A method and device for determining a knowledge graph abstract

Abstract

The application discloses a knowledge graph abstract determination method and device, comprising: determining an initial graph abstract of a current version of a knowledge graph and a first candidate triple corresponding to the initial graph abstract; obtaining a user portrait of a target user; determining first operation information corresponding to the first candidate triple according to the initial graph abstract and the user portrait; determining a target graph abstract of the current version of the knowledge graph according to the initial graph abstract and the first operation information; predicting in combination with the initial graph abstract and the user portrait to determine the first operation information; and operating the initial graph abstract according to the first operation information to obtain the target graph abstract, so that the target graph abstract can reflect key information in the knowledge graph and can also take into account the interests and demands of the user.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and particularly relates to a knowledge graph abstract determination method and device. BACKGROUND

[0002] A knowledge graph is a series of graphs showing the development process and structural relationship of knowledge. In other words, the knowledge graph includes a large number of nodes, each node includes certain information, and there is a connection relationship between the nodes. Generally, due to the large amount of information in the knowledge graph, the knowledge graph abstract is presented to the user in some cases, so as to facilitate the user to browse and operate.

[0003] The so-called knowledge graph abstract is a small number of nodes and their connection relationship in the knowledge graph, but it can represent the key information in the knowledge graph to a certain extent. In the prior art, a part of important nodes can be determined by a specific abstract calculation method, that is, the knowledge graph abstract is determined. However, the existing knowledge graph abstract determination method does not consider the interest degree and use feedback of the user for the knowledge graph abstract, so the effect needs to be improved. SUMMARY

[0004] The present application provides a knowledge graph abstract determination method and device.

[0005] In a first aspect, the present application provides a knowledge graph abstract determination method, comprising:

[0006] determining an initial graph abstract of a current version of a knowledge graph, and a first candidate triple corresponding to the initial graph abstract;

[0007] obtaining a user portrait of a target user;

[0008] determining first operation information corresponding to the first candidate triple according to the initial graph abstract and the user portrait;

[0009] determining a target graph abstract of the current version of the knowledge graph according to the initial graph abstract and the first operation information.

[0010] Preferably, the determination of the first candidate triple corresponding to the initial graph abstract comprises:

[0011] determining a first node included in the initial graph abstract; and a second node corresponding to the first node;

[0012] determining a plurality of first candidate triples according to the first node and the second node.

[0013] Preferably, the determining the first operation information corresponding to the first candidate triple according to the initial graph summary and the user portrait comprises:

[0014] determining a first summary feature according to the initial graph summary;

[0015] determining a first user feature according to the user portrait;

[0016] determining a first triple feature according to the first candidate triple;

[0017] determining the first operation information corresponding to the first candidate triple according to the first summary feature, the first user feature and the first triple feature.

[0018] Preferably, the determining the first operation information corresponding to the first candidate triple according to the first summary feature, the first user feature and the first triple feature comprises:

[0019] determining a first feature map corresponding to the first candidate triple according to the first summary feature, the first user feature and the first triple feature;

[0020] determining operation probabilities of each candidate operation for the target user according to the first feature map;

[0021] determining the first operation information according to the operation probabilities of each candidate operation.

[0022] Preferably, the determining the target graph summary of the current version of the knowledge graph according to the initial graph summary and the first operation information comprises:

[0023] performing a first operation on each of the first candidate triples corresponding to the initial graph summary according to the first operation information of each of the first candidate triples to determine the target graph summary.

[0024] Preferably, the first operation information comprises node expansion operation information or node retraction operation information; and the performing the first operation on each of the first candidate triples corresponding to the initial graph summary comprises:

[0025] when the first operation information is the node expansion operation information, expanding the first candidate triple corresponding to the first operation information;

[0026] when the first operation information is the node retraction operation information, retracting the first candidate triple corresponding to the first operation information.

[0027] Preferably, the method further comprises:

[0028] The target knowledge graph summary of the current version of the knowledge graph is determined as the initial knowledge graph summary of the next version of the knowledge graph.

[0029] In a second aspect, the present application provides a knowledge graph summary determination apparatus, comprising:

[0030] An initial summary determination module is configured to determine an initial knowledge graph summary of a current version of a knowledge graph and a first candidate triple corresponding to the initial knowledge graph summary information.

[0031] A user portrait determination module is configured to obtain a user portrait of a target user.

[0032] A first operation information determination module is configured to determine first operation information corresponding to the first candidate triple according to the initial knowledge graph summary and the user portrait.

[0033] A target summary determination module is configured to determine a target knowledge graph summary of the current version of the knowledge graph according to the initial knowledge graph summary and the first operation information.

[0034] In a third aspect, the present application provides a computer readable storage medium, which stores a computer program for executing the knowledge graph summary determination method provided by the present application.

[0035] In a fourth aspect, the present application provides an electronic device, comprising:

[0036] A processor;

[0037] A memory for storing executable instructions of the processor;

[0038] The processor is configured to read the executable instructions from the memory and execute the instructions to implement the knowledge graph summary determination method provided by the present application.

[0039] Compared with the prior art, the knowledge graph summary determination method and apparatus provided by the present application combines an initial knowledge graph summary and a user portrait to predict first operation information, and performs operation on the initial knowledge graph summary according to the first operation information to obtain a target knowledge graph summary, so that the target knowledge graph summary can reflect key information in the knowledge graph and also take into account the interests and demands of users. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 A flowchart of a knowledge graph summary determination method provided by an embodiment of the present application;

[0041] Figure 2 A schematic diagram of a knowledge graph summary in a knowledge graph summary determination method provided by an embodiment of the present application;

[0042] Figure 3 A structure diagram of an abstract determination network in a method for determining a knowledge graph abstract provided by an embodiment of the present application is shown in the following figure.

[0043] Figure 4 A flowchart of another method for determining a knowledge graph abstract provided by an embodiment of the present application is shown in the following figure.

[0044] Figure 5 A training method flowchart of an abstract determination network in a method for determining a knowledge graph abstract provided by an embodiment of the present application is shown in the following figure.

[0045] Figure 6 A structure diagram of a device for determining a knowledge graph abstract provided by an embodiment of the present application is shown in the following figure. DETAILED DESCRIPTION

[0046] In order to make the purposes, features and advantages of the present application more obvious and easy to understand, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0047] Because the information in a knowledge graph is huge, in some cases, a knowledge graph abstract is presented to a user for browsing and operation. The so-called knowledge graph abstract is a small number of nodes and their connection relationships in the knowledge graph, but it can represent the key information in the knowledge graph to a certain extent. In the prior art, a part of important nodes can be determined by a specific abstract calculation method, i.e., a knowledge graph abstract is determined.

[0048] However, the existing abstract calculation method mainly calculates the knowledge graph abstract according to the features of the nodes. That is, only the node features are considered in the calculation process, i.e., whether the information in the node is "key" is calculated and determined; and the interests and needs of the user are not considered, so that the part of nodes actually required by the user cannot be accurately presented.

[0049] Therefore, a method for determining a knowledge graph abstract is provided in the embodiments of the present application. As shown in the following figure, the method in the embodiments includes the following steps: Figure 1

[0050] Step 101, determining an initial graph abstract of a current version of a knowledge graph and a first candidate triple corresponding to the initial graph abstract.

[0051] ​In this embodiment, the process of determining the knowledge graph summary can be iterated through several versions based on actual circumstances. The knowledge graph summary determined in the previous version can be used as the initial knowledge graph summary for the current version. The initial knowledge graph summary can be as follows: Figure 2 As shown, Figure 2 The content enclosed within the solid-line rectangle is the initial spectral summary. In this embodiment, the initial spectral summary can be represented by Gt.

[0052] The initial graph summary will include several nodes and the connections between them. Figure 2 Each solid-lined circle in the diagram represents a node in the initial atlas summary. In this embodiment, the nodes in the initial atlas summary can also be referred to as the first node. Figure 2 Between the dashed rectangle and the solid rectangle, there are several nodes, such as... Figure 2 The node shown is a dashed circle. The node in the dashed circle is the next-hop node of the first node, meaning it has a direct connection to the first node. In this embodiment, the next-hop node of the first node is referred to as the second node.

[0053] In this field, two nodes and the connection between them can be logically referred to as a "triple". The connection between the two nodes can be directional or non-directional. When the connection is directional, the upstream node is called the head node of the triple, and the downstream node is called the tail node. For example... Figure 2 The dashed ellipse contains an exemplary "triple". Figure 2 Any triple formed by the first node and the second node included in the initial graph summary can be called a first candidate triple. That is, in this embodiment, the first node included in the initial graph summary and the second node corresponding to the first node can be determined based on the initial graph summary; multiple first candidate triples can be determined based on the first node and the second node. In this embodiment, any first candidate triple can be represented by ct.

[0054] Step 102: Obtain the user profile of the target user.

[0055] This embodiment will ultimately determine the target graph summary corresponding to the current version, which can be browsed by specific target users, ensuring that the target graph summary specifically meets the needs and interests of the target users. However, the graph summary determined in the previous version (i.e., the initial graph summary of the current version) may not meet the needs and interests of the target users. Therefore, the purpose of this embodiment is to adjust the initial graph summary according to the needs of the target users to obtain a targeted target graph summary.

[0056] Therefore, the user portrait of the target user can be obtained. The user portrait can include information of the target user in multiple dimensions. For example, the user portrait can include personal characteristics of the target user, such as identity, gender, age, occupation, education, device model used, performance information, and the like. In this embodiment, the user information included in the user portrait can be represented as a set Bt.

[0057] In step 103, first operation information corresponding to each first candidate triple is determined according to the initial graph summary and the user portrait.

[0058] In this embodiment, the so-called adjustment of the initial graph summary is actually an operation on each first candidate triple in the knowledge graph. That is, the operation that the target user will perform on each first candidate triple can be predicted by an artificial intelligence-based algorithm, so that the target user's desired operation is completed in advance, and a target graph summary that meets the user's demand is obtained. In this embodiment, the above prediction can be performed by using a pre-trained summary determination network. The structure of the summary determination network can refer to the network structure shown in Figure 3 of course, other network models with the same or similar functions can also be selected according to requirements, and the embodiment is not limited thereto.

[0059] By using the summary determination network, the first operation information corresponding to each first candidate triple can be obtained by predicting each first candidate triple in the initial graph summary one by one. The first operation information means the operation content that the target user will perform on the first candidate triple.

[0060] Specifically, the first summary feature can be determined according to the initial graph summary; the first user feature can be determined according to the user portrait; the first triple feature can be determined according to any first candidate triple; and the first operation information corresponding to the first candidate triple can be determined according to the first summary feature, the first user feature, and the first triple feature.

[0061] The network structure shown in Figure 3 The above process can be described as follows by combining the network structure shown in G (Gt) is obtained. The user portrait Bt is input into the first feedforward network structure (FNN1) of the summary determination network, and the first user feature Φ B(Bt). Any first candidate triple ct is input into a second feedforward network structure (FNN2) of the summary determining network. Specifically, the word vectors corresponding to the head node, tail node and connection relationship of the first candidate triple ct can be concatenated, and a triple vector is obtained, which is input into the second feedforward network structure, and a first triple feature Φ c (ct) is obtained.

[0062] Then, according to the first summary feature Φ G (Gt), the first user feature Φ B (Bt) and the first triple feature Φ c (ct), a first feature map corresponding to the first candidate triple can be determined. Specifically, the above three features can be concatenated by using a concatenation function to obtain the first feature map. The mathematical representation is: E = concat(Φ G (Gt), Φ B (Bt), Φ c (ct)); wherein, E represents the first feature map, and concat represents the concatenation function. In the embodiment, the specific content of the concatenation function is not limited, and any function capable of achieving the same or similar function can be used as the concatenation function concat in the overall technical solution of the embodiment.

[0063] According to the first feature map, the operation probability of each candidate operation of the target user can be determined. For example, in the knowledge graph, the operation on the node can include an expansion operation and a retraction operation, i.e., two candidate operations. The expansion operation can be represented as a1, and the retraction operation can be represented as a2. The first feature map is input into a third feedforward network structure (FNN3) of the summary determining network, and the operation probability of the target user performing each candidate operation can be predicted. The operation probability of the expansion operation a1 can be represented as Q1, and the operation probability of the retraction operation a2 can be represented as Q2.

[0064] In the embodiment, according to the prediction result, it can be considered that if the target user actually browses the initial graph summary, the target user will manually perform the candidate operation with the maximum operation probability on the first candidate triple ct. Therefore, according to the operation probability of each candidate operation, the first operation information can be determined; that is, the candidate operation with the maximum operation probability is determined as the information content of the first operation information. In other words, the information content of the first operation information is the operation to be performed by the target user.

[0065] Step 104, determining the target graph summary of the current version of the knowledge graph according to the initial graph summary and the first operation information.

[0066] After the first operation information corresponding to each first candidate triple in the initial graph summary is determined, the specific operation such as the expansion operation or the retraction operation can be performed on the corresponding first candidate triple according to each first operation information. Because the first operation information is the operation that the target user is predicted to perform, after the operation of the initial graph summary is completed, the operation process that the target user is about to perform is simulated, or in other words, the operation is completed in advance on behalf of the target user before the target user browses. The current version of the target graph summary is obtained after the operation, that is, the graph summary that the target user needs is obtained according to the prediction. Thus, the current version of the target graph summary in the knowledge graph is determined in the embodiment.

[0067] It should be noted that because the summary determination network combines the initial graph summary and the user portrait to make the prediction in the process of predicting the first operation information, it can consider whether the information in the node is "key" and can also consider the interests and needs of the user in the prediction. Therefore, the target graph summary obtained finally is more suitable for the needs of the user.

[0068] Through the above technical solution, it can be known that the embodiment has the beneficial effects that the initial graph summary and the user portrait are combined to make the prediction to determine the first operation information, and the initial graph summary is operated according to the first operation information to obtain the target graph summary, so that the target graph summary can reflect the key information in the knowledge graph and can also take into account the interests and needs of the user.

[0069] Figure 1 It is only the basic embodiment of the method described in the application, and certain optimization and expansion can be made on the basis thereof, and other preferred embodiments of the method can also be obtained.

[0070] As Figure 4 shown, it is another specific embodiment of the method for determining the knowledge graph summary described in the application. The embodiment is further expanded on the basis of the foregoing embodiment. The method specifically includes the following steps:

[0071] Step 401, determining an initial graph summary in a current version of a knowledge graph and first candidate triples corresponding to the initial graph summary.

[0072] Step 402, obtaining a user portrait of a target user.

[0073] Step 403, determining first operation information corresponding to the first candidate triple according to the initial graph summary and the user portrait.

[0074] The foregoing steps 401 to 403 are consistent with the corresponding contents of the embodiment shown in Figure 1 , and thus the description thereof is not repeated.

[0075] Step 404, performing a first operation on each first candidate triple corresponding to the initial graph summary according to the first operation information of each first candidate triple, to determine a target graph summary.

[0076] After determining the first operation information corresponding to each first candidate triple in the initial graph summary, the specific operation can be performed on the corresponding first candidate triple according to each first operation information.

[0077] For example, in the knowledge graph, the operation on the node can include an expansion operation and a retraction operation. The so-called expansion operation is to further expand and present the nodes under a certain node by operating the certain node; and the retraction operation is to hide the nodes under the certain node. Of course, the node in the knowledge graph can also include other operations. Various operations can be predicted as the information content of the first operation information.

[0078] When the information content of the first operation information is node expansion operation information, the operation performed is to expand the first candidate triple corresponding to the first operation information; when the first operation information is node retraction operation information, the operation performed is to retract the first candidate triple corresponding to the first operation information. By analogy, the corresponding operation can be performed on each first candidate triple, or each node, in the initial graph summary. Because the first operation information is predicted to be the operation that the target user can perform, after the operation on the initial graph summary is completed, the current version of the target graph summary is obtained.

[0079] In addition, it should be noted that, Figures 1 to 4 The method involved in the embodiments shown can be executed repeatedly in actual application, so that the target graph summary is iteratively updated for multiple versions, so as to adapt to different target users, or to the interests and habits of the same target user at different times. That is, the current version of the target graph summary in the knowledge graph can be determined as the initial graph summary of the next version of the knowledge graph. Then the method involved in the above embodiments is re-executed to obtain the target graph summary of the next version, and so on.

[0080] According to the actual operation of the target user on the target graph summary, it can be verified in reverse whether the prediction of the operation of the summary determination network on the target user is accurate, so that the summary determination network continues to learn and train in use. For example, if the target user does not perform any operation during the browsing of the target graph summary generated for him, it means that the target graph summary generated based on the prediction is accurate and meets the requirements of the target user. On the contrary, if the target user performs some operations, it means that the prediction result has certain deviation. The actual graph summary obtained by the target user after actual operation represents the knowledge graph summary actually required by the target user, that is, the "correct answer" of the prediction. Therefore, by the difference between the actual graph summary and the target graph summary, the loss in the prediction process of the summary determination network can be determined. Using this loss, the summary determination network can be further trained and updated, so as to improve the accuracy of the prediction and make it more meet the needs of the target user.

[0081] In fact, each target user will have his own unique habits and needs. During the multiple browsing of the target graph summary by the target user, the summary determination network can be continuously improved according to the operation of the target user, that is, the so-called online learning in the art. So as to further train the unique summary determination network that meets the target user. If it is an offline scene, the loop execution is to enable the summary determination network to learn from the operation / browsing records of the target user in the past period of time how to determine the initial graph summary of the next version. If it is an online scene, the loop execution is to enable the target graph summary to adapt to the real-time interest changes of the target user, so as to adapt to the interests and habits of the same target user at different times.

[0082] It should be noted that the summary determination network involved in the above embodiments will be pre-trained. The specific training process is not limited in the present application, and any training method that can achieve the same or similar effect can be combined in the overall technical solution of the present application.

[0083] For example, a supervised learning method can be used, that is, using the historical operation of the user on the knowledge graph as a labeled sample to supervise the learning and training of the first operation information predicted by the network.

[0084] Or it can also use an online learning training method, that is, according to the actual operation of the target user on the current version of the target graph summary or the target graph summary calculated by the subsequent version (including the target graph summary calculated by other networks in the subsequent version), it is inversely evaluated whether the effect of the current version of the target graph summary is ideal, so as to score each first operation information, determine the convergence degree of the network in the current version, and then realize training.

[0085] For example, Figure 5As shown, the following will be specifically described a training method for the summary determination network for reference:

[0086] Step 501, determine a current summary sample of a current version of the knowledge graph, and a second candidate triple corresponding to the current summary sample.

[0087] Similar to the above summary determination method, in the process of training the summary determination network, the summary of a version of the knowledge graph will be used as a current summary sample for training. The current summary sample can come from the historical log of the knowledge graph. Further, each triple in the current summary sample can be determined, which can be referred to as a second candidate triple here. In the present training method, the current summary sample can be represented as Gn, and any second candidate triple can be represented as cn.

[0088] In the historical log, the historical user can actually (at the historical time) perform some operations on the current summary sample, thereby obtaining a next version of the summary sample. The historical user actually performs operations on the current summary sample, which can be referred to as historical operations and represented by At. In some cases, the historical user and the target user described above can be the same user. That is, the training is performed according to the historical operations of the user.

[0089] It should be noted that because the historical log, that is, the browsing / operation records of the historical user in the past, is used in the training process of the network to be trained, the first operation information output by the summary determination network obtained by training can be more in line with the browsing / operation habits of the historical user (i.e., the target user) in the past, and can also be said to take into account the interests and needs of the user.

[0090] Step 502, obtain a user portrait sample.

[0091] Similarly, the user portrait sample of the historical user performing the historical operation will also be obtained in the training process. The specific content of the user portrait sample can be the same as the user portrait of the target user described above. This will not be repeated here. The user portrait sample will be represented by Bn.

[0092] Step 503, use the network to be trained to determine the second operation information corresponding to the second candidate triple according to the current summary sample and the user portrait sample.

[0093] The network to be trained is the form before the training of the summary determination network is completed. Its network structure is consistent with that of the summary determination network, but the specific network parameters will be gradually adjusted in the training process. That is, the network structure of the network to be trained can also refer to Figure 3 as shown.

[0094] In the training process, the to-be-trained network is inputted with the current summary sample Gn, the user portrait sample Bn and each second candidate triple cn represented as En, so as to predict the operation to be performed by the historical user on each second candidate triple, that is, to obtain the second operation information. The specific network calculation process and the prediction process are similar to step 103 described above. Specifically, that is:

[0095] The current summary sample Gn is inputted into the first convolutional network structure of the to-be-trained network, to obtain the second summary feature Φ G (Gn). The user portrait sample Bn is inputted into the first feedforward network structure of the to-be-trained network, to obtain the second user feature Φ B (Bn). Any second candidate triple cn is inputted into the second feedforward network structure of the to-be-trained network, to obtain the second triple feature Φ c (cn).

[0096] Then, according to the second summary feature Φ G (Gn), the second user feature Φ B (Bn) and the second triple feature Φ c (cn), the second feature map corresponding to the second candidate triple can be determined. Specifically, the above three can be spliced by using a splicing function to obtain the second feature map. The mathematical representation is: En = concat(Φ G (Gn), Φ B (Bn), Φ c (cn)); wherein, En represents the second feature map, and concat represents the splicing function.

[0097] According to the second feature map, the probability of the historical user performing various operations can be predicted. That is, the second feature map is inputted into the third feedforward network structure of the to-be-trained network, to obtain the operation probability of the user performing each candidate operation. The operation probability of the unfolding operation can be represented as Q’1, and the operation probability of the withdrawing operation can be represented as Q’2. According to the operation probability of each operation, the second operation information can be determined; that is, the candidate operation with the largest operation probability is determined as the information content of the second operation information. In other words, the information content of the second operation information is the operation to be performed by the historical user. For example, in the embodiment, Q’1>Q’2, which means that the to-be-trained network predicts that the historical user is likely to perform the unfolding operation on the second candidate triple.

[0098] Step 504, determining the target label corresponding to the current summary sample.

[0099] In reality, since the training of the network to be trained is not yet complete, its predictions may be correct or incorrect. However, the historical logs contain historical operation At, which are the operations actually performed by the user in the past, and can verify whether the second operation information predicted by the network to be trained is correct.

[0100] Alternatively, in other cases, other computational networks (such as those that can be reasonably considered capable of accurately predicting historical user actions) can be used to calculate the target operation As. The target operation As can also be used to verify whether the second operation information predicted by the network being trained is correct.

[0101] During the training process described above, the historical operation At or the target operation As can be used as the target label corresponding to the current summary sample.

[0102] Step 505: Using the second operation information and target labels, train the network to be trained on data, and determine the network to be trained after completing the data training as the summary determination network.

[0103] Using the target label and the second operation information, the network to be trained can be trained on data. For example, in the above example, the network to be trained predicts that a historical user might perform an unfolding operation on the second candidate triplet. The target label can be used to confirm whether this is actually the case. If the historical operation At or the target operation As in the target label is also an unfolding operation, it means that the second operation information prediction is correct, and a positive score R can be assigned to the second operation information. Conversely, if the historical operation At or the target operation As in the target label is not an unfolding operation, it means that the second operation information prediction is incorrect, and a negative score -R can be assigned to the second operation information.

[0104] Similarly, each second operation information corresponding to each second candidate triplet can obtain a similar score. A higher overall score for all second candidate triplets indicates more accurate predictions from the network to be trained; conversely, a lower overall score indicates lower accuracy. Therefore, gradient descent can be performed on the network to be trained based on the overall score to achieve data training. The trained network is then used as the summary determination network.

[0105] like Figure 6 The image shown is a specific embodiment of the knowledge graph summarization device described in this application. This embodiment of the device is used to perform... Figures 1 to 4 The physical apparatus of the method is essentially the same as that in the above embodiments, and the corresponding descriptions in the above embodiments also apply to this embodiment. The apparatus in this embodiment includes:

[0106] The initial abstract determination module 601 is configured to determine an initial graph abstract of a current version of the knowledge graph, and a first candidate triple corresponding to the initial graph abstract information.

[0107] The user portrait determination module 602 is configured to obtain a user portrait of a target user.

[0108] The first operation information determination module 603 is configured to determine first operation information corresponding to the first candidate triple according to the initial graph abstract and the user portrait.

[0109] The target abstract determination module 604 is configured to determine a target graph abstract of the current version of the knowledge graph according to the initial graph abstract and the first operation information.

[0110] In addition, based on the embodiment shown in the embodiment, preferably, the method further comprises: Figure 5

[0111] The initial abstract determination module 601 comprises:

[0112] The node determination unit 611 is configured to determine a first node included in the initial graph abstract, and a second node corresponding to the first node.

[0113] The first candidate triple determination module is configured to determine a plurality of first candidate triples according to the first node and the second node.

[0114] The first operation information determination module 603 comprises:

[0115] The first abstract feature determination unit 631 is configured to determine a first abstract feature according to the initial graph abstract.

[0116] The first user feature determination unit 632 is configured to determine a first user feature according to the user portrait.

[0117] The first triple feature determination unit 633 is configured to determine a first triple feature according to the first candidate triple.

[0118] The first operation information determination unit 634 is configured to determine first operation information corresponding to the first candidate triple according to the first abstract feature, the first user feature, and the first triple feature.

[0119] The first operation information determination unit 634 comprises:

[0120] The first feature map determination sub-unit 6341 is configured to determine a first feature map corresponding to the first candidate triple according to the first abstract feature, the first user feature, and the first triple feature.

[0121] ​The operation determination sub-unit 6342 is configured to determine, according to the first feature map, an operation probability of the target user for each candidate operation; and determine the first operation information according to the operation probability of each candidate operation.

[0122] The target abstract determination module 604 includes:

[0123] The expansion operation unit 641 is configured to, when the first operation information is node expansion operation information, expand the first candidate triple corresponding to the first operation information.

[0124] The retraction operation unit 642 is configured to, when the first operation information is node retraction operation information, retract the first candidate triple corresponding to the first operation information.

[0125] In addition to the method and the device described above, the embodiments of the present application can also be a computer program product, which includes computer program instructions, and the computer program instructions enable the processor to perform the steps in the method according to various embodiments of the present application described in the above “Exemplary Method” section of the present specification when the computer program instructions are run by the processor.

[0126] The computer program product can be written in any combination of one or more programming languages, including an object-oriented programming language, such as Java, C++, and the like, and conventional procedural programming languages, such as the "C" programming language, or the like. The program code can execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server.

[0127] In addition, the embodiments of the present application can also be a computer readable storage medium, which stores computer program instructions, and the computer program instructions enable the processor to perform the steps in the method according to various embodiments of the present application described in the above “Exemplary Method” section of the present specification when the computer program instructions are run by the processor.

[0128] The computer readable storage medium can be a combination of one or more computer readable media. The computer readable media can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium can include, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium include the following: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0129] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0130] The block diagrams of the devices, apparatuses, equipment, systems referred to in the present application are only illustrative examples and are not intended to require or imply that the connection, arrangement, configuration must be as shown in the block diagrams. As those skilled in the art will recognize, the devices, apparatuses, equipment, systems can be connected, arranged, configured in any manner. Words such as "include", "contain", "have", etc. are open-ended words, mean "including but not limited to", and can be used interchangeably with each other. The words "or" and "and" used herein mean the word "and / or", and can be used interchangeably with each other, unless the context clearly indicates otherwise. The word "such as" used herein means the phrase "such as but not limited to", and can be used interchangeably with each other.

[0131] It should also be noted that in the apparatuses, equipment and methods of the present application, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions of the present application.

[0132] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0133] The foregoing description has been presented for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of the application to the forms disclosed herein. Although various example aspects and embodiments have been discussed above, those of ordinary skill in the art will appreciate a variety of modifications, alternatives, permutations, additions, and sub-combinations, which fall within the scope of the application.

Claims

1. A method for determining a knowledge graph summary, comprising: determining an initial graph summary of a current version of a knowledge graph, and first candidate triples corresponding to the initial graph summary; obtaining a user portrait of a target user; using a pre-trained summary determination network, predicting each of the first candidate triples according to the initial graph summary and the user portrait, and determining first operation information corresponding to each of the first candidate triples; determining a target graph summary of the current version of the knowledge graph according to the initial graph summary and the first operation information; the determining of the first operation information corresponding to each of the first candidate triples according to the initial graph summary and the user portrait comprises: determining first summary features according to the initial graph summary; determining first user features according to the user portrait; determining first triple features according to the first candidate triples; determining the first operation information corresponding to each of the first candidate triples according to the first summary features, the first user features and the first triple features. 2.The method of claim 1, wherein the determining of the first candidate triples corresponding to the initial graph summary comprises: determining first nodes included in the initial graph summary; second nodes corresponding to the first nodes; and determining a plurality of the first candidate triples according to the first nodes and the second nodes. 3.The method of claim 1, wherein the determining of the first operation information corresponding to each of the first candidate triples according to the first summary features, the first user features and the first triple features comprises: determining a first feature map corresponding to each of the first candidate triples according to the first summary features, the first user features and the first triple features; determining operation probabilities of each of the first candidate triples according to the first feature map; determining the first operation information according to the operation probabilities of each of the first candidate triples. 4.The method of any one of claims 1 to 3, wherein the determining of the target graph summary of the current version of the knowledge graph according to the initial graph summary and the first operation information comprises: performing a first operation on each of the first candidate triples corresponding to the initial graph summary according to the first operation information of each of the first candidate triples, to determine the target graph summary.

5. The method of claim 4, wherein the first operation information includes node deployment operation information or node retraction operation information. the performing of the first operation on each of the first candidate triples corresponding to the initial graph summary comprises: when the first operation information is node expansion operation information, expanding the first candidate triple corresponding to the first operation information; and when the first operation information is node retraction operation information, retracting the first candidate triple corresponding to the first operation information. 6.The method of any one of claims 1 to 3, further comprising: determining the target graph summary of the current version of the knowledge graph as an initial graph summary of a next version of the knowledge graph. 7.A device for determining a knowledge graph summary, comprising: An initial summary determination module is configured to determine an initial graph summary of a current version of a knowledge graph, and a first candidate triple corresponding to the initial graph summary information; A user portrait determination module is configured to obtain a user portrait of a target user; A first operation information determination module is configured to, according to the initial graph summary and the user portrait, utilize a pre-trained summary determination network to predict the first candidate triples one by one, and determine first operation information corresponding to each of the first candidate triples; A target summary determination module is configured to determine a target graph summary of the current version of the knowledge graph according to the initial graph summary and the first operation information; The first operation information determination module is further configured to determine first summary features according to the initial graph summary, determine first user features according to the user portrait, determine first triple features according to the first candidate triples, and determine first operation information corresponding to each of the first candidate triples according to the first summary features, the first user features, and the first triple features. 8.A computer readable storage medium, the storage medium storing a computer program, the computer program being configured to execute the knowledge graph summary determination method according to any one of claims 1-6. 9.An electronic device, comprising: a processor; a memory for storing executable instructions of the processor; the processor is configured to read the executable instructions from the memory and execute the instructions to implement the knowledge graph summary determination method according to any one of claims 1-6.

Images