An object association recognition method, device and storage medium

Abstract

The application provides an object association recognition method and device and a storage medium, relates to the field of artificial intelligence, and the method comprises the following steps: obtaining target attribute features and target operation sequence data of a to-be-recognized object group; performing sequence pattern matching on the target operation sequence data based on a preset sequence pattern library to obtain a target operation sequence pattern of the to-be-recognized object group; performing sequence correlation processing on target operation features corresponding to the target operation sequence data by using a sequence correlation mining network of a preset relationship recognition model to obtain target sequence correlation features of the to-be-recognized object group; performing association relationship recognition on the to-be-recognized object group based on the target attribute features, the target operation features, the target operation sequence pattern and the target sequence correlation features by using an object group classification network of the preset relationship recognition model to obtain an association relationship category of the to-be-recognized object group. The application can effectively improve the generalization ability and timeliness of object association recognition.

Description

TECHNICAL FIELD

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

[0002] With the development of Internet technology, association relationship recognition technology has become a necessary link in the construction of object relationship chains, and can be applied to various scenarios such as relationship closeness recognition and relationship network construction. However, some current association relationship recognition methods need to manually set a large number of relationship rules and periodically maintain a relationship rule library to ensure the accuracy and timeliness of association relationship recognition. In various industrial application scenarios with flexible data changes, it is difficult to achieve automation, efficient generalization, and is easily affected by prior knowledge and subjective factors.

[0003] Therefore, it is necessary to provide an improved object association recognition scheme to solve the problems existing in the prior art and improve the generalization ability and timeliness of object association recognition. SUMMARY

[0004] The present application provides an object association recognition method, device and storage medium, which can effectively improve the generalization ability and timeliness of object association recognition.

[0005] In one aspect, the present application provides an object association recognition method, which comprises:

[0006] Obtaining target attribute features and target operation sequence data of a to-be-recognized object group; the target operation sequence data is data representing object interaction operations of the to-be-recognized object group;

[0007] Based on a preset sequence pattern library, performing sequence pattern matching on the target operation sequence data to obtain a target operation sequence pattern of the to-be-recognized object group;

[0008] Using a sequence correlation mining network of a preset relationship recognition model, performing sequence correlation processing on target operation features corresponding to the target operation sequence data to obtain target sequence correlation features of the to-be-recognized object group;

[0009] Using an object group classification network of the preset relationship recognition model, performing association relationship recognition on the to-be-recognized object group based on the target attribute features, the target operation features, the target operation sequence pattern and the target sequence correlation features to obtain an association relationship category of the to-be-recognized object group.

[0010] Another aspect provides an object association recognition device, which comprises:

[0011] The feature acquisition module is configured to acquire target attribute features of a to-be-identified object group and target operation sequence data, wherein the target operation sequence data is data representing object interaction operations of the to-be-identified object group.

[0012] The sequence pattern matching module is configured to perform sequence pattern matching on the target operation sequence data based on a preset sequence pattern library, to obtain a target operation sequence pattern of the to-be-identified object group.

[0013] The sequence correlation mining module is configured to perform sequence correlation processing on target operation features corresponding to the target operation sequence data by using a sequence correlation mining network of a preset relationship recognition model, to obtain target sequence correlation features of the to-be-identified object group.

[0014] The object association recognition module is configured to perform association relationship recognition on the to-be-identified object group based on the target attribute features, the target operation features, the target operation sequence pattern and the target sequence correlation features by using an object group classification network of the preset relationship recognition model, to obtain an association relationship category of the to-be-identified object group.

[0015] In another aspect, an object association recognition device is provided, which includes a processor and a memory, the device includes a processor and a memory, and the memory stores at least one instruction or at least one program, which is loaded and executed by the processor to implement the object association recognition method as described above.

[0016] In another aspect, a computer-readable storage medium is provided, which stores at least one instruction or at least one program, which is loaded and executed by a processor to implement the object association recognition method as described above.

[0017] In another aspect, a server is provided, which includes a processor and a memory, the device includes a processor and a memory, and the memory stores at least one instruction or at least one program, which is loaded and executed by the processor to implement the object association recognition method as described above.

[0018] The object association recognition method, device, equipment, storage medium and server provided by the present application have the following technical effects:

[0019] The application obtains target attribute features and target operation sequence data of a to-be-identified object group, performs sequence pattern matching on the target operation sequence data based on a preset sequence pattern library to obtain a target operation sequence pattern of the to-be-identified object group, and can realize automatic mining of the operation sequence pattern, does not need to set a large number of artificial rules, and has better generalization ability and timeliness. Then, the target operation sequence data is subjected to sequence correlation processing by using a sequence correlation mining network of a preset relationship identification model to obtain target sequence correlation features of the to-be-identified object group, and the to-be-identified object group is subjected to relationship identification based on the target attribute features, the target operation features, the target operation sequence pattern and the target sequence correlation features by using an object group classification network of the preset relationship identification model, and then a relationship category of the to-be-identified object group is obtained. The application can comprehensively identify the relationship based on various features, thereby improving the accuracy of the relationship identification result. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, and the advantages thereof, a brief introduction will be given to the drawings needed in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor based on these drawings.

[0021] Figure 1 is a schematic diagram of an application environment provided by an embodiment of the present application;

[0022] Figure 2 is a flowchart of an object relationship identification method provided by an embodiment of the present application;

[0023] Figure 3 is a flowchart of a target operation sequence data acquisition method provided by an embodiment of the present application;

[0024] Figure 4 is a flowchart of a target attribute feature acquisition method provided by an embodiment of the present application;

[0025] Figure 5 is a flowchart of a relationship identification method provided by an embodiment of the present application;

[0026] Figure 6 is a flowchart of a training method of a preset relationship identification model provided by an embodiment of the present application;

[0027] Figure 7 is a model structure diagram of an initial relationship identification model provided by an embodiment of the present application;

[0028] Figure 8is a structural schematic diagram of an object correlation recognition device provided by an embodiment of the present application.

[0029] Figure 9 is a hardware structure block diagram of a server of an object correlation recognition method provided by an embodiment of the present application.

[0030] Figure 10 is a structural schematic diagram of a blockchain system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0031] Cloud computing refers to a delivery and use model of IT infrastructure, which refers to obtaining required resources in a manner of on-demand and easy expansion through a network; broad-sense cloud computing refers to a delivery and use model of service, which refers to obtaining required service in a manner of on-demand and easy expansion through a network. Such service can be IT and software, Internet related, or other services. Cloud computing is a product of the development of grid computing, distributed computing, parallel computing, utility computing, network storage technologies, virtualization, load balancing and other traditional computer and network technologies.

[0032] The technical scheme of the embodiment of the present application relates to the field of big data technology of cloud computing. Big data refers to a data set that cannot be captured, managed and processed within a certain time range by conventional software tools, and is a mass, high growth rate and diversified information asset that needs a new processing mode to have stronger decision-making, insight discovery and process optimization capabilities.

[0033] With the advent of the cloud era, big data has attracted more and more attention. Big data needs special technology to effectively process a large amount of data over time. The technology suitable for big data includes large-scale parallel processing database, data mining, distributed file system, distributed database, cloud computing platform, Internet and scalable storage system. Data mining refers to the process of revealing hidden, previously unknown and potentially valuable information from a large amount of data in a database. Accurate object correlation recognition, i.e. related correlation relationship network construction, can provide an effective knowledge base for data mining of correlation relationship data.

[0034] With reference to the drawings and the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0035] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to the process, method, product, or device.

[0036] The various data collection, processing and application involved in the present application are performed with the consent and permission of the data owner, and the above operations comply with the relevant legal regulations.

[0037] Before the embodiments of the present application are further described in detail, the terms and phrases involved in the embodiments of the present application are explained, and the terms and phrases involved in the embodiments of the present application are applicable to the following explanations.

[0038] Recurrent Neural Network (RNN) is a kind of recursive neural network that takes sequence data as input and performs recursion in the evolution direction of the sequence, and all nodes (recurrent units) are connected in a chain form.

[0039] Long Short-Term Memory networks (LSTM) is a kind of time recurrent neural network, which is specially designed to solve the long-term dependence problem existing in general RNN (recurrent neural network), and all RNNs have a chain form of repeated neural network modules.

[0040] Artificial Intelligence (AI) is the theory, method, technology and application system of using digital computer or machine controlled by digital computer to simulate, extend and expand human intelligence, perceive environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making.

[0041] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software level technology. Artificial intelligence basic technology generally includes technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology and machine learning / deep learning, etc.

[0042] In recent years, with the research and progress of artificial intelligence technology, artificial intelligence technology has been widely applied in many fields. The scheme provided by the embodiments of the present application relates to machine learning / deep learning technology of artificial intelligence, which is specifically explained by the following embodiments.

[0043] Please refer to Figure 1 , Figure 1 is a schematic diagram of an application environment provided by the embodiments of the present application, as Figure 1 shown, the application environment can at least include a server 01 and a terminal 02. In actual application, the server 01 and the terminal 02 can be directly or indirectly connected through wired or wireless communication mode to realize the interaction between the terminal 02 and the server 01, which is not limited in the present application.

[0044] In the embodiments of the present application, the server 01 can be a stand-alone physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and basic cloud computing services such as big data and artificial intelligence platforms. Specifically, the server can include a physical device, which can specifically include a network communication unit, a processor, a memory, and the like, or can include software running in the physical device, which can specifically include an application program and the like. In the embodiments of the present application, the server 01 can be configured to provide an association relationship identification service according to the target operation sequence data and target attribute features of the to-be-identified object group in response to an object association identification request sent by the terminal 02. Specifically, the server 01 can also be configured to provide a training service of an initial relationship identification model based on a sample training set. Specifically, the server 01 can also be configured to provide a storage service, such as storing a preset sequence mode library.

[0045] In the embodiments of the present application, the terminal 02 can include a smart phone, a desktop computer, a tablet computer, a notebook computer, a digital assistant, an augmented reality (AR) / virtual reality (VR) device, a smart wearable device, a vehicle-mounted terminal device, and the like, or can include software running in the physical device, such as an application program and the like. Specifically, the terminal 02 can be configured to send an object association identification request and send relevant data of the object group to the server 01, such as sending operation data (such as behavior data) and object attribute data (such as attribute data) of each object in the object group to the server 01.

[0046] In addition, it should be noted that, Figure 1 The above-mentioned only shows an application environment of the object association identification method, and the application environment can include more or fewer nodes, which is not limited in the present application.

[0047] In some embodiments of the present application, the association relationship between the objects in the object group can be a social relationship. The social relationship is the bond of connection and interaction between people in a social network, and is also the basis of information dissemination in a social network. With the rapid development of online social applications and media, object association identification technology has become a necessary link for social portrait and relationship chain construction, and further realizes application in multiple scenarios, such as application in relationship closeness identification between users, and recommendation and precision marketing based on social network relationship.

[0048] The current social relationship recognition method mainly includes the method of setting rules manually and mining social network relationship based on graph model. However, the method of setting rules manually to mine social network relationship needs to formulate a large number of mapping relationship type rules based on prior knowledge, and regularly maintain the relationship rule library to ensure the accuracy and timeliness of the relationship mining. In various industrial application scenarios with flexible data changes, complete automation cannot be achieved. The method of mining relationship network based on graph model first needs to use network representation learning method to construct network model by taking relationship object as node. With the increase of the order of magnitude of the relationship object to be mined, the network complexity also increases dramatically, and the time of training network model is difficult to meet the actual application demand. The object association recognition method provided in the application can effectively solve the above problems.

[0049] The object association recognition method provided in the application is introduced based on the above application environment, and is applied to the server side. Figure 2 It is a flowchart of the object association recognition method provided in the embodiment of the application. The present specification provides method operation steps such as embodiments or flowcharts, but more or fewer operation steps can be included based on conventional or non-creative labor. The order of steps listed in the embodiment is only one of the many execution orders, and does not represent the only execution order. In actual system or server product execution, the method order shown in the embodiment or the drawing can be executed in sequence or in parallel (for example, in a parallel processor or multi-threaded processing environment). Specifically, as shown in Figure 2 The method can include:

[0050] S201: obtaining target attribute features and target operation sequence data of a to-be-recognized object group; the target operation sequence data is data representing object interaction operation of the to-be-recognized object group.

[0051] In the embodiment of the application, the association relationship category represents the association relationship between objects, and a plurality of association relationship categories can be preset. Correspondingly, when the object is a user account, such as a user social application account, the preset association relationship category can be specifically a preset social relationship category, which can include but is not limited to lover, parent-child, mother-child, colleague, friend and husband-wife relationship, etc. Specifically, the server can respond to the object association recognition request sent by the terminal to perform the association relationship recognition operation, or can also be based on the preset association relationship mining instruction to directly perform the association relationship recognition operation after receiving the related data of the object group.

[0052] In the embodiments of the present application, the to-be-identified object group can include at least two objects, the target attribute feature is generated based on the object portrait feature of each object in the to-be-identified object group, and the target attribute feature can represent the portrait feature of the to-be-identified object group. The target operation sequence data can be generated based on the object interaction operation data in a preset time period of the to-be-identified object group, wherein the preset time period can be any set time range. For example, when the object is a user account, the object interaction operation data can be the interaction behavior data corresponding to each user account in the to-be-identified object group, and correspondingly, the target operation sequence data is the behavior sequence data generated based on the interaction behavior operation data in the preset time period of the to-be-identified object group. Specifically, the preset time periods corresponding to different preset association relationship categories can be the same or different, and the same to-be-identified object group can also correspond to target operation sequence data of multiple different preset time periods. For example, the object group with a lover relationship usually has object interaction operations such as red envelope, transfer, or almost simultaneous publishing of a friend circle at multiple key time nodes such as Valentine's Day (Valentine's Day, White Valentine's Day, or Qixi Festival), birthdays of both parties, and Christmas, and generates corresponding operation data. The preset time period can be set as a time range of a period of time before and after the key time points, such as D days before and after Valentine's Day, i.e., [Valentine's Day-D, Valentine's Day+D].

[0053] In the embodiments of the present application, a method for acquiring target operation sequence data is also provided. Please refer to Figure 3 , Figure 3 A flowchart of a method for acquiring target operation sequence data provided by the embodiments of the present application is shown, which can specifically include the following steps:

[0054] S301: Obtain object interaction operation data and corresponding interaction operation time information in a preset time period of a to-be-identified object group.

[0055] In actual application, many operation data can be generated when each object performs an interaction operation. For example, when the object is a user account, many behavior data can be generated in the Internet activities of each user account. The operation data in one or more preset time periods of each object in the to-be-identified object group can be acquired, and the object interaction operation data between each object in the object group can be obtained by screening, and the interaction operation time information corresponding to the object interaction operation data is also acquired, such as the object interaction operation data corresponding to the interaction operations of sending a red envelope, transferring, publishing a friend circle near a key time node, and inviting to follow a theme public account and sharing a theme link, and the interaction operation time information of each interaction operation.

[0056] S303: Perform time sorting and data splicing processing on the object interaction operation data in the preset time period according to the interaction operation time information, and obtain target operation sequence data.

[0057] In actual applications, the timing of the object interaction operation data can be determined based on the interaction operation time information, and the object interaction operation data can be sorted and data spliced based on the timing to construct the target operation sequence data. For example, the object group to be identified includes user account A and user account B. At t0, user account B receives and receives 520 yuan transferred by user account A. User account A and user account B post a circle of friends at t1 and t2 respectively. At t3, user account A likes the circle of friends of user account B. At t4, user account A invites user account B to follow the wedding photography public number, and user account B follows the public number at t5. According to the timing of t0-t5, the above object interaction operation data can be sorted.

[0058] In some cases, it is also necessary to construct a mapping interaction operation label of the object interaction operation data to mark the operation sequence of the object interaction operation data. For example, for the object interaction operation data "user account B receives and receives 520 yuan transferred by user account A", for user account A, the mapping interaction operation label is marked as "A-520 transfer-B", and for user account B, the mapping interaction operation label is marked as "B-receive transfer 520-A", and the corresponding interaction operation is "transfer 520"; for the object interaction operation data "at t4, user account A invites user account B to follow the wedding photography public number, and user account B follows the public number at t5", for user account A, the mapping interaction operation label is marked as "A-invite to follow the wedding photography public number-B", and for user account B, the mapping interaction operation label is marked as "B-accept the invitation to follow the wedding photography public number-B", and the interaction operation is "follow the wedding photography public number".

[0059] In some cases, similar interaction operations can also be identified as the same type to perform uniform identifier identification. For example, transfer 520, transfer 5.20, transfer 52.0, transfer 99, transfer 1314, transfer 13.14, transfer 131.4 and transfer 999 have similar emotions, and the same type of transfer interaction operation can be identified as a uniform identifier.

[0060] In specific embodiments, the same interactive operation can also be identified by an operation identifier, for example, one or more characters can be used for unified identification, such as similar interactive operations such as transfer 520 can be identified as m, the interactive operation of publishing a circle of friends within a preset time period can be identified as kp, the interactive operation of following the same public number can be identified as u, and the like. Correspondingly, the above data splicing processing can be specifically: identifying the interactive operation corresponding to the object interactive operation data within the preset time period by an interactive operation identifier, and taking the time sequence obtained by time sorting as the data splicing order, splicing the interactive operation identifiers corresponding to the object interactive operation data to obtain the target identifier string representing the target operation sequence data. For example, the time nodes and interactive operation identifiers corresponding to the object interactive operation data of the to-be-identified object group are “t0 time / b, t5 time / c, t6 time / a, t 10 time / f, t 12 time / g, t 15 time / h”, and the corresponding target operation sequence data is represented by the target identifier string bcafgh.

[0061] It can be understood that different preset association relationship categories can correspond to different object interactive operation data, mapping interactive operation labels and interactive operation identifiers, and the like. The target operation sequence data corresponding to various preset association relationship categories can be constructed by using similar methods, which will not be enumerated herein.

[0062] In actual applications, the target operation sequence data can be further vectorized to obtain target operation features. Specifically, the target operation sequence data can be vectorized and encoded, including but not limited to one-hot encoding, to obtain target operation features.

[0063] Specifically, each interactive operation can have a corresponding code, such as the code corresponding to the operation of forwarding a circle of friends being [1, 0, 0, 0, 0, 0], the code corresponding to the operation of circle of friends liking being [0, 1, 0, 0, 0, 0], the code corresponding to the operation of forwarding an e-commerce shopping link being [0, 0, 1, 0, 0, 0], and the code corresponding to the operation of clicking to purchase being [0, 0, 0, 1, 0, 0], and the like. Based on the correspondence between the interactive operation and the code, the target operation features of the target operation sequence data can be obtained. For example, if the target operation sequence data of the to-be-identified object group is bca, the code corresponding to a is [1, 0, 0,...], the code corresponding to b is [0, 1, 0,...], and the code corresponding to c is [0, 0, 1,...], then the target operation features are [1, 0, 0,..., 0, 1, 0,..., 0, 0, 1,...].

[0064] In the embodiments of the present application, a method for obtaining target attribute features is also provided. Please refer to Figure 4 , Figure 4A flowchart of a method for obtaining target attribute features is shown, which can include the following steps:

[0065] S401: Obtain object attribute data of each object in the object group to be identified.

[0066] S403: Generate basic object information of each object based on the object attribute data.

[0067] S405: Perform vectorization processing and feature splicing processing on the basic object information of each object to obtain target attribute features of the object group to be identified.

[0068] In actual applications, the target attribute features are generated based on object portrait features of each object in the object group to be identified, and the vectorization processing on the basic object information can obtain the object portrait features. For example, when the object is a user account, the object portrait features can include but are not limited to user's basic portrait, wealth, life stage, consumption preference, etc. The basic portrait features can be, for example, user's age, gender, residence, etc., the wealth features can be, for example, asset score, consumption ability, risk tolerance, etc., the life stage features can be, for example, whether the user is married, has children, and has a house, etc., and the consumption preference features can be, for example, consumption preference merchants, consumption preference amounts, and consumption preference times, etc. In some cases, the object portrait features can also include the object's network address, etc. Specifically, the object attribute data can include but is not limited to various attribute data related to the above features, and the specific categories and obtaining methods of the object attribute data can be the same as the prior art, which will not be described here.

[0069] In specific embodiments, the object portrait features obtained by the vectorization processing are spliced to obtain the target attribute features. In some cases, the target operation features and the target attribute features can be dimensionally spliced to obtain target object group features. For example, the object portrait features of user account A in the object group to be identified are [0.2, -0.4, 0.09, 0.54, -2.5], the object portrait features of user account B are [0.7, -0.01, 0.3, 0.4, 9], and the target attribute features are [0.2, -0.4, 0.09, 0.54, -2.5, 0.7, -0.01, 0.3, 0.4, 9]. If the target operation features of the object group to be identified are [1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0], then the target object group features obtained by dimensionally splicing are [0.2, -0.4, 0.09, 0.54, -2.5, 0.7, -0.01, 0.3, 0.4, 9, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0].

[0070] S203: Perform sequence pattern matching on the target operation sequence data based on the preset sequence pattern library to obtain a target operation sequence pattern of the to-be-identified object group.

[0071] In actual application, the preset sequence pattern library pre-stores reference sequence patterns corresponding to each preset correlation relationship category, and the reference sequence patterns are obtained by performing sequence pattern mining on reference operation sequence data of a sample object group. Specifically, in the training process of the preset relationship identification model, the training sample set includes sample object groups belonging to various preset correlation relationship categories and reference operation sequence data of each sample object group. The reference operation sequence data of sample object groups of the same category is subjected to sequence pattern mining to obtain a reference sequence pattern corresponding to the preset correlation relationship category, and then reference sequence patterns corresponding to each preset correlation relationship category are obtained. It can be understood that each preset correlation relationship category can correspond to one or more reference sequence patterns. For example, the reference sequence patterns corresponding to the correlation relationship category 1 can include "bca, ca, agh, cagh…", wherein the English characters are the aforementioned interactive operation identifiers, and bca, ca, agh and cagh are each a reference sequence pattern. It can be understood that the reference sequence patterns corresponding to different preset correlation relationship categories can be completely the same, partially the same or completely different.

[0072] For example, when the object is a user account, the preset correlation relationship category is a preset social relationship category, the reference operation sequence data is reference behavior sequence data generated based on reference interaction behavior data within a preset time period of a sample object group, and the reference sequence pattern is obtained by performing sequence pattern mining on reference behavior sequence data corresponding to the sample object group.

[0073] Correspondingly, step S203 can include:

[0074] S2031: Determine, as the matched sequence pattern, the reference sequence pattern in the preset sequence pattern library for the sub-sequence of the target operation sequence data.

[0075] S2033: Generate the target operation sequence pattern according to the matched sequence pattern.

[0076] Specifically, the specific process of sequence pattern matching can be: searching for the sub-sequence of the target operation sequence data in all reference sequence patterns in the preset sequence pattern library, determining, as the matched sequence pattern, the reference sequence pattern of the sub-sequence of the target operation sequence data that is found, and then generating the target operation sequence pattern. In some embodiments, the target operation sequence pattern can be a simple combination or a weighted combination of the matched sequence pattern.

[0077] In some embodiments, the preset sequence pattern library further comprises a sequence pattern support corresponding to each reference sequence pattern, the sequence pattern support representing a frequency of occurrence of the reference sequence pattern in the corresponding preset association relationship category in the preset sequence pattern library. In an embodiment, the sequence pattern support is equal to a ratio of a number of samples in the sample object group corresponding to the corresponding preset association relationship category that appear the reference sequence pattern to a total number of samples in the sample object group corresponding to the preset association relationship category.

[0078] Correspondingly, before step S2033, there can be further included S2032: obtaining a sequence pattern support corresponding to each matched sequence pattern. Correspondingly, step S2033 can be: determining a weight coefficient of each matched sequence pattern based on the sequence pattern support, and performing a weighted operation on each matched sequence pattern based on the weight coefficient to obtain the target operation sequence pattern.

[0079] S205: using the sequence correlation mining network of the preset relationship recognition model to perform sequence correlation processing on the target operation feature corresponding to the target operation sequence data to obtain a target sequence correlation feature of the to-be-recognized object group.

[0080] In the embodiments of the present application, the interactive operations of the same object group have certain association and sequence correlation, and the corresponding operation features also retain the correlation, for example, user account A of a couple relationship pair forwards a friend circle of user account B at a key time node, user account A likes the friend circle of user account B, user account B forwards a certain e-commerce shopping link to user account A, user account A clicks the e-commerce shopping link and makes a purchase, etc. Based on the above object interactive operation data, operation sequence data can be generated, and then operation features retaining the sequence correlation between the above object interactive operation data can be generated.

[0081] In actual applications, the sequence correlation mining network can be a network layer constructed based on a recurrent neural network, for example, a network layer constructed based on LSTM, etc. The target operation feature is input into the sequence correlation mining network to output the corresponding target sequence correlation feature.

[0082] S207: using the object group classification network of the preset relationship recognition model to perform association relationship recognition on the to-be-recognized object group based on the target attribute feature, the target operation feature, the target operation sequence pattern and the target sequence correlation feature to obtain an association relationship category of the to-be-recognized object group.

[0083] In the embodiments of the present application, the preset relationship identification model is a model obtained by performing association relationship identification training on an initial relationship category model based on reference attribute features and reference operation sequence data of a plurality of sample object groups carrying association relationship category labels. In some embodiments, the initial relationship identification model is a multi-classification model constructed based on a recurrent neural network and a margin loss function, for example, a multi-classification model constructed based on LSTM and a margin softmax loss function.

[0084] Based on the above partial or all embodiments, in the embodiments of the present application, the object group classification network includes a feature encoding layer and a similarity calculation layer, please refer to Figure 5 , Figure 5 A flowchart of a method for identifying an association relationship provided by the embodiments of the present application is shown, and correspondingly, step S207 can include:

[0085] S2071: input the target attribute features, the target operation features, the target sequence correlation features and the target sequence pattern features corresponding to the target sequence pattern features into the feature encoding layer, perform encoding processing to obtain target encoding features of the object group to be identified.

[0086] In actual applications, the feature encoding layer is used for feature extraction on the input features to obtain the target encoding features. As described above, the preset relationship identification model is a multi-classification model, and object groups of the same preset association relationship category belong to the same classification output by the preset relationship identification model. In an embodiment, the preset relationship identification model is a multi-classification model constructed and trained based on Bi-LSTM and am-softmax loss function. Correspondingly, the expression of the feature encoding layer is shown in the following formula (I):

[0087]

[0088] In the application of association relationship identification, x represents the input features based on the target attribute features, the target operation features, the target sequence correlation features and the target sequence pattern features, and y represents the output target encoding features.

[0089] In some embodiments, the target attribute feature, the target operation feature, the target sequence correlation feature, and the target sequence pattern feature can be respectively input into a feature encoding layer; in other embodiments, the target attribute feature, the target operation feature, the target sequence correlation feature, and the target sequence pattern feature can also be spliced, such as dimension splicing, to generate object group spliced features of the to-be-identified object group, and then the object group spliced features are input into the feature encoding layer for encoding processing. For example, in an embodiment, the target attribute feature of the to-be-identified object group is [0.2, -0.4, 0.09, 0.54, -2.5, 0.7, -0.01, 0.3, 0.4, 9], the target operation feature is [1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0], the target sequence correlation feature is [0.1, -0.35, 0.578, 0.982…], and the target sequence pattern feature is [0.215, 0.47, 1.55, -0.588…], and the object group spliced features obtained based on a feature splicing manner can be [0.215, 0.47, 1.55, -0.588…, 0.2, -0.4, 0.09, 0.54, -2.5, 0.7, -0.01, 0.3, 0.4, 9, 0.1, -0.35, 0.578, 0.982…]. In an embodiment, after the object group spliced features are input into the feature encoding layer as x, the output target encoding features y are [0.579, 1.732, 0.9232, -0.2577…].

[0090] S2073: input the target encoding features into a similarity calculation layer to calculate the similarity between the target encoding features and each category encoding feature in the preset encoding feature set, and obtain each category similarity corresponding to the target encoding features.

[0091] In actual applications, the preset encoding feature set includes category encoding features corresponding to each preset association relationship category, and in some cases, the preset association relationship category and the category encoding feature are one-to-one corresponding. In some embodiments, after the preset relationship recognition model is obtained by training the initial relationship recognition model based on the training sample set corresponding to each preset association relationship category, the feature encoding layer of the preset relationship recognition model outputs the category encoding features corresponding to each preset association relationship category, and then the preset encoding feature set is obtained.

[0092] S2075: compare each category similarity with a similarity threshold value respectively to obtain a similarity comparison result.

[0093] In practical applications, the similarity of the target encoding feature and each category encoding feature in the preset encoding feature set is compared and ranked by similarity calculation to obtain a corresponding comparison and ranking result. It can be understood that the greater the similarity value, the higher the possibility that the to-be-identified object group is of the preset association relationship category corresponding to the category encoding feature. In an embodiment, the similarity of each category encoding feature can be ranked according to the similarity comparison result, and the preset association relationship category corresponding to the category encoding feature with the largest similarity value in the ranking is determined as the association relationship category of the to-be-identified object group. Specifically, cosine similarity or the like can be used for the similarity calculation.

[0094] S2077: determining the association relationship category of the to-be-identified object group according to the similarity comparison result.

[0095] Based on the above technical solution, the object association recognition method of the present application converts the classification problem of association relationship recognition into the similarity calculation problem of relationship feature recognition, uses a classification model as a training scheme, uses an object group classification network as a feature extraction model, and realizes comparison and ranking of category encoding features. When a multi-classification model is constructed and trained based on a margin-softmax loss function, the similarity calculation can be completed by introducing the margin loss of a face recognition model, which better makes the training result of the classification model approach the ranking result of the category similarity calculation of the association relationship category, so as to quickly improve the accuracy of the category similarity calculation and further improve the accuracy of the recognition result.

[0096] In some embodiments, step S2077 can include:

[0097] 1) If one or more category similarities are greater than or equal to a similarity threshold, the category encoding feature corresponding to the category similarity with the largest value in the category similarities is determined as the target category encoding feature.

[0098] 2) The preset association relationship category corresponding to the target category encoding feature is taken as the association relationship category of the to-be-identified object group.

[0099] In specific embodiments, if the category similarity of a category encoding feature and the target encoding feature exceeds the similarity threshold, the preset association relationship category corresponding to the category encoding feature is taken as the association relationship category of the to-be-identified object group. If the category similarities of multiple category encoding features and the target encoding feature exceed the similarity threshold, the preset association relationship category corresponding to the category similarity with the largest value is taken as the association relationship category of the to-be-identified object group.

[0100] In some embodiments, step S2077 can further include:

[0101] 3) If the similarity of each category is less than the similarity threshold, it is determined that the association relationship category of the to-be-identified object group is a to-be-updated relationship category.

[0102] 4) Update the preset sequence pattern library based on the target operation sequence pattern of the to-be-identified object group, and update the preset encoding feature set based on the target encoding feature of the to-be-identified object group.

[0103] In specific embodiments, if the category similarity of all category encoding features and the target encoding feature is less than the similarity threshold, it is determined that the association relationship category of the to-be-identified object group is not in the known preset association relationship category, and it is determined as a to-be-updated relationship category, and it is added to the preset association relationship category as a new association relationship category label. In addition, the target operation sequence pattern and the target encoding feature obtained in the identification process of the to-be-updated relationship category are updated to the preset sequence pattern library and the preset encoding feature set, respectively, so as to facilitate subsequent pattern matching and similarity calculation of the to-be-updated relationship category, and realize the update of the training sample set and the model.

[0104] In specific embodiments, after obtaining the new association relationship category label and adding it to the training sample set, in the subsequent object association identification process, new operation sequence patterns and category encoding features can be iteratively mined, thereby expanding the training sample set and realizing the automatic update and iteration of the model, and further improving the recall and accuracy of the model prediction.

[0105] In one embodiment, the category encoding feature A corresponding to the known couple relationship category is [0.579, 1.732, 0.9232, -0.2577…], the target encoding feature B of the to-be-identified object group is [0.472, 0.35, -0.58, 0.9224…], and the cosine distance similarity sim(A, B) of the above two encoding features is calculated. If sim(A, B) exceeds the similarity threshold , and the value of sim(A, B) is the maximum in each category similarity, it is considered that the association relationship category of the to-be-identified object group B is a couple relationship, and if sim(A, B) does not exceed the similarity threshold , it is determined that the to-be-identified object group is not a couple relationship.

[0106] Based on the above part or all embodiments, please refer to Figure 6 , the application also provides a training method of a preset relationship identification model, Figure 6 a flowchart of a training method of a preset relationship identification model provided by the embodiment is shown, the method can include:

[0107] S501: Construct an initial relationship identification model, and the initial relationship identification model includes a sequence correlation mining network and an object group classification network.

[0108] In actual application, the sequence correlation mining network is constructed based on a recurrent neural network. The recurrent neural network can better learn the correlation between the time-based interaction operations of the sample object groups, thereby improving the accuracy of model recognition.

[0109] In one embodiment, refer to Figure 7 , Figure 7 A model structure diagram of an initial relationship identification model is shown. The initial relationship identification model specifically includes a feature embedding layer, a sequence correlation mining network layer based on LSTM, and an object group classification network layer constructed based on LSTM / am-softmax. It should be noted that the model structure in the figure only shows the model structure of an initial relationship identification model. The initial relationship identification model can also include more or fewer network layers, which are not limited in the present application.

[0110] S503: Construct a training sample set based on the reference attribute features of the plurality of sample object groups, the reference operation sequence data, and the corresponding association relationship category labels. The reference operation sequence data is data representing the object interaction operations of the sample object groups.

[0111] In actual application, when the object is a user account, the association relationship category label can include, but is not limited to, a lover relationship label, a parent-child relationship label, a mother-child relationship label, a colleague relationship label, a friend relationship label, and a husband-wife relationship label, etc. In some embodiments, the training sample set can include an association relationship category label standard database <association relationship, object group>. Specifically, for each association relationship category label, N sample object groups are collected as training samples, and then the standard database <association relationship, object group> is constructed. The association relationship category label can be marked with a category id, such as 0, 1, 2, etc. The positive sample corresponding to each association relationship category label is the sample object group of the association relationship category, and the negative sample is the sample object group of other association relationship categories. The structure of the <association relationship, object group> standard database can be shown in Table 1 as follows:

[0112] Table 1

[0113]

[0114] In actual application, the corresponding reference attribute features can be constructed for each association relationship category. Similar to the foregoing target attribute features, the reference attribute features can include the object portrait features of the sample objects in the sample object groups. The object portrait features of the sample objects are obtained by vectorizing the basic object information, and the corresponding acquisition method is similar to the foregoing content, which will not be repeated here.

[0115] In practical applications, each association relationship category corresponds to multiple object portrait features, such as age, gender, place of residence, and wealth features such as asset score, consumption ability, and risk tolerance. Each object portrait feature can be preset with a preset portrait condition. In the process of obtaining the reference attribute feature, for each association relationship category, the object portrait features of the sample object group need to be preprocessed. The specific process can be: for each object portrait feature in each association relationship category, the total feature data missing amount corresponding to the object portrait feature that does not meet the corresponding preset portrait condition is counted; if the total feature data missing amount of any object portrait feature is less than the corresponding missing value filtering threshold, a feature filling value is determined based on the feature attribute of the object portrait feature; the object portrait feature that does not meet the corresponding preset portrait condition is replaced with the corresponding feature filling value. If the total feature data missing amount of any object portrait feature is greater than or equal to the corresponding missing value filtering threshold, the object portrait feature is discarded, i.e., the object portrait feature is not used as a feature for generating a reference attribute feature.

[0116] In one embodiment, if the feature attribute is discrete, a preset constant value can be used as the feature filling value, and if the feature attribute is continuous, the feature mean value corresponding to the object portrait feature can be used as the feature filling value.

[0117] The preset portrait condition can be set according to the needs of the application scenario, and a corresponding preset portrait condition can be set for each object portrait feature. When the attribute feature of a sample object does not meet the corresponding preset portrait condition, it is determined that the attribute feature of the sample object is missing, and is recorded as a feature data missing amount.

[0118] Specifically, the missing value filtering threshold can be determined based on the total feature data amount and a preset proportion value of the object portrait feature. The preset proportion value can be set based on the application scenario, and the preset proportion values corresponding to different object portrait features can be the same or different. Accordingly, the calculation expression of the missing value filtering threshold can be: missing value filtering threshold = total feature data amount X preset proportion value.

[0119] For example, the sample object is a sample user account, the number of sample object groups corresponding to the couple relationship category is 50, there are 100 sample objects in total, the object portrait feature is user age, the corresponding preset proportion value is 0.4, and the preset portrait condition is that the age is less than or equal to 60 years old. Correspondingly, the total feature data amount of the object portrait feature is 100, the missing value filtering threshold is 40, if 30 sample objects in the 100 sample objects are older than 60 years old, and the ages of the other sample objects are less than or equal to 60 years old, the total feature data missing amount is 30, which is less than the corresponding missing value filtering threshold 40, and data filling processing needs to be performed. The user age is a discrete feature, so the user age of the 30 sample objects can be replaced based on the corresponding preset constant value; if 50 sample objects in the 100 sample objects are older than 60 years old, and the ages of the other sample objects are less than or equal to 60 years old, the total feature data missing amount is 50, which is greater than the corresponding missing value filtering threshold 40, and the user age feature is discarded and the corresponding user age single-value feature is deleted.

[0120] Further, the preprocessing of the object portrait feature can also include feature outlier processing. In some cases, the feature distribution of the object portrait feature can be obtained, the object portrait feature values of the sample objects in the top preset proportion of the feature distribution are determined as outliers, and the outliers are filtered out. The feature value filling processing is performed using a corresponding feature filling value. The determination method of the feature filling value can be similar to the foregoing, which will not be described here again. It can be understood that the preset proportion can be set based on the requirements of the application scenario, which is not limited in the present application.

[0121] Further, the preprocessing of the object portrait feature can also include feature derivation processing. In some cases, methods including but not limited to feature transformation, feature squaring, feature addition and subtraction, etc. can be used to combine and derive the object portrait feature of the sample object to obtain a derived feature of the sample object, and the derived feature is added to the object portrait feature corresponding to the sample object.

[0122] In some cases, for continuous object portrait features, a binning discretization processing can be performed, and for discrete portrait features, an encoding processing can be performed based on one-hot or the like, to obtain the reference attribute feature of the sample object group.

[0123] It should be noted that the reference operation sequence data is obtained in a manner similar to the foregoing target operation sequence data, which will not be described here again.

[0124] S505: Sequence pattern mining is performed on the reference operation sequence data to obtain reference sequence patterns corresponding to a plurality of sample object groups.

[0125] In the embodiments of the present application, the reference operation sequence data of the sample object group corresponding to each preset association relationship category is subjected to sequence pattern mining. Specifically, the above sequence pattern mining can be performed based on the association algorithm of frequent item set mining. In one embodiment, the prefixspan algorithm and the multi-minimum support strategy can be used to mine the time-based reference sequence pattern of each preset association relationship category. Correspondingly, the calculation formula of the minimum support can be shown in the following formula (2):

[0126]

[0127] wherein n is the sample quantity of the sample object group corresponding to the same preset association relationship category, and a is the minimum support, which is set according to the sample quantity of the corresponding sample object group. The minimum supports corresponding to different preset association relationship categories can be the same or different.

[0128] For example, it is assumed that the sample quantity of the sample object group corresponding to the lover relationship category is 2, including sample object group A and sample object group B, the reference operation sequence data of sample object group A is represented by the identifier bcafgh, the reference operation sequence data of sample object group B is represented by the identifier bcdaghf, and the minimum support corresponding to the lover relationship category is 0.5. Correspondingly, the one-item prefix satisfying the minimum support and the corresponding projection data set are shown in Table 2 below:

[0129] Table 2

[0130]

[0131] Further, based on the results in Table 1 above, the two-item prefixes satisfying the minimum support and the corresponding projection data set are shown in Table 3 below:

[0132] Table 3

[0133]

[0134] Further, the three-item prefix, four-item prefix and five-item prefix satisfying the minimum support and the corresponding projection data set are shown in Tables 4 to 6 below:

[0135] Table 4

[0136]

[0137] Table 5

[0138]

[0139] Table 6

[0140]

[0141] Specifically, each prefix obtained is taken as a reference sequence pattern of the preset association relationship category, and then the reference sequence pattern corresponding to the preset association relationship category (the couple relationship category) is generated based on the above prefixes.

[0142] Further, according to the sequence pattern mining result, the sequence pattern support degree of each reference sequence pattern corresponding to the preset association relationship category can also be calculated. The frequency of the target reference sequence pattern in the preset association relationship category can be obtained by calculating the sequence pattern support degree. Specifically, the sequence pattern support degree can be equal to the ratio of the number of target reference sequence pattern samples to the total number of reference sequence pattern samples of the corresponding association relationship category. The number of target reference sequence pattern samples is the number of times the target reference sequence pattern appears in all reference sequence patterns corresponding to the sample object group of the preset association relationship category. For example, for the two samples of sample object group A and sample object group B, the reference sequence pattern corresponding to each of them contains bcagh, and the sequence pattern support degree of the reference sequence pattern is 1.

[0143] By performing sequence pattern mining on the reference sequence operation data, the reference sequence patterns of each preset association relationship category are obtained, without the need for manual rule making, the generalization ability of generating sequence patterns is stronger, and the preset sequence pattern library can be expanded through multiple rounds of iteration mining, thereby constantly improving the preset sequence pattern library and improving the recall rate of sequence pattern coverage. Since the process is automatically generated and does not require excessive human involvement, the method of the present application can well meet the online iterative automation needs. In addition, when sequence pattern mining is performed based on frequent item sets, the sequence mining result obtained can better represent the behavior commonality possessed in the preset association relationship category, further improving the accuracy of association relationship recognition.

[0144] S507: input the reference operation sequence data corresponding to the reference operation feature into the sequence correlation mining network, perform sequence correlation processing, and obtain the corresponding reference sequence correlation feature.

[0145] In some embodiments, the reference operation sequence data of each preset association relationship category can be vectorized, for example, for each sample object group of the same category, each object interaction operation data in the reference operation sequence data of the sample object group is one-hot encoded, and then each encoded object interaction operation data is spliced to obtain the reference operation feature of the sample object group, and further to obtain the reference operation feature corresponding to the preset association relationship category. For example, the operation sequence data of the sample object group is bcaf, the forwarding circle of friends operation b is encoded as [1, 0, 0, 0, 0, 0], the circle of friends like operation c is encoded as [0, 1, 0, 0, 0, 0], the forwarding e-commerce shopping link operation a is encoded as [0, 0, 1, 0, 0, 0], and the click link purchase operation f is encoded as [0, 0, 0, 1, 0, 0], and the reference operation feature of the sample object group is [1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0].

[0146] Further, the reference operation feature of each sample object group is input into the sequence correlation mining network of the initial relationship identification model to obtain the corresponding reference sequence correlation feature. In one embodiment, the sequence correlation mining network is a network layer constructed based on LSTM.

[0147] S509: Taking the reference attribute feature, the reference operation feature, the reference sequence mode and the reference sequence correlation feature as the input of the object group classification network, and taking the association relationship category label as the output of the initial relationship identification model, the initial relationship identification model is trained by association relationship identification constraint to obtain the preset relationship identification model.

[0148] In some embodiments, the object group classification network includes a feature encoding layer constructed based on a recurrent neural network, and the model loss in the training process of the preset relationship identification model is calculated based on a margin-softmax loss function to constrain the training of the association relationship identification.

[0149] In one embodiment, the object group classification network includes a feature encoding layer constructed based on Bi-LSTM, and the model loss in the training process is calculated based on am-softmax loss function. The expression of the corresponding object group classification network can be shown in formula (one) and formula (three):

[0150]

[0151]

[0152] In the training process, x represents reference attribute features, reference operation features, reference sequence correlation features, and reference sequence pattern features corresponding to reference sequence pattern features, y is a reference encoding feature output by the BI-LSTM layer, that is, a feature after encoding x, W is a preset association relationship category set, and the expression is , c n represents a preset association relationship category, that is, the above formula (three) can be transformed into .

[0153] Further, the expression of the am-softmax loss function is as shown in formula (four):

[0154]

[0155] wherein, represents the angle between and , and s and m are constants, the values of which can be set according to the application scenario and the number of sample object groups, and in an embodiment, s is 30 and m is 0.35.

[0156] In the model training process, the model is trained by minimizing the model loss based on the am-softmax loss function, and finally a preset relationship recognition model is obtained.

[0157] In actual application, according to the characteristics that the interactive operation of the object group has sequence correlation, that is, the feature output at a moment is not only related to the input feature at the moment, but also related to the input feature or output feature at the previous moment, the model constructed by using RNN\LSTM is used to mine sequence correlation features, and the model is trained by using margin softmax (such as am-softmax), the attribute features, operation features, sequence pattern features, and sequence correlation features of the object group of different association relationship categories are used for category similarity calculation and category recognition, so as to better recognize and distinguish the object groups of the same association relationship category, and compared with the existing method, the accuracy of association relationship mining can be better improved.

[0158] For example, for sample object group A, if its reference operational features are [1,0,0,0,0,0,0,1,0,0,0,0], reference sequence correlation features are [0.1,-0.35,0.578,0.982…], reference sequence pattern is bcagh, and the corresponding reference sequence pattern features are [0.215,0.47,1.55,-0.588…], and reference attribute features are [0.2,-0.4,0.09,0.54,-2.5, 0.7,-0.01,0.3,0.4,9]. The corresponding input x is [0.215,0.47,1.55,-0.588…, 0.2,-0.4,0.09,0.54,-2.5, 0.7,-0.01,0.3,0.4,9, 1,0,0,0,0,0,0,1,0,0,0,0, 0.1,-0.35,0.578,0.982…], and the resulting y is [0.579,1.732,0.9232,-0.2577…]. It should be noted that the above input x is only the result generated in one example. This x can be obtained using other feature concatenation and processing methods, such as weighted concatenation. This application is not limited to the above example.

[0159] Based on the above formula (iii), we can obtain p for each preset association category corresponding to y, and calculate the model loss based on the association category label and the am-softmax loss function. Then, we adjust the model parameters until the model converges to obtain the preset relationship recognition model.

[0160] In some embodiments, model convergence is determined by minimizing the model loss value during training. In some cases, the model convergence condition can be: the difference between the model loss value obtained in the current iteration and the model loss value obtained in the previous iteration is less than or equal to a preset threshold; or, the model loss value obtained in the current iteration is the inflection point of the loss curve corresponding to the model loss values ​​obtained in previous iterations. Accordingly, when the model loss value satisfies the model convergence condition, model convergence is determined, and the model parameters corresponding to the model loss value that satisfies the model convergence condition are used as the model parameters of the preset relation recognition model.

[0161] This application embodiment also provides an object association identification device 600, such as... Figure 8 As shown, Figure 8 This illustration shows a structural diagram of an object association identification device provided in an embodiment of this application. The device may include:

[0162] Feature acquisition module 10: used to acquire target attribute features and target operation sequence data of the object group to be identified; the target operation sequence data represents the object interaction operations of the object group to be identified;

[0163] The sequence pattern matching module 20 is configured to perform sequence pattern matching on the target operation sequence data based on a preset sequence pattern library to obtain a target operation sequence pattern of the to-be-identified object group.

[0164] The sequence correlation mining module 30 is configured to perform sequence correlation processing on the target operation feature corresponding to the target operation sequence data by using a sequence correlation mining network of a preset relationship recognition model to obtain a target sequence correlation feature of the to-be-identified object group.

[0165] The object association recognition module 40 is configured to perform association relationship recognition on the to-be-identified object group based on the target attribute feature, the target operation feature, the target operation sequence pattern and the target sequence correlation feature by using an object group classification network of a preset relationship recognition model to obtain an association relationship category of the to-be-identified object group.

[0166] In some embodiments, the object group classification network includes a feature encoding layer and a similarity calculation layer, and the object association recognition module 40 can include:

[0167] The encoding processing unit is configured to input the target attribute feature, the target operation feature, the target sequence correlation feature and the target sequence pattern feature corresponding to the target operation sequence pattern into the feature encoding layer to perform encoding processing to obtain a target encoding feature of the to-be-identified object group.

[0168] The similarity calculation unit is configured to input the target encoding feature into the similarity calculation layer to perform similarity calculation between the target encoding feature and each category encoding feature in a preset encoding feature set to obtain each category similarity corresponding to the target encoding feature.

[0169] The threshold comparison unit is configured to compare each category similarity with a similarity threshold respectively to obtain a similarity comparison result.

[0170] The association relationship category determination unit is configured to determine the association relationship category of the to-be-identified object group according to the similarity comparison result.

[0171] In some embodiments, the preset encoding feature set includes a category encoding feature corresponding to each preset association relationship category, and the association relationship category determination unit can include:

[0172] The category encoding feature determination subunit is configured to determine, if one or more category similarities are greater than or equal to the similarity threshold, a category similarity with the largest value among the category similarities as the target category encoding feature, and determine the preset association relationship category corresponding to the target category encoding feature as the association relationship category of the to-be-identified object group.

[0173] In some embodiments, the association relationship category determination unit can further include:

[0174] The to-be-updated relationship category determining subunit is configured to determine that the association relationship category of the to-be-identified object group is a to-be-updated relationship category if each category similarity is less than the similarity threshold.

[0175] The data updating subunit is configured to update the preset sequence pattern library based on the target operation sequence pattern of the to-be-identified object group, and update the preset encoding feature set based on the target encoding feature of the to-be-identified object group.

[0176] In some embodiments, the feature acquisition module 10 can be configured to acquire the target operation feature, and can specifically include:

[0177] The object interaction operation data acquisition unit is configured to acquire object interaction operation data and corresponding interaction operation time information of the to-be-identified object group within a preset time period.

[0178] The operation sequence data acquisition unit is configured to perform time sorting and data splicing processing on the object interaction operation data within the preset time period according to the interaction operation time information, to obtain target operation sequence data.

[0179] In some embodiments, the target object group feature further includes an attribute feature of the to-be-identified object group, and correspondingly, the feature acquisition module 10 can further include:

[0180] The object attribute data acquisition unit is configured to acquire object attribute data of each object in the to-be-identified object group.

[0181] The attribute feature generation unit is configured to generate basic object information of each object based on the object attribute data.

[0182] The attribute feature generation unit is configured to perform vectorization processing and feature splicing processing on the basic object information of each object to obtain the target attribute feature.

[0183] In some embodiments, the preset sequence pattern library pre-stores reference sequence patterns corresponding to each preset association relationship category, and the reference sequence pattern is obtained by performing sequence pattern mining on reference operation sequence data of a sample object group.

[0184] Correspondingly, the sequence pattern matching module 20 includes:

[0185] The matching sequence pattern determining unit is configured to determine the reference sequence pattern of the target operation sequence data as a matching sequence pattern in the preset sequence pattern library.

[0186] The operation sequence pattern generation unit is configured to generate a target operation sequence pattern according to the matching sequence pattern.

[0187] In some embodiments, the apparatus can further include:

[0188] The recognition model construction module is configured to construct an initial relationship recognition model, and the initial relationship recognition model comprises a sequence correlation mining network and an object group classification network.

[0189] The training sample set construction module is configured to construct a training sample set based on reference attribute features of a plurality of sample object groups, reference operation sequence data, and corresponding association relationship category labels.

[0190] The sequence pattern mining module is configured to perform sequence pattern mining on the reference operation sequence data to obtain reference sequence patterns corresponding to the plurality of sample object groups.

[0191] The sequence correlation feature mining module is configured to input reference operation features corresponding to the reference operation sequence data into the sequence correlation mining network to perform sequence correlation processing, and obtain corresponding reference sequence correlation features.

[0192] The model training module is configured to use the reference attribute features, the reference operation features, the reference sequence patterns, and the reference sequence correlation features as inputs of the object group classification network, use the association relationship category labels as outputs of the initial relationship recognition model, perform constraint training on the initial relationship recognition model based on association relationship recognition, and obtain a preset relationship recognition model.

[0193] The object association recognition device provided in the embodiments of the present application comprises a processor and a memory, the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the object association recognition method provided in the above method embodiments.

[0194] The memory can be used to store software programs and modules, and the processor can execute various functions, applications, and data processing by running the software programs and modules stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, application programs required for functions, etc.; and the data storage area can store data created according to the use of the device, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other volatile solid-state memory device. Accordingly, the memory can also include a memory controller to provide access of the processor to the memory.

[0195] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal, a server, or a similar computing device. Taking the case of running on a server as an example, Figure 9 is a hardware structure block diagram of a server of an object association recognition method provided in the embodiments of the present application. As shown in Figure 9As shown, the server 800 can vary greatly in configuration and performance, and can include one or more Central Processing Units (CPU) 810 (processor 810 can include, but is not limited to, a microprocessor, an Application-Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), etc.), a memory 830 for storing data, one or more storage media 820 (e.g., one or more mass storage devices) for storing applications 823 or data 822. The memory 830 and the storage media 820 can be of any type generally known or used in the art including, but not limited to, volatile and non-volatile memory, removable memory, and non-removable memory. The applications stored in the storage media 820 can include one or more modules, each of which can include a series of instructions for operating on the server. Further, the CPU 810 can be configured to communicate with the storage media 820 to execute a series of instructions in the storage media 820 on the server 800. The server 800 can also include one or more power supplies 860, one or more wired or wireless network interfaces 850, one or more input / output interfaces 840, and / or one or more operating systems 821, such as Windows Server TM , Mac OS X TM , Unix TM , Linux™, FreeBSD™, etc.

[0196] The input / output interface 840 can be configured to receive or transmit data via a network. Examples of the network can include a wireless network provided by a communication provider of the server 800. In one example, the input / output interface 840 includes a Network Interface Controller (NIC) that can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the input / output interface 840 can be a Radio Frequency (RF) module that is configured to communicate with the Internet through a wireless manner.

[0197] Those of ordinary skill in the art can understand that Figure 9 the structure shown is merely illustrative, and does not limit the structure of the electronic device described above. For example, the server 800 can include more or fewer components than those shown in Figure 9 or have a different configuration than that shown in Figure 9 .

[0198] The embodiment of the present application also provides a computer readable storage medium, which can be arranged in a server to store at least one instruction or at least one program for implementing an object association recognition method in the method embodiment, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the object association recognition method provided by the above method embodiment.

[0199] Optionally, in the embodiment, the storage medium can be located in at least one of a plurality of network servers of a computer network. Optionally, in the embodiment, the storage medium can include but is not limited to a U disk, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk and various media capable of storing program codes.

[0200] In the embodiment of the present application, the object association recognition method or device disclosed in the present application can run in one or more servers as Figure 9 described in the present application, wherein the plurality of servers can be composed of a block chain, and the object association recognition method or device can be provided with data storage and other services, such as one or more of the training sample set, the preset sequence pattern library, the preset encoding feature set, and the target operation sequence data and object attribute data of the object group to be recognized in the present application, which can be stored in the above block chain, and the server is a node on the block chain. Figure 10 is a structural schematic diagram of a block chain system provided by the embodiment of the present application. As Figure 10 shown, the server can be a node 920 in a distributed system 910, wherein the distributed system can be a block chain system, the block chain system can be a distributed system formed by connecting a plurality of nodes through network communication, the nodes can form a peer-to-peer (P2P) network, and any form of computer device, such as a server, a client 930 and other electronic devices, can become a node in the block chain system by joining the peer-to-peer network, wherein the block chain includes a series of blocks (Block) connected in the order of time of generation, once a new block is added to the block chain, it will not be removed, and the block records the record data submitted by the nodes in the block chain system.

[0201] The blockchain is a new application mode of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithm, and is essentially a decentralized database. The blockchain is a series of data blocks associated using a cryptography method, each data block contains information of a batch of network transactions, and is used to verify the validity (anti-fake) of the information and generate the next block. The blockchain can include a blockchain underlying platform, a platform product service layer, and an application service layer. The blockchain underlying platform can include user management, basic services, smart contracts, and operation detection processing modules. The user management module is responsible for identity information management of all blockchain participants, including maintaining public and private key generation (account management), key management, and user real identity and blockchain address correspondence maintenance (permission management), and in the case of authorization, detecting and auditing the transaction of certain real identities, and providing risk control rule configuration (risk audit). The basic service module is deployed on all blockchain node devices to verify the validity of business requests, and record to the storage after completing the consensus of the valid request. For a new business request, the basic service first performs interface adaptation analysis and authentication processing (interface adaptation), then encrypts the business information through a consensus algorithm (consensus management), and then transmits the complete and consistent information to the shared ledger (network communication) and records the storage. The smart contract module is responsible for contract registration and issuance, contract triggering and contract execution. Developers can define contract logic through a certain programming language, publish it to the blockchain (contract registration), trigger execution according to the logic of the contract terms, complete the contract logic, and also provide contract upgrade and cancellation functions. The operation detection module is mainly responsible for deployment, configuration modification, contract setting, cloud adaptation during product release, and real-time state visualization output during product operation, such as alarm, network detection, node device health state detection, and the like. The platform product service layer provides basic capabilities and implementation frameworks for typical applications. Developers can stack business characteristics based on these basic capabilities to complete the blockchain implementation of business logic. The application service layer provides application services based on the blockchain solution for business participants to use.

[0202] According to an aspect of the present application, a computer program product or computer program is provided, which includes computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the method provided in the various optional implementation manners described above.

[0203] It can be seen from the embodiments of the object association recognition method, device, equipment, server or storage medium provided in the application that the target attribute features and the target operation sequence data of the to-be-recognized object group are acquired; the target operation sequence data is subjected to sequence pattern matching based on a preset sequence pattern library to obtain a target operation sequence pattern of the to-be-recognized object group; automatic mining of the operation sequence pattern can be implemented, without the need to set a large number of artificial rules, and better generalization ability and timeliness are achieved; then the target operation features corresponding to the target operation sequence data are subjected to sequence correlation processing by using a sequence correlation mining network of a preset relationship recognition model to obtain target sequence correlation features of the to-be-recognized object group; and the to-be-recognized object group is subjected to association relationship recognition based on the target attribute features, the target operation features, the target operation sequence pattern and the target sequence correlation features by using an object group classification network of the preset relationship recognition model, and then the association relationship category of the to-be-recognized object group is obtained. Association relationship recognition can be performed by comprehensively combining various different features, and thus the accuracy of the relationship recognition result is improved.

[0204] It should be noted that the above-mentioned embodiments of the application are only for description, and do not represent the advantages and disadvantages of the embodiments. The above-mentioned embodiments of the application are described. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0205] Each of the embodiments in the application is described in a progressive manner, and the same or similar parts of each embodiment can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment and storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts are referred to the part of the method embodiment.

[0206] A person of ordinary skill in the art can understand that all or part of the steps of the above-mentioned embodiments can be completed by hardware, or a program can be used to instruct the related hardware to complete. The program can be stored in a computer readable storage medium. The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk.

[0207] The above is only a preferred embodiment of the application, and is not intended to limit the application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the application shall be included in the protection scope of the application.

Claims

1. A method for object association recognition, characterized in that, The method includes: Obtain the target attribute features of the group of objects to be identified; Obtain object interaction operation data and corresponding interaction operation time information within a preset time period of the group of objects to be identified; Based on the interaction operation time information, the object interaction operation data within the preset time period is sorted by time and spliced ​​to obtain target operation sequence data; the target operation sequence data is the data representing the object interaction operation of the object group to be identified, and the object interaction operation data is the operation data generated by each object in the object group to be identified when performing interaction operations. Based on a preset sequence pattern library, sequence pattern matching is performed on the target operation sequence data to obtain the target operation sequence pattern of the object group to be identified; the preset sequence pattern library pre-stores reference sequence patterns corresponding to each preset association category, and the reference sequence patterns corresponding to the preset association category are obtained by sequence pattern mining of the reference operation sequence data of the sample object group of the same category; the reference sequence patterns of the subsequences of the target operation sequence data in the preset sequence pattern library are sequence patterns that match the target operation sequence data; Using a sequence correlation mining network based on a preset relationship recognition model, sequence correlation processing is performed on the target operation features corresponding to the target operation sequence data to obtain the target sequence correlation features of the object group to be identified. Using the object group classification network of the preset relationship recognition model, the association relationship of the object group to be identified is performed based on the target attribute features, the target operation features, the target operation sequence pattern and the target sequence correlation features, so as to obtain the association relationship category of the object group to be identified.

2. The method according to claim 1, characterized in that, The object group classification network includes a feature encoding layer and a similarity calculation layer. Correspondingly, the object group classification network utilizing the preset relationship recognition model identifies the association relationships of the object group to be identified based on the target attribute features, the target operation features, the target operation sequence pattern, and the target sequence correlation features, obtaining the association relationship categories of the object group to be identified, including: The target attribute features, the target operation features, the target sequence correlation features, and the target sequence pattern features corresponding to the target operation sequence pattern are input into the feature encoding layer for encoding processing to obtain the target encoding features of the object group to be identified. The target coding feature is input into the similarity calculation layer to calculate the similarity between the target coding feature and each category of coding features in the preset coding feature set, thereby obtaining the category similarity corresponding to the target coding feature; The similarity of each category is compared with the similarity threshold to obtain the similarity comparison results; The association category of the group of objects to be identified is determined based on the similarity comparison results.

3. The method according to claim 2, characterized in that, The preset coding feature set includes category coding features corresponding to each preset association category, and determining the association category of the object group to be identified based on the similarity comparison result includes: If one or more category similarities are greater than or equal to the similarity threshold, the category coding feature corresponding to the category with the largest value among the category similarities is determined as the target category coding feature; The preset association category corresponding to the target category encoding feature is used as the association category of the object group to be identified.

4. The method according to claim 3, characterized in that, The method further includes: If the similarity of each category is less than the similarity threshold, the association category of the object group to be identified is determined as the relationship category to be updated; The preset sequence pattern library is updated based on the target operation sequence pattern of the object group to be identified, and the preset coding feature set is updated based on the target coding features of the object group to be identified.

5. The method according to claim 1, characterized in that, The method further includes: Obtain the object attribute data of each object in the group of objects to be identified; Generate basic object information for each object based on the object attribute data; The basic object information of each object is vectorized and feature concatenation is performed to obtain the target attribute features.

6. The method according to any one of claims 1-5, characterized in that, The method includes: An initial relationship identification model is constructed, which includes a sequence correlation mining network and an object group classification network; A training sample set is constructed based on the reference attribute features, reference operation sequence data, and corresponding association category labels of multiple sample object groups; the reference operation sequence data is data representing the object interaction operations of the sample object groups. Sequence pattern mining is performed on the reference operation sequence data to obtain the reference sequence patterns corresponding to the multiple sample object groups; The reference operation features corresponding to the reference operation sequence data are input into the sequence correlation mining network for sequence correlation processing to obtain the corresponding reference sequence correlation features. Using the reference attribute features, the reference operation features, the reference sequence pattern, and the reference sequence correlation features as inputs to the object group classification network, and using the association category label as the output of the initial relationship recognition model, the initial relationship recognition model is subjected to constraint training for association recognition to obtain the preset relationship recognition model.

7. An object association recognition device, characterized in that, The device includes: Feature acquisition module: used to acquire target attribute features of the object group to be identified; acquire object interaction operation data and corresponding interaction operation time information within a preset time period of the object group to be identified; perform time sorting and data splicing processing on the object interaction operation data within the preset time period according to the interaction operation time information to obtain target operation sequence data; the target operation sequence data is data characterizing the object interaction operation of the object group to be identified, and the object interaction operation data is the operation data generated by each object in the object group to be identified when performing interaction operations; Sequence pattern matching module: This module performs sequence pattern matching on the target operation sequence data based on a preset sequence pattern library to obtain the target operation sequence patterns of the object group to be identified. The preset sequence pattern library stores reference sequence patterns corresponding to each preset association category. The reference sequence patterns corresponding to the preset association category are obtained by sequence pattern mining of reference operation sequence data of sample object groups of the same category. The reference sequence patterns of subsequences of the target operation sequence data in the preset sequence pattern library are sequence patterns that match the target operation sequence data. Sequence correlation mining module: Used to process the sequence correlation of the target operation features corresponding to the target operation sequence data by using a sequence correlation mining network based on a preset relationship recognition model, so as to obtain the target sequence correlation features of the object group to be identified; Object association identification module: Used to identify the association relationship of the object group to be identified by the object group classification network of the preset relationship identification model based on the target attribute features, the target operation features, the target operation sequence pattern and the target sequence correlation features, and obtain the association relationship category of the object group to be identified.

8. The apparatus according to claim 7, characterized in that, The object group classification network includes a feature encoding layer and a similarity calculation layer. Correspondingly, the object association recognition module includes: Encoding processing unit: used to input the target attribute features, the target operation features, the target sequence correlation features, and the target sequence pattern features corresponding to the target operation sequence pattern into the feature encoding layer, perform encoding processing, and obtain the target encoding features of the object group to be identified; Similarity calculation unit: used to input the target coding feature into the similarity calculation layer, calculate the similarity between the target coding feature and each category of coding features in the preset coding feature set, and obtain the category similarity corresponding to the target coding feature; Threshold comparison unit: used to compare the similarity of each category with the similarity threshold respectively to obtain the similarity comparison result; Association category determination unit: used to determine the association category of the group of objects to be identified based on the similarity comparison results.

9. The apparatus according to claim 8, characterized in that, The preset coding feature set includes category coding features corresponding to each preset association category, and the association category determination unit includes: Category coding feature determination subunit: configured to, if one or more category similarities are greater than or equal to the similarity threshold, determine the category coding feature corresponding to the category with the largest numerical value among the category similarities as the target category coding feature; and configured to, The preset association category corresponding to the target category encoding feature is used as the association category of the object group to be identified.

10. The apparatus according to claim 9, characterized in that, The association category determination unit further includes: The subunit for determining the relationship category to be updated is used to determine the relationship category to be updated for the group of objects to be identified if the similarity of each category is less than the similarity threshold. Data update subunit: used to update the preset sequence pattern library based on the target operation sequence pattern of the object group to be identified, and to update the preset coding feature set based on the target coding features of the object group to be identified.

11. The apparatus according to claim 7, characterized in that, The feature acquisition module includes: Object attribute data acquisition unit: used to acquire object attribute data of each object in the group of objects to be identified; Attribute feature generation unit: used to generate basic object information for each object based on the object attribute data; Attribute feature generation unit: used to perform vectorization and feature concatenation processing on the basic object information of each object to obtain the target attribute features.

12. The apparatus according to any one of claims 7-11, characterized in that, The device includes: The identification model construction module is used to construct an initial relationship identification model, which includes a sequence correlation mining network and an object group classification network. Training sample set construction module: used to construct a training sample set based on reference attribute features, reference operation sequence data, and corresponding association category labels of multiple sample object groups; the reference operation sequence data is data representing the object interaction operations of the sample object group; Sequence pattern mining module: used to perform sequence pattern mining on the reference operation sequence data to obtain the reference sequence patterns corresponding to the multiple sample object groups; Sequence correlation feature mining module: used to input the reference operation features corresponding to the reference operation sequence data into the sequence correlation mining network, perform sequence correlation processing, and obtain the corresponding reference sequence correlation features; Model training module: used to perform constraint training on the initial relationship recognition model to obtain the preset relationship recognition model by taking the reference attribute features, the reference operation features, the reference sequence pattern and the reference sequence correlation features as input to the object group classification network, and taking the association category label as output to the initial relationship recognition model, so as to obtain the preset relationship recognition model.

13. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction or at least one program segment, which is loaded and executed by a processor using the object association identification method as described in any one of claims 1-6.

14. An object association recognition device, characterized in that, The device includes a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the object association recognition method as described in any one of claims 1-6.

15. A computer program product, characterized in that, The computer program product includes computer instructions stored in a computer-readable storage medium; a processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the object association identification method as described in any one of claims 1-6.

Images