Causal discovery using hyper-relational knowledge graph link prediction

Abstract

Causal discovery using hyper-relational knowledge graph link prediction is provided. Knowledge graph link prediction is used to perform causal discovery. In accordance with the mapping, information from the causal network is transformed into a causal knowledge graph comprising a plurality of causal links, where each causal link comprises a cause entity, a causal relationship and a result entity, and a potential medium. The causal knowledge graph is converted into an embedding, wherein the embedding comprises a potential vector space representation of the causal knowledge graph. The embedding is trained using a subset of causal links of the causal knowledge graph. Additional causal links are embedded that are used for causal discovery to predict a causal knowledge graph.

Description

Technical Field

[0001] The various aspects of this disclosure generally involve causal link prediction using hyperrelationship knowledge graphs. Background Technology

[0002] A knowledge graph is a graphical data model that captures the semantic relationships between entities, where entities can be events, objects, or concepts. Knowledge graphs can be used to capture causality in terms of cause and effect. This entity-based representation model achieves a broader search space by linking causal entities to related result entities or concepts in the knowledge graph. Summary of the Invention

[0003] In one or more illustrative examples, a method for causal discovery using knowledge graph link prediction includes: transforming information from a causal network into a causal knowledge graph comprising a plurality of causal links, wherein each causal link comprises a causal entity, a causal relation, and an outcome entity, and at least a subset of the causal links comprises causal hyperrelation links, the causal hyperrelation links further connecting the causal entity and the outcome entity to an association qualifier entity as a mediator in the cascaded causal connections between the causal entity and the outcome entity; transforming the causal knowledge graph into an embedding comprising a latent vector space representation of the causal knowledge graph; training the embedding using the causal links and causal hyperrelation links of the causal knowledge graph; and using the embedding for causal discovery to predict additional causal links in the causal knowledge graph.

[0004] In one or more illustrative examples, transforming a causal knowledge graph into an embedding involves using a StarE-based graph neural network approach, where a variable number of qualifier pairs can be configured to be associated with each causal link.

[0005] In one or more illustrative examples, transformations are performed in accordance with a causal ontology that defines concepts to structure a causal knowledge graph.

[0006] In one or more illustrative examples, the transformation also includes mapping nodes in a causal network to causal entities in a causal knowledge graph; and mapping edges in a causal network to causal links in a causal knowledge graph.

[0007] In one or more illustrative examples, the qualifier entity includes an intermediary entity concatenated between the cause entity and the result entity, such that the cause entity causes the result entity and / or the cause entity interprets the result entity.

[0008] In one or more illustrative examples, the causal hyperrelation link includes a qualifier relation, which includes one or more of the following: a hasMediator qualifier relation indicating the presence or absence of a medium, or a hasMediatorType qualifier relation indicating the type of medium.

[0009] In one or more illustrative examples, causal discovery includes causal interpretation to predict the type of causal entity given an outcome entity.

[0010] In one or more illustrative examples, causal discovery includes causal prediction to predict the type of an outcome entity given a causal entity.

[0011] In one or more illustrative examples, a system for causal discovery using knowledge graph link prediction includes one or more hardware computing devices configured to: transform information from a causal network into a causal knowledge graph comprising a plurality of causal links, wherein each causal link includes a causal entity, a causal relation, and an outcome entity, and at least a subset of the causal links includes causal hyperrelation links, which further connect the causal entities and outcome entities to association qualifier entities as mediators in the cascaded causal connections between the causal entities and outcome entities; transform the causal knowledge graph into embeddings comprising a latent vector space representation of the causal knowledge graph; train the embeddings using the causal links and causal hyperrelation links of the causal knowledge graph; and use the embeddings for causal discovery to predict additional causal links in the causal knowledge graph.

[0012] In one or more illustrative examples, transforming a causal knowledge graph into an embedding involves using a StarE-based graph neural network approach, where a variable number of qualifier pairs can be configured to be associated with each causal link.

[0013] In one or more illustrative examples, the transformation of information is performed in accordance with a causal ontology, where the causal ontology defines concepts to structure a causal knowledge graph.

[0014] In one or more illustrative examples, the transformation information also includes mapping nodes in a causal network to causal entities in a causal knowledge graph; and mapping edges in a causal network to causal links in a causal knowledge graph.

[0015] In one or more illustrative examples, for a causal explanation, the qualifier entity includes a mediator entity concatenated between the result entity and the cause entity, which explains the cause of the result entity.

[0016] In one or more illustrative examples, for causal prediction, the qualifier entity includes an intermediary entity concatenated between the cause entity and the result entity, which causes the result entity.

[0017] In one or more illustrative examples, the causal hyperrelationship includes a qualifier relation, which includes one or more of the following: a hasMediator qualifier relation indicating the presence or absence of a medium and a hasMediatorType qualifier relation indicating the type of medium.

[0018] In one or more illustrative examples, causal discovery includes causal interpretation to predict the type of causal entity given an outcome entity.

[0019] In one or more illustrative examples, causal discovery includes causal prediction to predict the type of an outcome entity given a causal entity.

[0020] In one or more illustrative examples, a non-transitory computer-readable medium includes instructions for predicting causal discovery using knowledge graph links. When executed by one or more computing devices, the instructions cause the one or more computing devices to perform the following operations: transforming information from a causal network into a causal knowledge graph comprising a plurality of causal links, wherein each causal link comprises a causal entity, a causal relation, and an outcome entity, and at least a subset of the causal links comprises causal hyperrelation links, which further connect the causal entities and outcome entities to association qualifier entities as mediators in the cascaded causal connections between the causal entities and outcome entities; transforming the causal knowledge graph into embeddings comprising a latent vector space representation of the causal knowledge graph; training the embeddings using the causal links and causal hyperrelation links of the causal knowledge graph; and using the embeddings for causal discovery to predict additional causal links in the causal knowledge graph.

[0021] In one or more illustrative examples, transforming a causal knowledge graph into an embedding involves using a StarE-based graph neural network approach, where a variable number of qualifier pairs can be configured to be associated with each causal link.

[0022] In one or more illustrative examples, the transformation of information is performed in accordance with a causal ontology, where the causal ontology defines concepts to structure a causal knowledge graph.

[0023] In one or more illustrative examples, the causal hyperrelationship link includes a qualifier relation, which includes one or more of the following: a hasMediator qualifier relation indicating the presence or absence of a medium and / or a hasMediatorType qualifier relation indicating the type of medium.

[0024] In one or more illustrative examples, the qualifier entity includes an intermediary entity concatenated between the cause entity and the result entity, such that the cause entity causes the result entity and / or the cause entity interprets the result entity. Attached Figure Description

[0025] Figure 1A The diagram illustrates an example of a cascaded causal connection, where A causes B, and ultimately B causes C; Figure 1B The diagram illustrates a chain of causal links encoded as knowledge graph links using the Resource Description Format (RDF) format. Figure 1C The diagram illustrates a causal link as a hyperrelation link, where the media entity is represented as a hyperrelation with a hyperrelation predicate hasMediator using the RDF-Star format; Figure 2A The flowchart illustrates the four stages of the disclosed method for finding missing causal links in incomplete causal networks. Figure 2B The illustration shows an example of causal link prediction that includes qualifier relations; Figure 2C The illustration shows an example of causal link prediction without qualifier relations; Figure 3 The diagram illustrates the instance materialization relationship within the context of CausalKG; Figure 4 The illustration shows an example StarE encoder; Figure 5 The diagram illustrates an example StarE architecture used for linking prediction models; Figure 6A The illustration shows an example CausalKG structure including a subgraph C with causal links that only have causal relationships. Figure 6B The illustration shows an example CausalKG structure including a subgraph CT, which has causal links with causal relationships and information about entity types; Figure 6C The diagram illustrates an example CausalKG structure that includes a subgraph CTP, which has causal relationships, entity type relationships, and information about the objects involved in causal events. Figure 7A The diagram illustrates the time... , and A snapshot of the collision event shown in the video; Figure 7B The diagram shows Figure 7A A snapshot of a causal event graph; Figure 7CThe causal and mediating representations of causal events in two different Causal KGs are shown; Figure 8A The illustration shows the results of an example experiment using subgraph C to perform a causal interpretation; Figure 8B The illustration shows example experimental results for performing causal interpretation using subgraph CT; Figure 8C The illustration shows example experimental results of performing causal interpretation using subgraph CTP; Figure 9A The illustration shows example experimental results for performing causal prediction using subgraph C; Figure 9B The illustration shows example experimental results for performing causal prediction using subgraph CT; Figure 9C The illustration shows example experimental results for performing causal prediction using subgraph CTP; Figure 10 The diagram illustrates an example process for causal discovery using hyperrelations; Figure 11 A schematic diagram depicts a control system configured to control a robotic assistant based on causal discovery; and Figure 12 The illustration shows an example manufacturing system used for anomaly detection and / or generation of synthetic anomaly data. Detailed Implementation

[0026] Detailed embodiments of the invention are disclosed herein as needed; however, it should be understood that the disclosed embodiments are merely illustrative of the invention, which can be embodied in various forms and alternative forms. The figures are not necessarily to scale; some features may be enlarged or reduced to show details of specific components. Therefore, the specific structural and functional details disclosed herein should not be construed as limiting, but merely as a representative basis for teaching those skilled in the art to employ the invention in different ways.

[0027] Traditionally, causal relationships are represented using Causal Bayesian Networks (CBNs), where nodes in a CBN represent events, and edges or links represent causal associations between two events. Having a complete network with all causal associations is important for many downstream applications. However, in practice, such causal networks are often incomplete, with missing causal links. Recent methods have successfully addressed this problem by encoding the causal network in a triple-based knowledge graph (such as a Resource Description Format (RDF)) and then using knowledge graph link prediction techniques to find missing causal links. While this approach works well for simple binary causal links, it does not support more complex links, such as mediated causal links.

[0028] Figures 1A-1C An example of the structure of a causal connection is illustrated. Figure 1A The diagram illustrates an example of a cascaded causal connection, where A causes B and ultimately B causes C. Figure 1B The diagram illustrates a chain of causal links encoded as knowledge graph links using RDF format. Figure 1C The diagram illustrates causal links as hyper-relational links, where mediating entities are represented as hyper-relational predicates with the term hasMediator (e.g., total causal result, natural direct result, natural indirect result). Causal links are encoded as knowledge graph links using RDF-Star format.

[0029] refer to Figure 1A A simple binary causal link can be defined such that A causes B. In this case, A can be called the cause, and B can be called the effect. This causal link can also be linked together, where A causes B, and then B causes C. In this more complex case, there is a causal link between A and C, linked by B. As a medium Nodes A and C can be referred to as cause and effect, respectively, while node B can be referred to as medium.

[0030] While existing methods using knowledge graph (KG) link prediction can predict direct binary causal links, such as A causing B, they cannot predict more complex mediated causal links, such as A being mediated by B causing C.

[0031] Figure 2A The diagram illustrates a flowchart of the four stages of the disclosed method for finding missing causal links in incomplete causal networks. This method, which may be called HyperCausalLP, supports and utilizes mediated links. This is achieved by using a hyper-relational knowledge graph to represent complex causal relationships. The four main stages are: causal network construction 202, causal knowledge graph creation 204, embedding generation 206, and causal discovery 208.

[0032] The causal network construction 202 may include finding known causal relationships and encoding them into the causal network 210. Figures 1A-1C The example illustrates how mediated causal links 308 can be encoded as superrelations. RDF-Star1 can be used to encode these causal links 308. The causal network construction 202 can be performed using observed data 212 and / or domain knowledge 214. Causal knowledge graph creation 204 may include transforming the causal network 210 into a CausalKG 216 conforming to a causal ontology 218.

[0033] Embedding generation 206 may include KG embedding models 220A and 220B that learn CausalKG 216. The super-relation KG is then used to train a KG embedding (KGE) model. This can be achieved using the StarE algorithm, which employs a neural network-based message-passing framework. Finally, the KGE model is used to predict new causal links 308. In this method, embeddings include CausalKG-Base without mediating relations and CausalKG-M, a super-relation graph-based embedding with mediating relations as super-relations.

[0034] Causal discovery 208 may include embedding a knowledge graph for the causal discovery 208 task. An example of such a task is predicting new causal links 308 in CausalKG 216. More specifically, two causal tasks are performed: (1) causal explanation, in which an outcome event is given and its cause is predicted, and (2) causal prediction, in which a cause event is given and its outcome is predicted.

[0035] This approach for finding missing causal links with a medium can be evaluated using the CLEVRER-Humans benchmark dataset for causal inference. The dataset provides a set of causal networks representing collision events in simulated videos. Based on this set of causal networks, a hyperrelational causal KG is generated, a KG embedding model is trained, and two link prediction operations are performed, one for causal explanation and the other for causal prediction. The evaluation results show that using a hyperrelational KG embedding model to encode causal link predictions with a medium outperforms baseline methods using standard ternary basis KG embedding models without a medium.

[0036] Causal link prediction (308) can be formalized as a KG link prediction problem. The following definitions define the main concepts, including causal relation (304), causal link (308), causal entity (302), qualifier entity (312), causal super-relation link (316), and causal knowledge graph.

[0037] Causal Knowledge Graph It is a super-relation KG, which includes causal knowledge in the form of causal relation 304 and causal entity 302. Let... ,in: : Represents a set of nodes of an entity; A set of labels representing relationships; : Represents a set of edges that link pairs of entities. Each link is a triple. ,in It is the head entity. It's a relationship. It is the tail section; : Represents a set of nodes for causal entity 302; : A set of tags indicating causality (304 error); Indicate the power set; and : A set of edges, representing the causal super-relation link 316 connecting the pairs of causal entities 302.

[0038] Causal entity 302, This is the entity that serves as the head or tail of causal link 308. There are two types of causal entities 302: Cause Entity ( )and Result Entity ( ), making Cause Entity cause Result Entity However, in the case of hyperrelationship links, causal entity 302 can also be qualifier entity 312 ( ).

[0039] Causal relationship 304 This represents a causal relationship between entities. There are four types of causal relationships: 304 From Cause Entity arrive Result Entity 304; From Result Entity arrive Cause Entity The causal relationship 304; that is The reverse; From Cause Entity arrive Result Entity Type of causal relationship 304; and From Result Entity arrive Cause Entity Type of causal relationship 304.

[0040] Causal link 308 , is an edge in causalKG 216 connecting a pair of causal entities 302 with causal relationship 304. Causal link 308 is a triple. ,in It is the head causal entity 302. It is a causal relationship (304), and It is the causal entity at the tail, 302.

[0041] Qualifier for 310, , is a super relation in causalKG 216, which connects causal link 308 to its super relation entity pair. It is a set of qualifiers for 310 ( ), with qualifier relationship , and qualifier entity 312, .

[0042] Qualifier entity 312, , is causal entity 302, which is part of qualifier pair 310. In a given cascaded causal connection, qualifier entity 312 (i.e., the medium) is sequentially connected. Cause Entity and Result Entity The entity between them is also referred to as the medium. In this disclosure, qualifier entity 312 refers to the medium in a causal connection. In the context of this disclosure, word qualifier entity 312 and medium are used interchangeably.

[0043] Qualifier relation 314, This represents the association between causal link 308 and qualifier entity 312 (e.g., media entity). There are two types of qualifier relationships 314: It is the qualifier relation from causal link 308 to the mediating entity 314; and It is the qualifier relation from causal link 308 to media entity type 314.

[0044] Causal hyperrelation link 316 , is an edge in causalKG 216 that connects a pair of causal entities 302 to causal relation 304 and their associated medium (qualifier entity 312). Each causal hyperrelation link 316 is a triple. ,in It is the head causal entity 302. It's a causal relationship (304). It is the causal entity 302 at the tail. It is a set of qualifiers for 310 ( ), with qualifier relation 314, , and qualifier entity 312, .

[0045] Causal relation 304 extraction is a task of finding new causal links 308 within Causal KG 216. Given Causal KG 216, G, this task can be achieved using knowledge graph link prediction. There are two types of causal relation 304 extraction—causal prediction and causal explanation: Causal prediction: Given a causal entity ( ), relation( ) and qualifier pair 310 ( ), find the type of the associated result entity ( ), making Established.

[0046] Causal explanation: Given a result entity ( ), relation( ) and qualifier pair 310 ( ), find the type of the associated cause entity ( ), making Established.

[0047] Return to reference Figure 2A HyperCausalLP is structured into four main stages: causal network construction 202, in which known causal relationships 304 are identified and encoded into a causal network 210; causal KG creation 204, in which the causal network 210 is transformed into a CausalKG 216, conforming to a super-relational causal ontology 218 incorporating qualifier pairs 310; embedding generation 206, during which super-relational KG embeddings 224 are learned for the CausalKG 216; and causal discovery 208, in which new causal links 308 are predicted in the CausalKG 216.

[0048] A causal network (210) is a graphical model known as a causal Bayesian network, which is structured as a directed acyclic graph. In this model, nodes symbolize events, and edges represent causal links between these events (308). It is labeled as... The network includes nodes and edge The direction of each edge in the network indicates the direction of causality. Given a three-node causal network 210, the causal links 308 can have three different orientation structures: concatenation, bifurcation, and collision. A concatenation structure is one in which causal associations are traversed sequentially, such as the first event causing the second event, and the second event causing the third event. In a bifurcation structure, the first event causes both the second and third events. In a collision structure, two independent events together cause the third event. However, this disclosure focuses on... Figure 1A The diagram shows a chain structure. The first node is considered the cause entity, the second node is the mediator entity, and the third node is the result entity.

[0049] The process of transforming data from causal network 210 into a super-relational causal knowledge graph (CausalKG 216) involves several direct transformations: The nodes in causal network 210 become causal entities 302 in CausalKG 216. The mediator nodes in causal network 210 become mediator entities in CausalKG 216, which are represented as qualifier entities 312.

[0050] The edges in causal network 210 become causal links 308 in CausalKG 216, in the form of .

[0051] CausalKG 216 also incorporates other causal relations 304 and qualifier relations 314, such as: , , , and CausalKG 216 includes information from causal network 210 and conforms to causal ontology 218. Causal ontology 218 can be rooted in concepts from causal artificial intelligence (AI), such as causal Bayesian networks and do-calculus. Causal ontology 218 can be used to define the semantics and structure of nodes in causal relations 304 and CBN. The ontology can define the main concepts used to structure CausalKG 216, including causal entities 302, causal relations 304, and mediators.

[0052] CausalKG 216 is used for causal link 308 prediction using KG link prediction. There are two causal link 308 prediction tasks: causal explanation and causal prediction. The goal of causal explanation is to predict the type of the causal entity linked to the resulting entity. The goal of causal prediction is to predict the type of the resulting entity linked to the causal entity. The goal of these two tasks is not to predict specific instances of causal entities (in the case of causal explanation) or resulting entities (in the case of causal prediction), but rather the types of these corresponding entities.

[0053] The causal entity (in the case of causal explanation) and the resultant entity (in the case of causal prediction) are not directly linked to the causal entity type and the resultant entity, respectively. They are separated by two hops: for causal explanation, ; and for causal prediction, The embedding model makes predictions about directly linked entities.

[0054] Figure 3Example 300 illustrates the materialization relation in the context of CausalKG 216. Example 300 shows the materialized causal relation 304, causesType, and causedByType. causedByType is the materialization relation from the result entity instance to the cause entity type. causesType is the materialization relation from the cause entity instance to the result entity type. Example 300 also illustrates two qualifier relations 314 associated with the causal relation: hasMediator and hasMediatorType. Qualifier relation 314 is also associated with the causedBy relation, which is the inverse of the causal relation. Such example 300 can be used for causal prediction: This is used to add a link connecting the cause entity to the result entity type. In another example, it is used for causal explanation: This adds links connecting the resulting entity to the causal entity type. Along with all the above knowledge, CausalKG 216 can also integrate additional domain knowledge 214 associated with entities not explicitly mentioned in the causal network 210.

[0055] Return to reference Figure 2A CausalKG 216 can be transformed into a low-dimensional continuous latent vector space representation, which can be called KGE 224. KGE 224 can be used for downstream tasks such as link prediction, entity classification, and triad classification. The HyperCausalLP method can use the KG embedding algorithm to generate embeddings that can be used for causal link 308 prediction.

[0056] This method learns two types of KGE 224 embeddings for CausalKG 216: 1) CausalKGE-Base224B embeddings without a mediator (no superrelation), and 2) CausalKGE-M224A embeddings with a mediator as a superrelation (represented using qualifier pair 310). The CausalKGE-Base 224B embeddings can be trained using causal links 308, thus ignoring the mediator associated with each link. Conversely, the CausalKGE-M 224A embeddings can be trained using causal links 308 with a mediator. Both CausalKGE-Base 224B and CausalKGE-M 224A embeddings can be evaluated on a causal link 308 prediction task using KG link prediction techniques. In the example, the CausalKG 216 embeddings for CausalKGE-Base 224B can be generated using KG embedding algorithms available in the Ampligraph library2.

[0057] Figure 2BAn example of a causal link 308 prediction including qualifier relation 314 is illustrated. This can be performed, for example, using a CausalKGE-M 224A embedding with a medium as a superrelation. Figure 2C The illustration shows an example of a causal link 308 prediction without qualifier relation 314. This can be performed, for example, using CausalKGE-Base 224B embeddings.

[0058] CausalKGE-M 224A can be generated as a super-relation KGE 224 model based on a graph neural network, such as StarE. StarE is a graph neural network-based approach. StarE allows a variable number of qualifier pairs 310 to be associated with causal links 308. StarE combines the causal relationship 304 embedding with a fixed-length vector representing the associated qualifier pairs 310. StarE integrates qualifier pairs with causal links 308 into the message passing process. The StarE model consists of two parts— Figure 4 The StarE encoder shown and as Figure 5 The diagram shows the StarE architecture, which includes a transformer-based decoder. The StarE encoder and the transformer-based decoder can be trained jointly.

[0059] Figure 4 The illustration shows an example StarE encoder 400. The StarE encoder 400 encodes the superrelations of causal relationships. Superrelation qualifier pairs (or mediator pairs) are encoded using composition functions. They are passed together and weighted. Perform the transformation. The transformed vector and... and Merge. The final node, i.e., the cause entity, combines messages from all super-relationships. As defined in StarE: It is a composite function of the nodes and their respective relationships. These are direction-specific shared parameters for outgoing, incoming, and self-looping relationships, and It is to include the main relationship ( ) indicates its qualifier ( The representation of a function is a combination of .

[0060] The StarE encoder 400 can be used in KGE 224 models or relational inference networks. In the example, (Reasons) and include , , , The interaction of media entities, where these relationships are transformed through various transformations ( ) and aggregation ( The model calculates a weighted sum or transformation of these mediation relationships and outputs an encoded representation of the causal entity. This encoding indicates how it captures the result entity ( (and its media). As shown, Indicates a set of reasons, Represents a set of media, and It represents a set of media. The function refers to the transformation of the input vector or embedding, and Nodes represent transformation functions or mappings, such as neural network activation or parameterization functions that encode relationships between entities. It can refer to a change specific to a medium, while This can refer to the transformation applied after the summation step of aggregating intermediate information. The summation node (composed of...) (Identification) Aggregates inputs from various media, which can combine information from different media paths to integrate different relational signals before further processing. and The weights in a neural network correspond to learnable parameters that adjust the influence of aggregated information. It can be applied to the aggregation of media, and Possible causes - entity path. This represents a gating mechanism or activation function that modulates the encoded information before it is passed to the final summation and prediction steps.

[0061] Figure 5 The illustration shows an example StarE architecture 500 used for link prediction models. As shown, the StarE architecture 500 updates the N and R matrices and then encodes them using these matrices before passing the relations in a given query through the transformer layer, pooling layer, and fully connected layer. The fixed-dimensional output is then compared with N, and the result is passed through a sigmoid function to produce a probability distribution over the entities.

[0062] The StarE method initializes two embedding matrices, R (relationships) and E (entities). StarE iteratively updates the embeddings by message-passing across edges in the training set. For the link prediction task, the query is first linearized, and then the updated embeddings are used to encode the relations and entities. The data is then passed through a transformer. The output of the transformer is averaged to obtain a fixed-dimensional vector representation of the query. The vector is passed through a fully connected layer, multiplied with the entities, and passed through a sigmoid function to obtain a probability distribution over all entities. The top n candidate entities for the link prediction query are obtained.

[0063] The disclosed method, HyperCausalLP, formalizes the problem of predicting causal links 308 into a KG link prediction task. Trained CausalKG 216 embedding models, namely CausalKGE-Base 224B and CausalKGEM, are used to predict missing causal links 308 between causal entities 302 in a KG. More specifically, HyperCausalLP is used for the tasks of causal explanation and causal prediction. Causal explanation aims to predict the cause of an outcome, and causal prediction aims to predict the outcome of a cause. For a given causal link 308, the causal explanation prediction form... The link, and the causal prediction prediction form The link.

[0064] For a given dataset, HyperCausalLP can be used to create CausalKG 216, generating and learning KGE 224, using the mediators associated with causal entities 302, causal relationships 304, and causal links 308 between entities. The generated KGE 224 can then be used for causal link 308 prediction in the form of causal explanations and causal predictions.

[0065] The CLEVRER-Humans dataset (a benchmark dataset for causal inference) can be used to evaluate HyperCausalLP. More specifically, the KG link prediction task can be used to evaluate the HyperCausalLP-based hypergraph causal link 308 prediction method for 1) causal interpretation, given the outcome entity prediction form. The causal link 308 refers to the type of the causal entity, and 2) causal prediction, given the form of the causal entity prediction. The type of the result entity of the causal triple (see Figure 3 The above evaluation can be demonstrated using the CLEVRER-Humans benchmark dataset for causal inference. This section describes in detail the CLEVRER-Humans dataset, data preprocessing steps, creation of CausalKG 216 from the dataset, experimental setup, evaluation metrics, and evaluation of different CausalKG 216 variants.

[0066] The initial steps in generating the CLEVRER-Humans CausalKG 216 involve preprocessing causal event graphs (CEGs). CEGs act as proxies for causal networks 210, and their preprocessing is crucial to ensuring they conform to the definition of causal networks 210. In causal networks 210, edges represent causal links 308 between nodes. The first step in this process is to remove edges with a score of 1, indicating no causal responsibility between the two nodes. Next, to maintain the structure of the directed acyclic graph, edges that create cycles in the CEG are removed. Finally, a CEG is excluded if it has no remaining causal links 308 or if the depth from the root node to a leaf node is less than 2. After preprocessing, the CLEVRER-Humans dataset has 764 CEGs remaining.

[0067] Regarding event extraction, the CLEVRER-Humans dataset features 27 distinct events, such as collision, entry, exit, stop, and proceed. These events can be categorized into two types: binary events and single events. Binary events involve two participating objects and include actions such as collision, impact, hit, bounce, and side-bump. Single events involve only one object and include actions such as entry, exit, and stop. Information about the event type and participating objects can be extracted from the node descriptions in the CEG by parsing the CEG JavaScript object annotation (JSON) file provided by the dataset. To capture the root forms of event labels (e.g., collision, hit, push) rather than their verb forms (e.g., collided, hit, pushed), the Berkeley Neural Semantic Analyzer and the Natural Language Toolkit (NLTK) lemmatizer can be used (in the example). Nodes describing multiple events, such as "the red ball collides with the blue ball and hits the yellow cylinder," can be removed from the CEG because they describe more than one event. Instead, focus can be placed on nodes describing a single event.

[0068] Regarding object and object property extraction, in addition to extracting event types, information about the participating objects and their characteristics, including color, shape, and material, can be collected. Some object characteristics in the dataset may be mislabeled, such as an object being labeled as gold instead of yellow. These mislabeling issues can be identified, and the terminology can be normalized accordingly.

[0069] CausalKG 216 can be created according to CLEVRER-Humans by encoding causal information within a CEG in RDF format, following the causal ontology 218. As described above, the disclosed method creates two different KGs: CausalKG-Base 220B and CausalKG-M 220A. CausalKG-Base 220B can be a simple KG with causal links 308, while CausalKG-M 220A is a super-relation KG composed of mediators as super-relations (qualifiers). Super-relations with mediator information between two given nodes in a CEG can be encoded using the RDF-star format as discussed herein.

[0070] KG can include causal relationships and, in some examples, details about events (such as hits, collisions, pushes, etc.), the objects involved, and their properties. CEG can serve as a graphical representation of events in the video. To represent information from CEG, three ontologies can be used: causal ontology 218, scene ontology (prefixed with "so:"), and semantic sensor network ontology (prefixed with "ssn:"). Causal ontology 218 can be used for events (as causal entities 302), causal relationships 304, and their corresponding causal media (i.e., qualifier pairs 310). Scene and sensor ontology describe additional video information, such as scenes, objects, and object properties. Each video can be described as a scene (so:Scene) using the scene ontology concept. This includes representing and connecting events within the scene (using the so:includes relation), the objects involved (using the so:hasParticipant relation), and object properties (using the ssn:hasProperty relation). In summary, CausalKG 216 from CLEVRER-Humans contains over 48K links, 5664 entities, 31 entity types, and 10 relationships.

[0071] For tasks involving causal explanation and causal prediction, CausalKGE-Base 224B and CausalKGE-M 224A embeddings can be generated and evaluated on different CLEVRER-Humans CausalKG 216 subgraph structures, such as... Figures 6A-6C As illustrated in the diagram. In the case of CausalKG-M and a given subgraph, the super relation (qualifier pair 310) can be associated with the cause and causeBy causal relations 304, as shown in the diagram. Figure 3 As shown in the diagram. When different types of information are available in CausalKG 216, various graph structures can be used to evaluate the performance of HyperCausalLP. For example, three different subgraph structures can be defined using increasing levels of expression (such as...). Figures 6A-6CAs shown in the image).

[0072] Figure 6A The illustration shows an example CausalKG 216 structure, which includes a subgraph C with causal links 308, each having only causal relationships 304. These causal relationships 304 may include, for example, causes, causedBy, causesType, and causedByType. Figure 6B The illustration shows an example CausalKG 216 structure, which includes a subgraph CT with causal links 308, causal links 308 having causal relationships 304 and information about entity types. These entity types can include, for example, rdf:type. Figure 6C The diagram illustrates an example CausalKG 216 structure, which includes a subgraph CTP containing a causal relationship 304, entity type relations, and information about the objects involved in the causal events (e.g., hasParticipant). In the case of CausalKGE-M224A, the superrelation (qualifier pair 310) is associated with the causes and causedBy causal relationships 304.

[0073] The hyperparameters of each of these graph structures can be optimized for both causal explanation and prediction tasks. The CausalKGE-Base 224B model for each graph structure can be trained on its respective optimized hyperparameters. The CausalKGE-M 224A model can be trained on the StarE hyperparameters. Then, using link prediction methods, the trained CausalKGE can be used for the causal link 308 prediction task.

[0074] The HyperCausalLP can be evaluated using KG link prediction experimental design settings. For the set of causal links given in CausalKG 216 308, By changing a set of causal links 308 using another causal entity 302 in KG tail or head To generate a set of broken links Such as using The result of replacing the header is generated and use Replace the tail result This model represents a real link. and broken links Assign scores. Scores can be sorted to rank genuine links. Filtering evaluation settings and filtering for broken links. It can be used to exclude links present in both the training and validation sets. The performance of HyperCausalLP can be evaluated using two metrics: Mean Reciprocal Rank (MRR) and Hits@K (where Hits@K = ...). MRR is the average of the lowest-ranked test links. Hits@k is the percentage of test links among the top k ranked links. Higher values ​​for these two metrics indicate better model performance.

[0075] Figure 7C The illustration shows example snapshots of CausalKG-Base and CausalKG-M representations. For example... Figure 7A As shown, the snapshot of the collision event is displayed in time from the video from CLEVRER-Humans. , and There exist three consecutive collision events: A: the red cube collides with the yellow ball; B: the yellow ball hits the blue cylinder; and C: the blue cylinder moves. A, B, and C are causal entities 302. A.Type is collision, B.Type is hit, and C.Type is movement. Figure 7B It shows Figure 7A A snapshot of the causal event graph. Figure 7C The causal and mediating (qualifier pair 310) linking representations in two different CausalKG 216 are shown.

[0076] Figures 8A-8C The accompanying illustrations show example experimental results from CausalKGE-M 224A and CausalKGE-Base224B used to perform causal interpretation. Figures 9A-9C The accompanying illustrations show example experimental results for CausalKGE-M 224A and CausalKGE-Base 224B used to perform causal prediction. More specifically, Figures 8A-8C and Figures 9A-9C The figure illustrates the MRR and Hit@K (k=1, 3, 10) of five KGE 224 models evaluated on different CausalKG 216 subgraphs. These include, for example, Figure 6A The C subgraph shown is as follows: Figure 6B The CT sub-image shown, and as... Figure 6C The CTP subgraph shown.

[0077] In these results, HyperCausalLP was evaluated on CausalKG 216, generated from the CLEVRER-Humans dataset, for causal link 308 prediction. The method was also evaluated on CausalKG-M using StarE on three distinct CausalKG 216 subgraphs C, CT, and CTP with different hyperrelationships. CausalKG-HasMediator: hasMediator as a qualifier relation 314 CausalKG-HasMediatorType: hasMediatorType as a qualifier relation 314 CausalKG-HasMediatorInstanceType: hasMediator and hasMediatorType are related as qualifiers (314).

[0078] The results (i.e., MRR, HitK) show that CausalKGE-M 224A significantly improves performance over CausalKGE-Base 224B, a baseline model with only links and no hyperrelationships (or mediating information). CausalKGE-Base 224B was evaluated using four KGE 224 models (TransE, DistMult, HolE, and ComplEx). The addition of supplementary knowledge (i.e., CT, CTP) in CausalKGE-M 224A demonstrates improved performance over the simpler C-subgraph.

[0079] The addition of more knowledge improved the KG link prediction performance for both causal explanation and causal prediction tasks. When enriched with additional knowledge for causal prediction (i.e., CTP), the CausalKGE-M 224A with hasMediator outperformed C by 13.77%. When enriched with additional knowledge for causal prediction (i.e., CTP), the CausalKGE-M 224A with hasMediatorType outperformed C by 12.55%. When enriched with additional knowledge for causal prediction (i.e., CTP), the CausalKGE-M 224A with both hasMediator and hasMediatorType outperformed C by 8.36%. When enriched with additional knowledge for causal explanation (i.e., CTP), the CausalKGE-M 224A with hasMediator outperformed C by 7.04%. When enriched with additional knowledge for causal explanation (i.e., CTP), the CausalKGE-M 224A with hasMediatorType outperforms C by 6.28%. When enriched with additional knowledge for causal explanation (i.e., CTP), the CausalKGE-M 224A with both hasMediator and hasMediatorType outperforms C by 10.13%. Combining mediator with causal link 308 provides additional knowledge, which is crucial for the causal link 308 prediction task. The super-relation hasMediator performs best compared to other super-relations hasMediatorType, as well as considering both hasMediator and hasMediatorType together. Super-relation-based KG performs better with increasing number of qualifiers. It has been successfully demonstrated that incorporating knowledge in super-relations (qualifiers) significantly improves causal link 308 prediction.

[0080] Therefore, a method for finding missing causal links 308 in an incomplete causal network 210 is disclosed. The disclosed method incorporates mediating information from the CBN as a superrelation into the KG. The KGE 224 model trained with qualifiers (mediating or superrelation) outperforms all baseline KGE 224 metrics without qualifiers. The results show that efficient fusion of causal links 308 with qualifiers (mediating or superrelation) in the KG can facilitate the completion of the incomplete causal network 210.

[0081] Figure 10An example process 1000 for causal discovery using hyperrelations is illustrated. Process 1000 can be implemented to perform causal discovery using knowledge graph link prediction, addressing a key gap in the prior art by considering media information and causal links 308. Using process 1000, it can be seen that the KGE model trained with media information outperforms the baseline KGE metric without media information. The results show that incorporating knowledge in hyperrelations significantly improves causal link 308 prediction.

[0082] At operation 1002, causal network construction 202 is performed. Causal network construction 202 may include finding known causal relationships and encoding them into a causal network 210. This causal network construction 202 may be performed using observation data 212 and / or using domain knowledge 214.

[0083] At operation 1004, causal knowledge graph creation 204 is performed. Causal knowledge graph creation 204 may include transforming causal network 210 into a CausalKG 216 conforming to causal ontology 218. CausalKG 216 may include multiple causal links, each causal link 308 including a cause entity, a causal relationship, and a result entity. In the example, information from causal network 210 may be transformed into CausalKG 216 according to a mapping. The mapping may include mapping nodes in causal network 210 to causal entities 302 in CausalKG 216, and mapping edges in causal network 210 to causal links 308 in CausalKG 216.

[0084] At operation 1006, embedding learning 206 is performed. Embedding learning 206 may include learning KG embedding models 220A and 220B of CausalKG 216 using a training set, and evaluating the training using a test set. This can be performed in two different ways. In the first approach, CausalKGE-Base 224B is generated using embeddings without mediators (without superrelations). The CausalKGE-Base 224B embeddings can be trained using causal links 308, thus ignoring the mediators associated with each link. In the second approach, CausalKGE-M 224A is generated using embeddings with mediators as superrelations (represented using qualifier pair 310). The CausalKGE-M 224A embeddings can be trained using causal links 308 with mediators. The CausalKGE-Base 224B and CausalKGE-M 224A embeddings can be evaluated on a task of predicting causal links 308 using KG link prediction techniques. In the examples, the CausalKG 216 embeddings for CausalKGE-Base 224B can be generated using KG embedding algorithms available in Ampligraph library2. CausalKGE-M 224A can be generated as a super-relation KGE 224 model based on a graph neural network, such as StarE. In many examples, the application of qualifiers from causal network 210 to 310 outperforms the baseline KGE metric that is not trained on the super-relation.

[0085] At operation 1008, causal discovery 208 is performed. Causal discovery 206 may include embedding knowledge graphs 224A, 224B for the causal discovery task. An example of such a task is predicting new causal links 308 in CausalKG 216. In some examples, causal discovery 206 includes causal explanations to make predictions given the types of result and cause entities. In some examples, causal discovery 206 includes causal predictions to make predictions given the types of cause and result entities. After operation 1008, process 1000 ends.

[0086] Figure 11 A schematic diagram depicts the interaction between the computer-controlled machine 1102 and the control system 1112. The computer-controlled machine 1112 can realize various aspects of causal discovery 208 and the use of predicted causal information. (Reference) Figure 10 And refer to Figures 1A-9CThe methods discussed herein can be performed within the context of such a computer-controlled machine 1112 and control system 1112. The computer-controlled machine 1102 includes actuators 1114 and sensors 1116. Actuators 1114 may include one or more actuators, and sensors 1116 may include one or more sensors. Sensors 1116 are configured to sense the condition of the computer-controlled machine 1102. Sensors 1116 may be configured to encode the sensed condition into a sensor signal 1118 and transmit the sensor signal 1118 to the control system 1112. Non-limiting examples of sensors 1116 include video, radar, lidar, ultrasonic, and motion sensors. In one embodiment, sensor 1116 is an optical sensor configured to sense an optical image of the environment in proximity to the computer-controlled machine 1102.

[0087] The control system 1112 is configured to receive sensor signals 1118 from the computer control machine 1102. As described below, the control system 1112 can also be configured to calculate actuator control commands 1120 based on the sensor signals 1118 and transmit the actuator control commands 1120 to the actuator 1114 of the computer control machine 1102.

[0088] like Figure 11 As shown, the control system 1112 includes a receiving unit 1112. The receiving unit 1112 can be configured to receive sensor signals 1118 from sensor 1116 and transform the sensor signals 1118 into input signals X. In an alternative embodiment, the sensor signals 1118 are received directly as input signals X without the need for the receiving unit 1112. Each input signal x can be a portion of each sensor signal 1118. The receiving unit 1112 can be configured to process each sensor signal 1118 to generate each input signal x. The input signals x can include data corresponding to the image recorded by sensor 1116.

[0089] The control system 1112 includes a machine learning (ML) process 1124. The ML process 1124 can be configured to learn, classify, infer, generate, etc., using one or more models, such as those described in detail above. In this example, the ML process 1124 is configured to determine an output signal Y based on an input signal X. Each output signal Y includes information that assigns one or more labels to each input signal X. The ML process 1124 can transmit the output signal Y to a conversion unit 1128. The conversion unit 1128 is configured to convert the output signal Y into an actuator control command 1120. The control system 1112 is configured to transmit the actuator control command 1120 to an actuator 1114, which is configured to actuate the computer-controlled machine 1102 in response to the actuator control command 1120. In another embodiment, the actuator 1114 is configured to actuate the computer-controlled machine 1102 directly based on the output signal Y.

[0090] When actuator 1114 receives actuator control command 1120, actuator 1114 is configured to perform an action corresponding to the associated actuator control command 1120. Actuator 1114 may include control logic configured to transform actuator control command 1120 into a second actuator control command 1120 used to control actuator 1114. In one or more embodiments, actuator control command 1120 may be used to control a display, instead of actuator 1114 or other devices.

[0091] In another embodiment, instead of the computer-controlled machine 1102 including sensor 1116 or otherwise, the control system 1112 includes sensor 1116. Instead of the computer-controlled machine 1102 including actuator 1114 or otherwise, the control system 1112 may also include actuator 1114.

[0092] like Figure 11 As shown, the control system 1112 also includes a processor 1130 and a memory 1132. The processor 1130 may include one or more processors. The memory 1132 may include one or more memory devices. The causal hyperrelationship links determined by one or more embodiments can be implemented by the control system 1112, which includes a non-volatile storage device 1126, the processor 1130, and the memory 1132.

[0093] Non-volatile storage device 1126 may include one or more persistent data storage devices, such as hard disk drives, optical disk drives, magnetic tape drives, non-volatile solid-state devices, cloud storage devices, or any other device capable of persistently storing information. Processor 1130 may include one or more devices selected from a high-performance computing (HPC) system, including high-performance cores, microprocessors, microcontrollers, digital signal processors, microcomputers, central processing units, field-programmable gate arrays, programmable logic devices, state machines, logic circuits, analog circuits, digital circuits, or any other device that manipulates (analog or digital) signals based on computer-executable instructions residing in memory 1132. Memory 1132 may include a single memory device or multiple memory devices, including but not limited to random access memory (RAM), volatile memory, non-volatile memory, static random access memory (SRAM), dynamic random access memory (DRAM), flash memory, cache memory, or any other device capable of storing information.

[0094] Processor 1130 may be configured to read memory 1132 and execute computer-executable instructions residing in non-volatile storage device 1126 and embodying one or more ML algorithms and / or method techniques of one or more embodiments. Non-volatile storage device 1126 may include one or more operating systems and applications. Non-volatile storage device 1126 may store compiled and / or interpreted computer programs created using a wide variety of programming languages ​​and / or techniques, including but not limited to, individually or in combination, Java, C, C++, C#, Objective C, Fortran, Pascal, JavaScript, Python, Perl, and PL / SQL.

[0095] When executed by processor 1130, the computer-executable instructions of non-volatile storage device 1126 can enable control system 1112 to implement one or more ML algorithms and / or method techniques as disclosed herein. Non-volatile storage device 1126 may also include ML data (including data parameters) supporting the functionality, features, and processes of one or more embodiments described herein.

[0096] Program code embodying the algorithms and / or methodologies described herein can be distributed individually or collectively as a program product in a wide variety of different forms. The program code can be distributed using a computer-readable storage medium having computer-readable program instructions thereon for causing a processor to implement aspects of one or more embodiments. A computer-readable storage medium—which is inherently non-transitory—can include volatile and non-volatile, as well as removable and non-removable tangible media implemented in any method or technology, for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer-readable storage media can also include RAM, ROM, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other solid-state memory technologies, portable optical disc read-only memory (CD-ROM) or other optical storage devices, magnetic tape cassettes, magnetic tape, disk storage devices or other magnetic storage devices, or any other medium that can be used to store desired information and is readable by a computer. Computer-readable program instructions can be downloaded from the computer-readable storage medium to a computer, another type of programmable data processing device, or another device, or downloaded via a network to an external computer or external storage device.

[0097] Computer-readable program instructions stored in a computer-readable medium can be used to direct a computer, other type of programmable data processing apparatus, or other device to operate in a particular manner, causing the instructions stored in the computer-readable medium to produce an article of art, including instructions that implement the functions, actions, and / or operations specified in a flowchart or figure. Consistent with one or more embodiments, in some alternative embodiments, the functions, actions, and / or operations specified in the flowchart and figure can be reordered, serialized, and / or processed simultaneously. Furthermore, any flowchart and / or figure may include more or fewer nodes or blocks than those illustrated consistent with one or more embodiments.

[0098] Processes, methods, or algorithms may be embodied, in whole or in part, using suitable hardware components such as application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), state machines, controllers, or other hardware components or devices, or a combination of hardware, software, and firmware components.

[0099] Figure 12 An example manufacturing system 1200 for anomaly detection and / or generation of synthetic anomaly data is illustrated. System 1200 can be configured to control manufacturing machines 1202, such as punch presses, cutting machines, or gun drills, as part of a production line.

[0100] System 1200 can be configured to control actuator 1114, which is configured to control manufacturing machine 1202. Sensor 1116 of system 1200 can be configured to capture one or more properties of manufactured product 1204. ML processing 1124 can be configured to determine the state of manufactured product 1204 based on one or more captured properties. Actuator 1114 can be configured to control system 1200 (e.g., manufacturing machine) based on the determined state of manufactured product 1204 for subsequent manufacturing steps of manufactured product 1204. In particular, actuator 1114 can be configured to control system 1200 (e.g., manufacturing machine) on a subsequent manufactured product 1206 of system 1200 (e.g., manufacturing machine) based on the determined state of manufactured product 1204.

[0101] For example, system 1200 can use CausalKG 216 to predict the causes of problems in manufacturing system 1200, such as what the problem is. causedBy (For example, causedByType Alternatively, system 1200 can utilize CausalKG 216 to predict the outcome that should be addressed, such as whether the sensed input might cause a problem, for example... causesType .

[0102] The processes, methods, or algorithms disclosed herein may be transferable to / implemented by a processing device, controller, or computer, which may include any existing programmable electronic control unit or dedicated electronic control unit. Similarly, processes, methods, or algorithms may be stored in many forms as data and instructions executable by a controller or computer, including but not limited to information persistently stored on non-writable storage media such as read-only memory (ROM) devices, and information changeably stored on writable storage media such as floppy disks, magnetic tapes, optical discs (CDs), RAM devices, and other magnetic and optical media. Processes, methods, or algorithms may also be implemented in a software executable object. Alternatively, suitable hardware components, such as ASICs, FPGAs, state machines, controllers, or other hardware components or devices, or a combination of hardware, software, and firmware components, may be used to embody these processes, methods, or algorithms, wholly or partially.

[0103] One or more embodiments of causal discovery predict causal links between physical objects or components. Such causal links between physical objects or components can be dynamic and / or static interactions. These causal links between physical objects or components can be mechanical, electrical, and / or chemical interactions. For example, and without limitation on the physical objects or components that can be applied, the physical objects or components can be combined... Figure 7AThe cube and sphere are described, and the causal link can be a dynamic mechanical interaction between the cube and the sphere. As another non-limiting example, the physical objects or components can be manufacturing machine 1202 and manufactured product 1204, and the causal link can be a mechanical interaction between these components.

[0104] While exemplary embodiments have been described above, these embodiments are not intended to describe all possible forms covered by the claims. The terms used in this specification are descriptive rather than restrictive, and it will be understood that various changes may be made without departing from the spirit and scope of this disclosure. As described above, features of various embodiments may be combined to form further embodiments of the invention that may not be explicitly described or illustrated. While various embodiments may have been described as providing an advantage or superiority over other embodiments or prior art implementations with respect to one or more desired features, those skilled in the art will recognize that one or more features or characteristics may be waived to achieve desired overall system properties, depending on the particular application and implementation. These properties may include, but are not limited to, strength, durability, lifespan, merchantability, appearance, packaging, size, suitability, weight, manufacturability, ease of assembly, etc. Thus, if any embodiment is described to some extent as less desirable than other embodiments or prior art implementations with respect to one or more features, then such embodiments do not exceed the scope of this disclosure and may be desirable for a particular application.

Claims

1. A method for causal discovery using knowledge graph link prediction, comprising: Information from a causal network is transformed into a causal knowledge graph, which includes multiple causal links, each of which includes a cause entity, a causal relationship, and an effect entity. At least a subset of the causal links includes a causal hyperrelation link, which further connects the cause entity and the effect entity to an associated qualifier entity as a medium, the medium being the intermediary in the serial causal connection between the cause entity and the effect entity. The causal knowledge graph is transformed into an embedding, the embedding including the latent vector space representation of the causal knowledge graph; The embedding is trained using the causal links and causal hyperrelation links of the causal knowledge graph; as well as The embedding is used for causal discovery to predict additional causal links in the causal knowledge graph.

2. The method according to claim 1, wherein, Transforming the causal knowledge graph into an embedding involves using a method based on a StarE graph neural network, where a variable number of qualifier pairs can be configured to be associated with each of the causal links.

3. The method according to claim 1, wherein, The transformation is performed in accordance with a causal ontology, which defines concepts to structure the causal knowledge graph.

4. The method according to claim 1, wherein, The transformation also includes: Mapping nodes in the causal network to causal entities in the causal knowledge graph; and The edges in the causal network are mapped to causal links in the causal knowledge graph.

5. The method according to claim 4, wherein, The qualifier entity includes a mediating entity concatenated between the cause entity and the result entity, such that the cause entity causes the result entity and / or the cause entity interprets the result entity.

6. The method according to claim 1, wherein, The causal hyperrelationship link includes qualifier relations, which include one or more of the following: a hasMediator qualifier relation indicating the presence or absence of a medium and / or a hasMediatorType qualifier relation indicating the type of the medium.

7. The method according to claim 1, wherein, The causal discovery includes causal interpretation to predict the type of the causal entity given the resulting entity.

8. The method according to claim 1, wherein, The causal discovery includes causal prediction to predict the type of the resulting entity given a causal entity.

9. A system for causal discovery using knowledge graph link prediction, comprising: One or more hardware computing devices are configured to: Information from a causal network is transformed into a causal knowledge graph, which includes multiple causal links, each of which includes a cause entity, a causal relationship, and an effect entity. At least a subset of the causal links includes a causal hyperrelation link, which further connects the cause entity and the effect entity to an associated qualifier entity as a medium, the medium being the intermediary in the serial causal connection between the cause entity and the effect entity. The causal knowledge graph is transformed into an embedding, the embedding including the latent vector space representation of the causal knowledge graph; The embedding is trained using the causal links and causal hyperrelation links of the causal knowledge graph; as well as The embedding is used for causal discovery to predict additional causal links in the causal knowledge graph.

10. The system of claim 9, wherein converting the causal knowledge graph into an embedding comprises using a method based on a StarE graph neural network, wherein a variable number of qualifier pairs can be configured to be associated with each of the causal links.

11. The system according to claim 9, wherein, The transformation of information is performed in accordance with a causal ontology, wherein the causal ontology defines concepts to structure the causal knowledge graph.

12. The system of claim 9, wherein the converted information further includes: Map the nodes in the causal network to the causal entities in the causal knowledge graph; as well as The edges in the causal network are mapped to causal links in the causal knowledge graph.

13. The system according to claim 12, wherein, The qualifier entity includes a mediating entity concatenated between the cause entity and the result entity, such that the cause entity causes the result entity and / or the cause entity interprets the result entity.

14. The system according to claim 9, wherein, The causal hyperrelationship link includes qualifier relations, which include one or more of the following: a hasMediator qualifier relation indicating the presence or absence of a medium and / or a hasMediatorType qualifier relation indicating the type of the medium.

15. The system according to claim 9, wherein, The qualifier entity includes a mediating entity concatenated between the cause entity and the result entity, such that the cause entity causes the result entity and / or the cause entity interprets the result entity.

16. A non-transitory computer-readable medium comprising instructions for causal discovery using knowledge graph linking prediction, wherein, when executed by one or more computing devices, the instructions cause the one or more computing devices to perform the following operations: Information from a causal network is transformed into a causal knowledge graph, which includes multiple causal links, each of which includes a cause entity, a causal relationship, and an effect entity. At least a subset of the causal links includes a causal hyperrelation link, which further connects the cause entity and the effect entity to an associated qualifier entity as a medium, the medium being the intermediary in the serial causal connection between the cause entity and the effect entity. The causal knowledge graph is transformed into an embedding, the embedding including the latent vector space representation of the causal knowledge graph; The embedding is trained using the causal links and causal hyperrelation links of the causal knowledge graph; as well as The embedding is used for causal discovery to predict additional causal links in the causal knowledge graph.

17. The medium according to claim 16, wherein, Transforming the causal knowledge graph into an embedding involves using a method based on a StarE graph neural network, where a variable number of qualifier pairs can be configured to be associated with each of the causal links.

18. The medium according to claim 16, wherein, The transformation of information is performed in accordance with a causal ontology, wherein the causal ontology defines concepts to structure the causal knowledge graph.

19. The medium according to claim 16, wherein, The causal hyperrelationship link includes qualifier relations, which include one or more of the following: a hasMediator qualifier relation indicating the presence or absence of a medium and / or a hasMediatorType qualifier relation indicating the type of the medium.

20. The medium according to claim 16, wherein, The qualifier entity includes a mediating entity concatenated between the cause entity and the result entity, such that the cause entity causes the result entity and / or the cause entity interprets the result entity.

Images