature vector of customer behavior as a "semi-finished product" of feature engineering in the data warehouse, the time of feature engineering in targeted customer recognition application is greatly compressed, and the implementation efficiency of the overall scheme is significantly improved.

[0132] In addition, in the embodiment, different granularity feature rule extraction can also be realized. By adjusting the preset parameters, the granularity of the corresponding time series can be adjusted to any scale under the condition that the computing power meets the requirements, meeting the needs of different applications.

[0133] As can be seen, through the technical scheme of the present application, the purpose of identifying the target object group is achieved, thereby realizing the technical effect of improving the efficiency of identifying the target object group, and further solving the technical problem in the prior art that the customer recognition model is constructed for a single product and the construction time is too long, which leads to the inability to balance the effectiveness and timeliness of multi-product marketing activities.

[0134] Embodiment 2

[0135] According to an embodiment of the present invention, an embodiment of an object identification device is provided, wherein, Figure 13 This is a schematic diagram of an optional object identification device according to an embodiment of the present invention, such as... Figure 13 As shown, the device includes: an acquisition module 1301 for acquiring transaction behavior data of at least one object; a first processing module 1302 for performing time-series processing on the transaction behavior data to obtain a first time series, wherein the first time series is used to characterize the behavioral characteristics of at least one object; a second processing module 1303 for performing data dimensionality reduction processing on the first time series to obtain a first feature vector; a calculation module 1304 for calculating the similarity between the first feature vector and a second feature vector to obtain a calculation result, wherein the second feature vector is a feature vector of a reference object, and the feature vector of the reference object characterizes the data characteristics corresponding to participation in the target marketing activity; and a determination module 1305 for determining at least one target object from at least one object based on the calculation result, thereby obtaining a target object group composed of at least one target object, wherein the target object group is a potential customer group participating in the target marketing activity.

[0136] It should be noted that the above-mentioned acquisition module 1301, first processing module 1302, second processing module 1303, calculation module 1304 and determination module 1305 correspond to steps S101 to S105 in the above embodiments. The five modules and the corresponding steps implement the same examples and application scenarios, but are not limited to the content disclosed in the above embodiment 1.

[0137] Optionally, the first processing module includes: a third processing module for performing format conversion processing on the transaction behavior data to obtain an initial time series; and a fourth processing module for performing normalization processing on the initial time series to obtain a first time series.

[0138] Optionally, the fourth processing module includes: a fifth processing module for normalizing the discrete features in the initial time series to obtain normalized values ​​of the discrete features; a sixth processing module for normalizing the continuous features in the initial time series to obtain normalized values ​​of the continuous features; and a seventh processing module for concatenating the normalized values ​​of the discrete features and the normalized values ​​of the continuous features to obtain the first time series.

[0139] Optionally, the second processing module includes: an encoding processing module for encoding the first time series based on the first residual network to obtain an initial first feature vector; and a transformation module for generating a first feature vector based on the initial first feature vector.

[0140] Optionally, the apparatus for identifying objects further comprises: a receiving module, configured to receive an object list of reference objects; a first calculating module, configured to obtain a reference time sequence based on the object list; and a second calculating module, configured to determine the second feature vector based on the reference time sequence.

[0141] Optionally, the calculating module comprises: a second calculating module, configured to calculate mathematical expectation values of the first feature vector and the second feature vector respectively to obtain a first expectation value and a second expectation value; and a third calculating module, configured to calculate a similarity between the first expectation value and the second expectation value to obtain the calculation result.

[0142] Optionally, the determining module comprises: a fourth calculating module, configured to input the calculation result into a first clustering model to perform clustering calculation to obtain mathematical expectation values of the plurality of feature vectors; an eighth processing module, configured to perform sorting processing on the mathematical expectation values of the plurality of feature vectors, and determine a feature vector corresponding to a maximum mathematical expectation value as a target feature vector; and a first determining module, configured to determine at least one target object corresponding to the target feature vector from the at least one object to obtain a target object group.

[0143] Optionally, the apparatus for identifying objects further comprises: a first obtaining module, configured to obtain object numbers corresponding to at least one target object in the target object group to generate a target object list; and a sending module, configured to send the target object list to a preset platform, wherein the preset platform is configured to recommend a target marketing activity to objects corresponding to the numbers based on the target object list.

[0144] Embodiment 3

[0145] According to another aspect of the embodiments of the present application, a computer readable storage medium is provided, and the computer readable storage medium stores a computer program, wherein the computer program is configured to execute the above-mentioned method for identifying objects when running.

[0146] Embodiment 4

[0147] According to another aspect of the embodiments of the present application, an electronic device is provided, and the electronic device comprises: Figure 14 is a schematic diagram of an optional electronic device according to an embodiment of the present application, as Figure 14As shown, the electronic device includes one or more processors; a memory for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement a program for running, wherein the program is configured to perform the above-mentioned method of identifying an object when running. The processor implements the following steps when executing the program: obtaining transaction behavior data of at least one object; performing time series processing on the transaction behavior data to obtain a first time series, wherein the first time series is used to represent the behavior characteristics of the at least one object; performing data dimension reduction processing on the first time series to obtain a first feature vector; calculating the similarity between the first feature vector and a second feature vector to obtain a calculation result, wherein the second feature vector is a feature vector of a reference object, and the feature vector of the reference object represents data characteristics participating in a target marketing activity; and determining at least one target object from the at least one object according to the calculation result to obtain a target object group consisting of the at least one target object, wherein the target object group is a potential customer group participating in the target marketing activity.

[0148] Optionally, the processor further implements the following steps when executing the program: performing format conversion processing on the transaction behavior data to obtain an initial time series; and performing normalization processing on the initial time series to obtain the first time series.

[0149] Optionally, the processor further implements the following steps when executing the program: performing normalization processing on discrete features in the initial time series to obtain normalized values of the discrete features; performing normalization processing on continuous features in the initial time series to obtain normalized values of the continuous features; and performing splicing processing on the normalized values of the discrete features and the normalized values of the continuous features to obtain the first time series.

[0150] Optionally, the processor further implements the following steps when executing the program: performing encoding processing on the first time series based on a first residual network to obtain an initial first feature vector; and generating the first feature vector based on the initial first feature vector.

[0151] Optionally, the processor further implements the following steps when executing the program: receiving an object list of a reference object; obtaining a reference time series based on the object list; and determining the second feature vector based on the reference time series.

[0152] Optionally, the processor further implements the following steps when executing the program: respectively calculating mathematical expectation values of the first feature vector and the second feature vector to obtain a first expectation value and a second expectation value; and calculating the similarity between the first expectation value and the second expectation value to obtain a calculation result.

[0153] Optionally, the processor, when executing the program, further implements the following steps: inputting the calculation result into the first clustering model to perform clustering calculation, to obtain mathematical expectation values of the plurality of feature vectors; performing sorting processing on the mathematical expectation values of the plurality of feature vectors, and determining a feature vector corresponding to a maximum mathematical expectation value as a target feature vector; determining at least one target object corresponding to the target feature vector from the at least one object, to obtain a target object group.

[0154] Optionally, the processor, when executing the program, further implements the following steps: obtaining object numbers corresponding to the at least one target object in the target object group, to generate a target object list; and sending the target object list to a preset platform, where the preset platform is configured to recommend the target marketing activity to the objects corresponding to the numbers based on the target object list.

[0155] The device herein can be a server, a PC, a PAD, a mobile phone, etc.

[0156] The above-mentioned embodiment numbers of the application are only for description, and do not represent the advantages and disadvantages of the embodiments.

[0157] In the above-mentioned embodiments of the application, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0158] In several embodiments provided in the present application, it should be understood that the disclosed technical contents can be implemented by other ways. Among them, the above-mentioned device embodiments are only schematic, for example, the division of the units can be a logical function division, and actual implementation can have another division way, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units or modules shown or discussed can be indirect coupling or communication connection through some interfaces, units or modules, which can be electrical or other forms.

[0159] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed to multiple units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0160] In addition, each functional unit in each embodiment of the application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be realized in the form of hardware or in the form of software functional unit.

[0161] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application, essentially or in other words, the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to make a computer device (which can be a personal computer, a server or a network device, etc.) execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: a U disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store program codes.

[0162] The above is only the preferred embodiment of the present application, it should be pointed out that, for those skilled in the art, without departing from the principles of the present application, can make a number of improvements and refinements, these improvements and refinements should also be considered as the protection scope of the present application.

Claims

1. A method of identifying an object, characterized by, The method comprises the following steps: acquiring transaction behavior data of at least one object; performing time series processing on the transaction behavior data to obtain a first time series, wherein the first time series is used to represent the behavior characteristics of the at least one object; performing data dimension reduction processing on the first time series to obtain a first feature vector; calculating the similarity between the first feature vector and a second feature vector to obtain a calculation result, wherein the second feature vector is a feature vector of a reference object, and the feature vector of the reference object represents data characteristics of participating in a target marketing activity; determining at least one target object from the at least one object according to the calculation result to obtain a target object group composed of the at least one target object, wherein the target object group is a potential customer group participating in the target marketing activity; determining at least one target object from the at least one object according to the calculation result to obtain a target object group composed of the at least one target object, comprising: inputting the calculation result into a first clustering model for clustering calculation to obtain a mathematical expectation value of a plurality of feature vectors; performing sorting processing on the mathematical expectation value of the plurality of feature vectors, and determining a feature vector corresponding to a maximum mathematical expectation value as a target feature vector; determining the at least one target object corresponding to the target feature vector from the at least one object to obtain the target object group.

2. The method of claim 1, wherein, performing time series processing on the transaction behavior data to obtain a first time series, comprising: performing format conversion processing on the transaction behavior data to obtain an initial time series; performing normalization processing on the initial time series to obtain the first time series.

3. The method of claim 2, wherein, performing normalization processing on the initial time series to obtain the first time series, comprising: performing normalization processing on discrete features in the initial time series to obtain a normalized value of the discrete features; performing normalization processing on continuous features in the initial time series to obtain a normalized value of the continuous features; performing splicing processing on the normalized value of the discrete features and the normalized value of the continuous features to obtain the first time series.

4. The method of claim 3, wherein, performing data dimension reduction processing on the first time series to obtain a first feature vector, comprising: performing encoding processing on the first time series based on a first residual network to obtain an initial first feature vector; generating the first feature vector based on the initial first feature vector.

5. The method of claim 1, wherein, Before calculating the similarity between the first feature vector and a second feature vector to obtain a calculation result, the method further comprises: receiving an object list of a reference object; acquiring a reference time series based on the object list; determining the second feature vector based on the reference time series.

6. The method of claim 5, wherein, calculating the similarity between the first feature vector and a second feature vector to obtain a calculation result, comprising: respectively calculating mathematical expectation values of the first feature vector and the second feature vector to obtain a first expectation value and a second expectation value; calculating the similarity between the first expectation value and the second expectation value to obtain the calculation result.

7. The method of claim 1, wherein, After determining at least one target object from the at least one object according to the calculation result, a target object group composed of the at least one target object is obtained, and the method further comprises: obtaining object numbers corresponding to the at least one target object in the target object group, and generating a target object list; sending the target object list to a preset platform, wherein the preset platform is configured to recommend the target marketing activity to objects corresponding to the object numbers based on the target object list.

8. An apparatus for recognizing an object, the apparatus comprising: Comprise: an acquisition module configured to acquire transaction behavior data of at least one object; a first processing module configured to perform time series processing on the transaction behavior data to obtain a first time series, wherein the first time series is used to represent behavior characteristics of the at least one object; a second processing module configured to perform data dimension reduction processing on the first time series to obtain a first feature vector; a calculation module configured to calculate a similarity between the first feature vector and a second feature vector to obtain a calculation result, wherein the second feature vector is a feature vector of a reference object, and the feature vector of the reference object represents data characteristics of participating in a target marketing activity; a determination module configured to determine at least one target object from the at least one object according to the calculation result to obtain a target object group composed of the at least one target object, wherein the target object group is a potential customer group participating in the target marketing activity; the determination module comprises: a fourth calculation module configured to input the calculation result into a first clustering model to perform clustering calculation to obtain mathematical expectation values of a plurality of feature vectors; an eighth processing module configured to perform sorting processing on the mathematical expectation values of the plurality of feature vectors, and determine a feature vector corresponding to a maximum mathematical expectation value as a target feature vector; and a first determination module configured to determine at least one target object corresponding to the target feature vector from the at least one object to obtain the target object group.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, wherein the computer program is configured to execute the method for identifying an object in any one of claims 1 to 7 when running.

10. An electronic device, comprising: The electronic device comprises one or more processors; a memory configured to store one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement a program for running, wherein the program is configured to execute the method for identifying an object in any one of claims 1 to 7 when running.

Images