Data completion method and device, equipment and storage medium

Abstract

The invention discloses a data completion method and device, equipment and a storage medium, and the method comprises the steps: obtaining a plurality of tetrad association features in a multi-association knowledge graph of to-be-trained data, each tetrad association feature comprises a moment, a head entity, a relationship and a tail entity, considering the dynamic change of a to-be-trained entity in the to-be-trained data, and obtaining a plurality of to-be-trained entities in the to-be-trained data; mining a plurality of tetrad association features formed based on moments, head entities, relationships and tail entities; for each tetrad association feature, under the condition of tetrad increment structure knowledge corresponding to the moment, the head entity, the relationship and the tail entity, calculating a tetrad establishment probability of the tetrad association feature, and representing establishment possibilities of the head entity, the tail entity and a dynamic relationship between the head entity and the tail entity; based on a plurality of tetrad establishment probabilities corresponding to the plurality of tetrad association features, accurately determining entity association features from the plurality of tetrad association features; and complementing the to-be-trained data based on the entity association features to obtain more accurate complemented data.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a data completion method, device, equipment and storage medium. Background Art

[0002] At present, in artificial intelligence scenarios, pre-trained models are generally constructed through a large amount of data to be trained. When there is missing data in the data to be trained, it is necessary to complete the data to be trained to obtain completed data in order to more accurately construct the pre-trained model and improve the accuracy of the pre-trained model.

[0003] In the related art, the data completion method for the data to be trained refers to: for the entities to be trained in the data to be trained, based on the fixed neighborhood entities or random adjacent entities of the entities to be trained in a single static knowledge graph of the data to be trained, completing the data to be trained to obtain the completed data.

[0004] However, the association relationship between the entities to be trained in the training data changes dynamically. The above data completion method relies on a single static knowledge graph, which easily leads to the completed data obtained by completing the data to be trained being inaccurate, thereby causing the pre-trained model built based on the completed data to be inaccurate. Summary of the invention

[0005] In view of this, the embodiments of the present application provide a data completion method, apparatus, device and storage medium to accurately complete the data to be trained to obtain completed data, so that the completed data is more accurate, thereby improving the accuracy of the pre-trained model constructed based on the completed data.

[0006] On the one hand, an embodiment of the present application provides a data completion method, the method comprising:

[0007] Obtain multiple quadruple association features in the multi-association knowledge graph of the data to be trained; each quadruple association feature includes time, head entity, relationship and tail entity;

[0008] For each of the four-tuple association features, probability calculation is performed on the four-tuple association features under the condition of the four-tuple incremental structure knowledge corresponding to the time, the head entity, the relationship, and the tail entity to obtain the four-tuple establishment probability of the four-tuple association features;

[0009] Determining an entity association feature from the multiple quadruple association features according to the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features;

[0010] The data to be trained is completed according to the entity association features to obtain completed data.

[0011] Optionally, the performing probability calculation on the quadruple association feature under the condition of the quadruple incremental structure knowledge corresponding to the time, the head entity, the relationship, and the tail entity to obtain the quadruple establishment probability of the quadruple association feature includes:

[0012] Under the condition of the quadruple incremental structural knowledge, a probability calculation is performed on the head entity in the quadruple association feature to obtain the head entity establishment probability of the head entity;

[0013] Under the conditions of the quadruple incremental structure knowledge and the head entity, a probability calculation is performed on the tail entity in the quadruple association feature to obtain a tail entity establishment probability of the tail entity;

[0014] Under the conditions of the quadruple incremental structure knowledge, the head entity and the tail entity, probability calculation is performed on the relationship in the quadruple association feature to obtain the relationship establishment probability of the relationship;

[0015] The probability of the four-tuple association feature is calculated according to the head entity establishment probability, the tail entity establishment probability and the relationship establishment probability to obtain the four-tuple establishment probability.

[0016] Optionally, determining the entity association feature from the multiple quadruple association features according to the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features includes:

[0017] According to the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features, the quadruple association feature corresponding to the maximum establishment probability is determined from the multiple quadruple association features as the entity association feature.

[0018] Optionally, the quadruple incremental structural knowledge includes the incremental structural knowledge of the head entity corresponding to the quadruple incremental structural knowledge, the global structural knowledge of the relationship, and the global structural knowledge of the tail entity.

[0019] Optionally, the incremental structural knowledge of the head entity includes incremental attribute knowledge and incremental relationship knowledge of the head entity; the global structural knowledge of the tail entity includes global attribute knowledge and global relationship knowledge of the tail entity.

[0020] Optionally, the method further includes:

[0021] For each entity to be trained in the completed data, incremental attribute knowledge and known attribute knowledge of the entity to be trained are obtained; the incremental attribute knowledge indicates that the incremental attribute of the entity to be trained has an incremental attribute value, and the known attribute knowledge indicates that the known attribute of the entity to be trained has a known attribute value;

[0022] Under the condition that the known attribute of the entity to be trained has the known attribute value, a probability calculation is performed on the incremental attribute of the entity to be trained having the incremental attribute value to obtain the attribute value establishment probability that the incremental attribute of the entity to be trained has the incremental attribute value;

[0023] If the probability of the attribute value being true is greater than or equal to the preset probability of being true, performing a correlation calculation based on the incremental attribute value and the known attribute value to obtain the attribute value correlation between the incremental attribute value and the known attribute value;

[0024] If the attribute value correlation is greater than or equal to a preset correlation, attribute compression is performed on the global attribute knowledge of the entity to be trained to obtain compressed attribute knowledge of the entity to be trained.

[0025] Optionally, under the condition that the known attribute of the entity to be trained has the known attribute value, performing probability calculation on the incremental attribute of the entity to be trained having the incremental attribute value to obtain the attribute value establishment probability that the incremental attribute of the entity to be trained has the incremental attribute value includes:

[0026] Based on the attribute value conditional probability that the incremental attribute has the incremental attribute value under the condition that the known attribute has the known attribute value, and the attribute correlation between the incremental attribute and the known attribute, the probability that the incremental attribute of the entity to be trained has the incremental attribute value is calculated to obtain the probability that the attribute value is established.

[0027] On the other hand, an embodiment of the present application provides a data completion device, the device comprising: an acquisition unit, a calculation unit, a determination unit and a completion unit;

[0028] The acquisition unit is used to acquire multiple quadruple association features in the multi-association knowledge graph of the training data; each quadruple association feature includes a time, a head entity, a relationship and a tail entity;

[0029] The calculation unit is used to perform probability calculation on each of the four-tuple association features under the condition of the four-tuple incremental structure knowledge corresponding to the time, the head entity, the relationship and the tail entity to obtain the four-tuple conditional probability of the four-tuple association feature;

[0030] The determining unit is used to determine the entity association feature from the multiple quadruple association features according to the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features;

[0031] The completion unit is used to complete the data to be trained according to the entity association features to obtain completed data.

[0032] Optionally, the computing unit is used to:

[0033] Under the condition of the quadruple incremental structural knowledge, a probability calculation is performed on the head entity in the quadruple association feature to obtain the head entity establishment probability of the head entity;

[0034] Under the conditions of the quadruple incremental structure knowledge and the head entity, a probability calculation is performed on the tail entity in the quadruple association feature to obtain a tail entity establishment probability of the tail entity;

[0035] Under the conditions of the quadruple incremental structure knowledge, the head entity and the tail entity, probability calculation is performed on the relationship in the quadruple association feature to obtain the relationship establishment probability of the relationship;

[0036] The probability of the four-tuple association feature is calculated according to the head entity establishment probability, the tail entity establishment probability and the relationship establishment probability to obtain the four-tuple establishment probability.

[0037] Optionally, the determining unit is used to:

[0038] According to the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features, the quadruple association feature corresponding to the maximum establishment probability is determined from the multiple quadruple association features as the entity association feature.

[0039] Optionally, the quadruple incremental structural knowledge includes the incremental structural knowledge of the head entity corresponding to the quadruple incremental structural knowledge, the global structural knowledge of the relationship, and the global structural knowledge of the tail entity.

[0040] Optionally, the incremental structural knowledge of the head entity includes incremental attribute knowledge and incremental relationship knowledge of the head entity; the global structural knowledge of the tail entity includes global attribute knowledge and global relationship knowledge of the tail entity.

[0041] Optionally, the device further comprises: a compression unit;

[0042] The acquisition unit is further used to acquire, for each entity to be trained in the completed data, incremental attribute knowledge and known attribute knowledge of the entity to be trained; the incremental attribute knowledge indicates that the incremental attribute of the entity to be trained has an incremental attribute value, and the known attribute knowledge indicates that the known attribute of the entity to be trained has a known attribute value;

[0043] The calculation unit is further used to calculate the probability that the incremental attribute of the entity to be trained has the incremental attribute value under the condition that the known attribute of the entity to be trained has the known attribute value, so as to obtain the attribute value establishment probability that the incremental attribute of the entity to be trained has the incremental attribute value;

[0044] The calculation unit is further configured to, if the probability of the attribute value being true is greater than or equal to a preset probability of true, perform a correlation calculation based on the incremental attribute value and the known attribute value to obtain the attribute value correlation between the incremental attribute value and the known attribute value;

[0045] The compression unit is used to perform attribute compression on the global attribute knowledge of the entity to be trained if the attribute value correlation is greater than or equal to a preset correlation, so as to obtain compressed attribute knowledge of the entity to be trained.

[0046] Optionally, the computing unit is used to:

[0047] Based on the attribute value conditional probability that the incremental attribute has the incremental attribute value under the condition that the known attribute has the known attribute value, and the attribute correlation between the incremental attribute and the known attribute, the probability that the incremental attribute of the entity to be trained has the incremental attribute value is calculated to obtain the probability that the attribute value is established.

[0048] On the other hand, an embodiment of the present application provides a computer device, the computer device comprising a processor and a memory:

[0049] The memory is used to store a computer program and transmit the computer program to the processor;

[0050] The processor is configured to execute the method described in the above aspects according to the instructions in the computer program.

[0051] On the other hand, an embodiment of the present application provides a computer-readable storage medium, characterized in that the computer-readable storage medium is used to store a computer program, and the computer program is used to implement the method described in the above aspects when executed by a processor.

[0052] Compared with the prior art, this application has at least the following advantages:

[0053] By adopting the technical solution of the embodiment of the present application, first, multiple quadruple association features are obtained in the multi-association knowledge graph of the data to be trained, wherein each quadruple association feature includes a time, a head entity, a relationship, and a tail entity; secondly, for each quadruple association feature, under the condition of the quadruple incremental structural knowledge corresponding to the time, the head entity, the relationship, and the tail entity, the quadruple establishment probability of the quadruple association feature is calculated; then, based on the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features, the entity association feature is determined from the multiple quadruple association features; finally, the data to be trained is completed based on the entity association features to obtain the completed data. It can be seen that this method considers the dynamic changes of the entities to be trained in the training data through the multi-association knowledge graph of the training data, and mines multiple four-tuple association features formed based on time, head entity, relationship and tail entity; and under the condition of the four-tuple incremental structural knowledge corresponding to the time, head entity, relationship and tail entity, calculates the four-tuple establishment probability of each four-tuple association feature, indicating the possibility of the establishment of the head entity, the tail entity, and the dynamic relationship between the head entity and the tail entity in each four-tuple association feature; accurately selects entity association features from multiple four-tuple association features, so as to accurately complete the data to be trained through the entity association features to obtain the completed data, so that the completed data is more accurate, thereby improving the accuracy of the pre-trained model built based on the completed data. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0055] Figure 1 A schematic diagram of a system framework involved in an application scenario in an embodiment of the present application;

[0056] Figure 2 A flowchart of a data completion method provided in an embodiment of the present application;

[0057] Figure 3 A schematic diagram of the structure of a multi-association knowledge graph provided in an embodiment of the present application;

[0058] Figure 4 A schematic diagram of the structure of a quadruple incremental structure knowledge provided in an embodiment of the present application;

[0059] Figure 5 A schematic diagram of the structure of a data completion device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0060] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0061] At present, when constructing a pre-trained model through data to be trained with missing data, it is usually firstly aimed at the entities to be trained in the data to be trained, based on the fixed neighborhood entities or random adjacent entities of the entities to be trained in the single static knowledge graph of the data to be trained, to complete the data to be trained to obtain completed data, and then build a pre-trained model based on the completed data. However, after research, the inventors found that the association relationship between the entities to be trained in the data to be trained is dynamically changing, and relying on a single static knowledge graph can easily lead to the completed data obtained by completing the data to be trained being inaccurate, thereby leading to the pre-trained model constructed based on the completed data being inaccurate.

[0062] In order to solve this problem, in an embodiment of the present application, the dynamic changes of the entities to be trained in the training data are considered through the multi-association knowledge graph of the data to be trained, and multiple four-tuple association features formed based on time, head entity, relationship and tail entity are mined; and under the condition of the four-tuple incremental structure knowledge corresponding to the time, head entity, relationship and tail entity, the four-tuple establishment probability of each four-tuple association feature is calculated, indicating the possibility of the establishment of the head entity, the tail entity, and the dynamic relationship between the head entity and the tail entity in each four-tuple association feature; entity association features are accurately selected from multiple four-tuple association features, so as to accurately complete the data to be trained through the entity association features to obtain the completed data, so that the completed data is more accurate, thereby improving the accuracy of the pre-trained model constructed based on the completed data.

[0063] For example, one of the scenarios of the embodiments of the present application may be applied to Figure 1 In the scenario shown. The scenario includes a terminal 101 and a server 102, the terminal 101 determines the data to be trained and sends it to the server, the server 102 executes the implementation method provided by the embodiment of the present application, completes the data to be trained to obtain completed data, and builds a pre-trained model based on the completed data.

[0064] First, in the above application scenario, although the action description of the implementation method provided in the embodiment of the present application is executed by the server 102; however, the embodiment of the present application is not limited in terms of the execution subject, as long as the actions disclosed in the implementation method provided in the embodiment of the present application are executed.

[0065] Secondly, the above scenario is only an example scenario provided by the embodiment of the present application, and the embodiment of the present application is not limited to this scenario.

[0066] The specific implementation of the data completion method, device, equipment and storage medium in the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0067] First, the specific implementation method of the data completion method in the embodiment of the present application is described in detail through an embodiment.

[0068] See also Figure 2 ,Should Figure 2 A flowchart of a data completion method provided in an embodiment of the present application is shown in FIG.

[0069] S201: Obtain multiple quadruple association features in the multi-association knowledge graph of the data to be trained; each quadruple association feature includes a time, a head entity, a relationship, and a tail entity.

[0070] Because in the related technology, for the entities to be trained in the data to be trained, the fixed neighborhood entities or random adjacent entities of the entities to be trained in the single static knowledge graph of the data to be trained are used to complete the data to be trained to obtain the completed data, without considering that the association relationship between the entities to be trained in the data to be trained is dynamically changing. This relies on a single static knowledge graph, which may easily lead to the problem that the completed data obtained by completing the data to be trained is not accurate enough, thereby causing the pre-trained model built based on the completed data to be not accurate enough.

[0071] Therefore, in an embodiment of the present application, in order to solve the above problems, the dynamic changes of the entities to be trained in the training data are considered through the multi-association knowledge graph of the data to be trained, and multiple four-tuple association features formed based on time, head entity, relationship and tail entity are mined; that is, multiple four-tuple association features are obtained in the multi-association knowledge graph of the data to be trained, wherein each four-tuple association feature includes time, head entity, relationship and tail entity.

[0072] Among them, the multi-association knowledge graph of the data to be trained includes multiple knowledge graphs with association relationships. The multi-association knowledge graph has rich in-and-out degree information and path information, and the information will continue to change and enrich over time; each quadruple association feature means that at each moment in the quadruple association feature, the head entity links the tail entity through a relationship, and the head entity and the tail entity belong to different knowledge graphs in the multi-association knowledge graph.

[0073] See also Figure 3 ,Should Figure 3A schematic diagram of the structure of a multi-association knowledge graph provided in an embodiment of the present application. The multi-association knowledge graph includes multiple knowledge graphs KG1, KG2, KG3 and KG4 with association relationships. 1 Node starts to observe its link structure: within the range of “one-hop path”, h 1 There are 4 out-degree nodes t 1 ,t 2 ,t 3 、h 2 ,h 1 There are 4 outgoing edges r 1 、r 2 、r 3 、r 4 ,h 1 There is 1 in-degree node h 0 ,h 1 There is 1 incoming edge r 0 ; In the "Multiple Paths" range, h 1 There are 4 two paths r 4 →r 5 、r 4 →r 6 、r 4 →r 7 、r 4 →r 8 ,h 1 There are 4 profile related entities t 4 ,t 5 ,t 10 、h 3 ,h 1 There is 1 intermediary entity h 2 ,h 1 There are 5 three paths r 4 →r 8 →r 9 、r 4 →r 8 →r 10 、r 4 →r 8 →r 11 、r 4 →r 8 →r 12 、r 4 →r 8 →r 13 ,h 1 There are 5 profile related entities t 6 ,t 7 ,t 8 ,t 9 ,t 10 ,h 1 There are two intermediary entities h 2 、h 3KG1, KG2, KG3 and KG4 will make the entity h 1 The number of occurrences of h increases, due to 1 One more outgoing degree h 2 The local structure has been changed and a new global structure "multi-hop" path has been added, indicating that the relationship between entities in the multi-association knowledge graph is dynamically changing.

[0074] As an example of S201, multiple quadruple association features in the multi-association knowledge graph G of the training data are obtained as multiple (h, r, t, τ), h∈G in the knowledge graph G i , t∈G in the knowledge graph G j , G i ≠G j , where each (h, r, t, τ) includes time τ, head entity h, relation r, and tail entity t.

[0075] S202: For each quadruple association feature, a probability calculation is performed on the quadruple association feature under the condition of quadruple incremental structure knowledge corresponding to the time, head entity, relationship and tail entity to obtain the quadruple establishment probability of the quadruple association feature.

[0076] In an embodiment of the present application, after executing S201 to obtain multiple quadruple association features based on time, head entity, relationship and tail entity in the multi-association knowledge graph of the data to be trained, it is also necessary to clarify the possibility of the establishment of the head entity, the tail entity, and the dynamic relationship between the head entity and the tail entity in each quadruple association feature; therefore, for each quadruple association feature, under the condition of the quadruple incremental structure knowledge corresponding to the time, head entity, relationship and tail entity, the quadruple establishment probability of the quadruple association feature is calculated.

[0077] Among them, the quadruple incremental structural knowledge corresponding to the moment, head entity, relationship and tail entity refers to the structural knowledge added by the head entity at the moment in the quadruple association feature, linking the tail entity through the relationship; the quadruple establishment probability of the quadruple association feature refers to the possibility of the establishment of the head entity, the tail entity, and the dynamic relationship between the head entity and the tail entity in the quadruple association feature.

[0078] As an example of S202, based on the above S201 example, for each (h, r, t, τ), the probability of (h, r, t, τ) is calculated under the condition of the four-tuple incremental structure knowledge C(h, r, t, τ) corresponding to τ, h, r and t, and the probability of the four-tuple establishment of (h, r, t, τ) is obtained as P((h, r, t, τ)|C(h, r, t, τ); θ), where θ represents the space of G.

[0079] S203: Determine an entity association feature from the multiple quadruple association features according to the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features.

[0080] In an embodiment of the present application, after executing the above-mentioned S202 to calculate the quadruple establishment probability of the head entity, the tail entity, and the possibility of the dynamic relationship between the head entity and the tail entity in each quadruple association feature, since the larger the quadruple establishment probability corresponding to the quadruple association feature, the greater the possibility of the dynamic relationship between the head entity, the tail entity, and the tail entity in the quadruple association feature; therefore, based on the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features, the entity association features can be accurately determined from the multiple quadruple association features.

[0081] The entity association feature refers to a quadruple association feature with a high probability of establishing a head entity, a tail entity, and a dynamic relationship between the head entity and the tail entity among multiple quadruple association features.

[0082] S204: Complete the training data according to the entity association features to obtain completed data.

[0083] In the embodiment of the present application, after executing the above S203 to determine the entity association feature with a high possibility of establishing a dynamic relationship between the head entity and the tail entity from multiple quadruple association features, the data to be trained can be accurately completed based on the entity association feature to obtain the completed data.

[0084] The completed data refers to the data to be trained after completion based on entity-related features.

[0085] Through the various implementation methods provided in this embodiment, the data completion method refers to: obtaining multiple quadruple association features in the multi-association knowledge graph of the data to be trained, wherein each quadruple association feature includes a time, a head entity, a relationship, and a tail entity; for each quadruple association feature, under the condition of the quadruple incremental structural knowledge corresponding to the time, the head entity, the relationship, and the tail entity, calculating the quadruple establishment probability of the quadruple association feature; based on the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features, determining the entity association feature from the multiple quadruple association features; completing the data to be trained based on the entity association features to obtain the completed data. It can be seen that this method considers the dynamic changes of the entities to be trained in the training data through the multi-association knowledge graph of the training data, and mines multiple four-tuple association features formed based on time, head entity, relationship and tail entity; and under the condition of the four-tuple incremental structural knowledge corresponding to the time, head entity, relationship and tail entity, calculates the four-tuple establishment probability of each four-tuple association feature, indicating the possibility of the establishment of the head entity, the tail entity, and the dynamic relationship between the head entity and the tail entity in each four-tuple association feature; accurately selects entity association features from multiple four-tuple association features, so as to accurately complete the data to be trained through the entity association features to obtain the completed data, so that the completed data is more accurate, thereby improving the accuracy of the pre-trained model built based on the completed data.

[0086] In an embodiment of the present application, in the above S202, for each quadruple association feature, under the condition of the quadruple incremental structural knowledge corresponding to the time, head entity, relationship and tail entity, the quadruple establishment probability of the quadruple association feature is calculated specifically, considering that the quadruple incremental structural knowledge corresponding to the time, head entity, relationship and tail entity refers to the structural knowledge added by the head entity at the time in the quadruple association feature by linking the tail entity through the relationship, and the head entity is established first, then the tail entity is established, and finally the relationship is established, which indicates the establishment of the dynamic relationship between the head entity and the tail entity in the quadruple association feature; based on this, it is necessary to first calculate the head entity establishment probability of the head entity in the quadruple association feature under the condition of the quadruple incremental structural knowledge, and then calculate the tail entity establishment probability of the tail entity in the quadruple association feature under the condition of the quadruple incremental structural knowledge and the head entity, and finally calculate the relationship establishment probability of the relationship in the quadruple association feature under the conditions of the quadruple incremental structural knowledge, the head entity and the tail entity, so as to calculate the quadruple establishment probability of the quadruple association feature through the head entity establishment probability, the tail entity establishment probability and the relationship establishment probability. Therefore, in an optional implementation of the embodiment of the present application, S202 may include, for example, the following S202a-S202d (not shown in the figure).

[0087] S202a: Under the condition of the quadruple incremental structural knowledge, the probability of the head entity in the quadruple association feature is calculated to obtain the head entity establishment probability of the head entity.

[0088] S202b: Based on the quadruple incremental structure knowledge and the head entity, the probability of the tail entity in the quadruple association feature is calculated to obtain the tail entity establishment probability of the tail entity.

[0089] S202c: Based on the quadruple incremental structure knowledge, the head entity and the tail entity, the probability of the relationship in the quadruple association feature is calculated to obtain the probability of the relationship being established.

[0090] S202d: Probability calculation is performed on the quadruple association features according to the head entity establishment probability, the tail entity establishment probability and the relationship establishment probability to obtain the quadruple establishment probability.

[0091] Among them, the quadruple incremental structural knowledge corresponding to the moment, head entity, relationship and tail entity refers to the structural knowledge added by the head entity at the moment in the quadruple association feature, linking the tail entity through the relationship. Considering that at the moment in the quadruple association feature, the head entity may have increased structural knowledge, and the relationship and tail entity are added, then in an optional implementation of the embodiment of the present application, the quadruple incremental structural knowledge needs to include the incremental structural knowledge of the head entity corresponding to the quadruple incremental structural knowledge, the global structural knowledge of the relationship, and the global structural knowledge of the tail entity.

[0092] Specifically, the structural knowledge of an entity can be divided into the attribute knowledge of the entity and the relationship knowledge of the entity. In an optional implementation of the embodiment of the present application, the incremental structural knowledge of the head entity needs to include the incremental attribute knowledge and incremental relationship knowledge of the head entity, and the global structural knowledge of the tail entity needs to include the global attribute knowledge and global relationship knowledge of the tail entity.

[0093] In addition, the head entity establishment probability of the head entity refers to the possibility of the head entity in the quadruple association feature; the tail entity establishment probability of the tail entity refers to the possibility of the tail entity in the quadruple association feature; the relationship establishment probability of the relationship refers to the possibility of the dynamic relationship between the head entity and the tail entity in the quadruple association feature.

[0094] As an example of S202a-S202d, based on the above S202 example, under the condition of C(h, r, t, τ), the probability of h in (h, r, t, τ) is calculated, and the probability of the head entity of h is obtained as P(h|C(h, r, t, τ); θ); under the conditions of C(h, r, t, τ) and h, the probability of t in (h, r, t, τ) is calculated, and the probability of the tail entity of t is obtained as P(t|C(h, r, t, τ); h, θ); under the conditions of C(h, r, t, τ), h and t, the probability of r in (h, r, t, τ) is calculated. Calculate the probability that the relationship is established at r, which is P(r|C(h, r, t, τ); h, t, θ); calculate the probability of (h, r, t, τ) based on P(h|C(h, r, t, τ); θ), P(t|C(h, r, t, τ); h, θ), and P(r|C(h, r, t, τ); h, t, θ), and obtain P((h, r, t, τ)|C(h, r, t, τ); θ) = P(h|C(h, r, t, τ); θ)·P(t|C(h, r, t, τ); h, θ)·P(r|C(h, r, t, τ); h, t, θ).

[0095] See also Figure 4 ,Should Figure 4 A schematic diagram of a structure of a quadruple incremental structure knowledge provided in an embodiment of the present application. Based on the above S202 example, C(h, r, t, τ) needs to include the incremental attribute knowledge C of h corresponding to C(h, r, t, τ). a (h) and incremental relational knowledge C r (h) r’s global structural knowledge C p (r), and t’s global attribute knowledge C a (t) and global relational knowledge C r (t).

[0096] In an embodiment of the present application, when the above S203 is specifically implemented to determine the entity association feature from multiple quadruple association features based on the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features, it is considered that the quadruple establishment probability corresponding to the quadruple association feature is the largest, indicating that the possibility of the establishment of the head entity, the tail entity, and the dynamic relationship between the head entity and the tail entity in the quadruple association feature is the largest; based on this, based on the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features, the quadruple association feature corresponding to the maximum establishment probability among the multiple quadruple association features is determined as the entity association feature. Therefore, in an optional implementation of the embodiment of the present application, S203 may, for example, include the following S203a (not shown in the figure): according to the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features, the quadruple association feature corresponding to the maximum establishment probability among the multiple quadruple association features is determined as the entity association feature.

[0097] In an embodiment of the present application, after executing the above S201-S204 to complete the data to be trained to obtain the completed data, for each entity to be trained in the completed data, the known attributes of the entity to be trained have known attribute values, and the incremental attributes of the entity to be trained may have incremental attribute values, so that the entity to be trained in the completed data has a complex and rich attribute structure; in order to quickly and efficiently build a pre-trained model based on the completed data later, it is necessary to calculate the probability that the incremental attribute of the entity to be trained has the incremental attribute value under the condition that the known attributes of the entity to be trained have known attribute values. When the attribute value establishment probability is greater than or equal to the preset establishment probability, it indicates that the incremental attribute of the entity to be trained has the incremental attribute value; in this case, the attribute value correlation between the incremental attribute value and the known attribute value of the entity to be trained is further calculated. When the attribute value correlation is greater than or equal to the preset correlation, it indicates that the incremental attribute value of the entity to be trained is highly correlated with the known attribute value, so that the global attribute knowledge of the entity to be trained can be compressed to obtain the compressed attribute knowledge of the entity to be trained, so as to simplify the global attribute knowledge of the entity to be trained, thereby quickly and efficiently building a pre-trained model based on the completed data. Therefore, in an optional implementation of the embodiment of the present application, the method may also include the following S1-S4 (not shown in the figure).

[0098] S1: For each entity to be trained in the completed data, the incremental attribute knowledge and known attribute knowledge of the entity to be trained are obtained; the incremental attribute knowledge means that the incremental attribute of the entity to be trained has an incremental attribute value, and the known attribute knowledge means that the known attribute of the entity to be trained has a known attribute value.

[0099] S2: Under the condition that the known attributes of the entity to be trained have known attribute values, the probability of the incremental attributes of the entity to be trained having incremental attribute values ​​is calculated to obtain the probability that the incremental attributes of the entity to be trained have incremental attribute values.

[0100] S3: If the probability of the attribute value being true is greater than or equal to the preset probability of being true, a correlation calculation is performed based on the incremental attribute value and the known attribute value to obtain the attribute value correlation between the incremental attribute value and the known attribute value.

[0101] S4: If the attribute value correlation is greater than or equal to the preset correlation, the global attribute knowledge of the entity to be trained is compressed to obtain the compressed attribute knowledge of the entity to be trained.

[0102] Among them, the probability that the incremental attribute of the entity to be trained has an incremental attribute value refers to the possibility that the incremental attribute of the entity to be trained has an incremental attribute value, and the preset probability refers to the lower limit probability; the attribute value correlation between the incremental attribute value and the known attribute value refers to the attribute value correlation between the incremental attribute value and the known attribute value, and the preset correlation refers to the lower limit of the correlation; the compressed attribute knowledge of the entity to be trained refers to the global attribute knowledge of the entity to be trained after compressing the incremental attribute value.

[0103] In the embodiment of the present application, when the above S2 is specifically implemented to calculate the probability of the incremental attribute of the entity to be trained having the incremental attribute value under the condition that the known attribute of the entity to be trained has the known attribute value, it is actually based on the attribute correlation between the incremental attribute and the known attribute, considering the attribute value conditional probability that the incremental attribute has the incremental attribute value under the condition that the known attribute has the known attribute value; based on this, the attribute value conditional probability that the incremental attribute has the incremental attribute value under the condition that the known attribute has the known attribute value, and the attribute correlation between the incremental attribute and the known attribute, the attribute value establishment probability that the incremental attribute of the entity to be trained has the incremental attribute value can be calculated. Therefore, in an optional implementation of the embodiment of the present application, S2 may include the following S2a (not shown in the figure): according to the attribute value conditional probability that the incremental attribute has the incremental attribute value under the condition that the known attribute has the known attribute value, and the attribute correlation between the incremental attribute and the known attribute, the incremental attribute of the entity to be trained has the incremental attribute value, and the probability of the attribute value establishment is obtained.

[0104] The attribute correlation between the incremental attribute and the known attribute refers to the attribute correlation between the incremental attribute and the known attribute.

[0105] Next, the specific implementation method of the data completion device in the embodiment of the present application is described in detail through an embodiment.

[0106] See also Figure 5 , Figure 5 A schematic diagram of the structure of a data completion device provided in an embodiment of the present application. In this embodiment, the device may specifically include: an acquisition unit 501, a calculation unit 502, a determination unit 503 and a completion unit 504;

[0107] An acquisition unit 501 is used to acquire multiple quadruple association features in a multi-association knowledge graph of the training data; each quadruple association feature includes a time, a head entity, a relationship, and a tail entity;

[0108] A calculation unit 502 is used to perform probability calculation on each quadruple association feature under the condition of quadruple incremental structure knowledge corresponding to the time, head entity, relationship and tail entity to obtain a quadruple conditional probability of the quadruple association feature;

[0109] A determination unit 503, configured to determine an entity association feature from a plurality of quadruple association features according to a plurality of quadruple establishment probabilities corresponding to the plurality of quadruple association features;

[0110] The completion unit 504 is used to complete the training data according to the entity association features to obtain completed data.

[0111] In an optional implementation of the embodiment of the present application, the calculation unit 502 is used to:

[0112] Under the condition of the quadruple incremental structure knowledge, the probability of the head entity in the quadruple association feature is calculated to obtain the head entity establishment probability of the head entity;

[0113] Under the conditions of the quadruple incremental structure knowledge and the head entity, the probability of the tail entity in the quadruple association feature is calculated to obtain the tail entity establishment probability of the tail entity;

[0114] Under the conditions of the quadruple incremental structure knowledge, the head entity and the tail entity, the probability of the relationship in the quadruple association feature is calculated to obtain the probability of the relationship being established;

[0115] The probability of the four-tuple association features is calculated according to the head entity establishment probability, the tail entity establishment probability and the relationship establishment probability to obtain the four-tuple establishment probability.

[0116] In an optional implementation of the embodiment of the present application, the determining unit 503 is configured to:

[0117] According to the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features, the quadruple association feature corresponding to the maximum establishment probability is determined as the entity association feature from the multiple quadruple association features.

[0118] In an optional implementation of the embodiment of the present application, the quadruple incremental structural knowledge includes the incremental structural knowledge of the head entity corresponding to the quadruple incremental structural knowledge, the global structural knowledge of the relationship, and the global structural knowledge of the tail entity.

[0119] In an optional implementation of the embodiment of the present application, the incremental structural knowledge of the head entity includes the incremental attribute knowledge and incremental relationship knowledge of the head entity; the global structural knowledge of the tail entity includes the global attribute knowledge and global relationship knowledge of the tail entity.

[0120] In an optional implementation of the embodiment of the present application, the device further includes: a compression unit;

[0121] The acquisition unit 501 is further used to acquire, for each entity to be trained in the completed data, incremental attribute knowledge and known attribute knowledge of the entity to be trained; the incremental attribute knowledge means that the incremental attribute of the entity to be trained has an incremental attribute value, and the known attribute knowledge means that the known attribute of the entity to be trained has a known attribute value;

[0122] The calculation unit 502 is further used to calculate the probability that the incremental attribute of the entity to be trained has the incremental attribute value under the condition that the known attribute of the entity to be trained has the known attribute value, so as to obtain the attribute value establishment probability that the incremental attribute of the entity to be trained has the incremental attribute value;

[0123] The calculation unit 502 is further configured to, if the probability of the attribute value being true is greater than or equal to the preset probability of true, perform a correlation calculation based on the incremental attribute value and the known attribute value to obtain the attribute value correlation between the incremental attribute value and the known attribute value;

[0124] The compression unit is used to compress the global attribute knowledge of the entity to be trained if the attribute value correlation is greater than or equal to the preset correlation, so as to obtain the compressed attribute knowledge of the entity to be trained.

[0125] In an optional implementation of the embodiment of the present application, the calculation unit 502 is used to:

[0126] According to the attribute value conditional probability that the incremental attribute has the incremental attribute value under the condition that the known attribute has the known attribute value, and the attribute correlation between the incremental attribute and the known attribute, the probability that the incremental attribute of the training entity has the incremental attribute value is calculated to obtain the probability of the attribute value being true.

[0127] Through various implementation methods provided in this embodiment, the data completion device includes an acquisition unit, a calculation unit, a determination unit and a completion unit. Among them, the acquisition unit acquires multiple quadruple association features in the multi-association knowledge graph of the data to be trained, each quadruple association feature includes time, head entity, relationship and tail entity; the calculation unit calculates the quadruple establishment probability of the quadruple association feature for each quadruple association feature under the condition of the quadruple incremental structure knowledge corresponding to the time, head entity, relationship and tail entity; the determination unit determines the entity association feature from the multiple quadruple association features based on the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features; the completion unit completes the data to be trained based on the entity association feature to obtain the completed data. It can be seen that the device considers the dynamic changes of the entities to be trained in the training data through the multi-association knowledge graph of the training data, and mines multiple four-tuple association features formed based on time, head entity, relationship and tail entity; and under the condition of the four-tuple incremental structure knowledge corresponding to the time, head entity, relationship and tail entity, calculates the four-tuple establishment probability of each four-tuple association feature, indicating the possibility of the establishment of the head entity, the tail entity, and the dynamic relationship between the head entity and the tail entity in each four-tuple association feature; accurately selects entity association features from multiple four-tuple association features, so as to accurately complete the data to be trained through the entity association features to obtain the completed data, so that the completed data is more accurate, thereby improving the accuracy of the pre-trained model built based on the completed data.

[0128] The present application also provides a computer device, which includes a processor and a memory:

[0129] The memory is used to store a computer program and transmit the computer program to the processor;

[0130] The processor is configured to execute the method described in the above embodiment according to the instructions in the computer program.

[0131] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, it is used to implement the method described in the above embodiment.

[0132] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0133] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0134] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "include", "comprises" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprises a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0135] The above is only a preferred embodiment of the present application and does not constitute any formal limitation to the present application. Although the present application has been disclosed as above with a preferred embodiment, it is not intended to limit the present application. Any technician familiar with the art can make many possible changes and modifications to the technical solution of the present application using the methods and technical contents disclosed above without departing from the scope of the technical solution of the present application, or modify it into an equivalent embodiment of equivalent changes. Therefore, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present application without departing from the content of the technical solution of the present application still fall within the scope of protection of the technical solution of the present application.

Claims

1. A data completion method, characterized in that: include: Obtain multiple quadruple association features in the multi-association knowledge graph of the data to be trained; Each quadruple association feature includes time, head entity, relation, and tail entity; For each of the four-tuple association features, probability calculation is performed on the four-tuple association features under the condition of the four-tuple incremental structure knowledge corresponding to the time, the head entity, the relationship, and the tail entity to obtain the four-tuple establishment probability of the four-tuple association features; Determining an entity association feature from the multiple quadruple association features according to the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features; The data to be trained is completed according to the entity association features to obtain completed data.

2. The method according to claim 1, characterized in that The performing probability calculation on the four-tuple association feature under the condition of the four-tuple incremental structure knowledge corresponding to the time, the head entity, the relationship and the tail entity to obtain the four-tuple establishment probability of the four-tuple association feature includes: Under the condition of the quadruple incremental structural knowledge, a probability calculation is performed on the head entity in the quadruple association feature to obtain the head entity establishment probability of the head entity; Under the conditions of the quadruple incremental structure knowledge and the head entity, a probability calculation is performed on the tail entity in the quadruple association feature to obtain a tail entity establishment probability of the tail entity; Under the conditions of the quadruple incremental structure knowledge, the head entity and the tail entity, probability calculation is performed on the relationship in the quadruple association feature to obtain the relationship establishment probability of the relationship; The probability of the four-tuple association feature is calculated according to the head entity establishment probability, the tail entity establishment probability and the relationship establishment probability to obtain the four-tuple establishment probability.

3. The method according to claim 1, characterized in that The determining the entity association feature from the multiple quadruple association features according to the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features includes: According to the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features, the quadruple association feature corresponding to the maximum establishment probability is determined from the multiple quadruple association features as the entity association feature.

4. The method according to any one of claims 1 to 3, characterized in that: The quadruple incremental structural knowledge includes the incremental structural knowledge of the head entity, the global structural knowledge of the relationship, and the global structural knowledge of the tail entity corresponding to the quadruple incremental structural knowledge.

5. The method according to claim 4, characterized in that The incremental structural knowledge of the head entity includes incremental attribute knowledge and incremental relationship knowledge of the head entity; the global structural knowledge of the tail entity includes global attribute knowledge and global relationship knowledge of the tail entity.

6. The method according to claim 1, characterized in that The method further comprises: For each entity to be trained in the completed data, incremental attribute knowledge and known attribute knowledge of the entity to be trained are obtained; the incremental attribute knowledge indicates that the incremental attribute of the entity to be trained has an incremental attribute value, and the known attribute knowledge indicates that the known attribute of the entity to be trained has a known attribute value; Under the condition that the known attribute of the entity to be trained has the known attribute value, a probability calculation is performed on the incremental attribute of the entity to be trained having the incremental attribute value to obtain the attribute value establishment probability that the incremental attribute of the entity to be trained has the incremental attribute value; If the probability of the attribute value being true is greater than or equal to the preset probability of being true, performing a correlation calculation based on the incremental attribute value and the known attribute value to obtain the attribute value correlation between the incremental attribute value and the known attribute value; If the attribute value correlation is greater than or equal to a preset correlation, attribute compression is performed on the global attribute knowledge of the entity to be trained to obtain compressed attribute knowledge of the entity to be trained.

7. The method according to claim 6, characterized in that Under the condition that the known attribute of the entity to be trained has the known attribute value, performing probability calculation on the incremental attribute of the entity to be trained having the incremental attribute value to obtain the attribute value establishment probability that the incremental attribute of the entity to be trained has the incremental attribute value includes: Based on the attribute value conditional probability that the incremental attribute has the incremental attribute value under the condition that the known attribute has the known attribute value, and the attribute correlation between the incremental attribute and the known attribute, the probability that the incremental attribute of the entity to be trained has the incremental attribute value is calculated to obtain the probability that the attribute value is established.

8. A data completion device, characterized in that: include: Acquisition unit, calculation unit, determination unit and completion unit; The acquisition unit is used to acquire multiple quadruple association features in the multi-association knowledge graph of the training data; Each quadruple association feature includes time, head entity, relation, and tail entity; The calculation unit is used to perform probability calculation on each of the four-tuple association features under the condition of the four-tuple incremental structure knowledge corresponding to the time, the head entity, the relationship and the tail entity to obtain the four-tuple conditional probability of the four-tuple association feature; The determining unit is used to determine the entity association feature from the multiple quadruple association features according to the multiple quadruple establishment probabilities corresponding to the multiple quadruple association features; The completion unit is used to complete the data to be trained according to the entity association features to obtain completed data.

9. A computer device, characterized in that: The computer device comprises a processor and a memory: The memory is used to store a computer program and transmit the computer program to the processor; The processor is configured to execute the method according to any one of claims 1 to 7 according to the instructions in the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, it is used to implement the method according to any one of claims 1 to 7.

Images