Transformer-based model for knowledge graph link prediction

Abstract

Systems, products, and methods are provided for improving knowledge graph (KG) link prediction using a transformer-based artificial neural network. A first topic model is utilized on a first dataset derived from a KG that includes a plurality of first triples. The first triples include a first entity and a first edge connecting the first entity to represent a relationship between the first connected entities. A first similarity function is applied to the first connected entities of the first triples to provide respective first similarity scores. A first subset of the one or more first triples is selected from the plurality of first triples based on the first similarity scores. The artificial neural network is trained using the selected first subset of the one or more first triples.

Description

[Technical Field]

[0001] The present embodiments relate to improving the performance of artificial neural network-based language models, and in particular transformer-based models, for predictions related to supplementing or completing knowledge graphs. [Background technology]

[0002] Artificial Intelligence (AI) relates to the field of computer science that focuses on computers and their behavior in relation to humans. AI refers to intelligence when a machine is capable of making informed decisions, thereby maximizing its chances of success in a given topic. More specifically, AI can learn from datasets to solve problems and provide relevant recommendations. For example, in the field of artificial intelligence (AI) computer systems, natural language systems (such as the IBM Watson® artificial intelligence computer system or other natural language question-answering systems) process natural language based on system-acquired knowledge. To process natural language, a system may be trained using data derived from a knowledge database or corpus, but the resulting outcomes may be inappropriate or inaccurate for various reasons.

[0003] Machine learning (ML), a subset of AI, uses algorithms to learn from data and create perspectives based on this data. AI refers to intelligence when a machine is able to make informed decisions, thereby maximizing the chances of success in a given topic. More specifically, AI can learn from datasets to solve problems and provide relevant recommendations. Cognitive computing is a blend of computer science and cognitive science. Cognitive computing uses self-learning algorithms that use minimal data, visual recognition, and natural language processing to solve problems and optimize human processing.

[0004] At the core of AI and associated reasoning lies the concept of similarity. The process of understanding natural language and objects requires reasoning in terms of relationships, which can be challenging. Structures, including static and dynamic structures, define the determined output or action for a given deterministic input. More specifically, the determined output or action is based on explicit or inherent relationships within the structure. To build these structures, appropriate datasets are relied upon. An example of such a dataset is a Knowledge Graph (KG), which integrates data using a graph-structured data model or topology.

[0005] Many AI systems rely on knowledge graphs (KGs) as a key resource for performing tasks such as question answering (QA), semantic search, and recommendation generation. A KG typically represents a collection or dataset of interlinked data composed of triples (also called triplets) of information. A KG contains descriptions of entities, represented by vertices or nodes, linked or connected to each other by relationships or predicates, which are represented by edges linking or connecting vertices or nodes in the KG. The interlinked entities of a triple are sometimes referred to as subjects and objects, with the relationship represented by the triple relationship. Entities may include, for example, real-world objects, events, situations, or abstract concepts. A KG can be constructed by identifying associations between different entities. KGs have evolved to the point where some contain millions or even billions of entries.

[0006] Some KGs are used in combination with ontologies, which include the representation, formal naming, and definition of the concepts, data, and categories, properties, and relationships between entities that underpin one, many, or all domains of discourse. Ontologies are created in every discipline to limit complexity and organize information into data and knowledge. When a new ontology is created, its use is expected to improve problem-solving within that domain.

[0007] KGs often contain a large amount of factual information with little formal semantics. For example, a KG in the genomics domain may contain entities such as genes, gene variants, drugs, and diseases, and may show relationships between the entities, such as the association of a particular gene variant with a particular disease. Traditional approaches to building KGs require domain-expert researchers to study new publications to discover relationships between entities. While KGs provide researchers with insight into the interrelationships of entities, manually assembling and updating KGs can be very complex and time-consuming. Another problem associated with KGs is that the data they contain is often incomplete.

[0008] On the other hand, Open KG does not require a pre-specified ontology, making it more adaptable. In Open KG, the same potential entity can be represented by different nodes labeled with different noun phrases (NPs). For example, the entity "President of the United States" can be represented by the first node "President of the United States" and the second node "President". Similarly, in Open KG, the same potential relationship can be represented by different relation phrases (RPs). For example, the RPs "born" and "born" refer to the same underlying relationship.

[0009] By using a link prediction model, subject matter experts (SMEs) can be at least partially removed from the KG preparation process. The purpose of the link prediction model is to predict missing entities (nodes) or relationships (edges) in the KG. The prediction model evaluates the validity of triples not present in the KG to supplement or complete the KG. For example, link prediction may involve answering queries that identify subjects and relationships to predict missing objects. As a specific example, link prediction may provide a link prediction model with a query for "First Lady" as the subject and "Spouse" as the relationship to identify "President of the United States" as the object. In another embodiment, the KG may include a link between "First Lady" and "President of the United States," but may lack a similar link between "First Lady" and "United States." The prediction model in this example identifies an edge representing a relationship between the First Lady and the President of the United States (or "President"), entities whose relationship is missing from the incomplete KG.

[0010] Liang Yao et al., KG-BERT: BERT for Knowledge Graph Completion, incorporated herein by reference, proposes using a pre-trained language model known as Knowledge Graph Bidirectional Encoder Representation by Transfer (KG-BERT) to model triples. Yao describes BERT as a pre-trained contextual language representation model built on a multi-layer bidirectional transformer encoder. According to Yao, entity and relationship descriptions of triples are received as text-order inputs to the KG-BERT model, which then calculates a scoring function for the triples. This method is said to achieve "state-of-the-art results in triple classification, relationship prediction, and link prediction tasks." Triple classification aims to determine whether a given subject-relation-object (s, r, o) triple is correct. Link prediction predicts the missing element (subject or object) by considering other elements (object or subject) and relationships, i.e., (s, r, ?) or (?, r, o). Relation prediction aims to predict relations from subjects and objects, i.e., (s, ?, o).

[0011] However, the inventors have discovered significant technical problems related to scalability in connection with training and using predictive models on large datasets. The execution time for training, testing (or validation), and using a predictive model can be long, especially when the associated KG is complex, such as when the KG contains millions or billions of entries. The execution time for training and testing, especially for large KGs or datasets, can theoretically take days, months, or even years, thus creating significant technical problems. Summary of the Invention

[0012] Embodiments include systems, computer program products, and methods for Natural Language Processing (NLP), including NLP directed to improving the performance of language models, particularly Transformer-based models, in knowledge graph link prediction and / or relation prediction.

[0013] In one aspect, a system includes a processing unit operably coupled to a memory and an artificial intelligence (AI) platform in communication with the processing unit and the memory. The AI ​​platform includes one or more tools supporting modification of a knowledge graph, the tools including a topic model manager and a machine learning manager (MLM). The topic manager is configured to utilize a first topic model on a first dataset derived from a knowledge graph (KG) including a plurality of first triples. The first triple includes a first entity and a first edge connecting the first entity to represent a relationship between the first connected entities. The topic model is further configured to apply a first similarity function to the first connected entities of the first triples to provide respective first similarity scores and select a first subset of one or more first triples from the plurality of first triples based on the first similarity scores. The MLM is configured to train an artificial neural network using the selected first subset of the one or more first triples.

[0014] In another aspect, a computer program product is provided. The computer program product includes a computer-readable storage device having program code embodied therein that is executable by a processor. The program code is executable by the processor and utilizes a first topic model on a first dataset derived from a knowledge graph (KG) that includes a first entity and a plurality of first triples, the first triples including a first edge connecting the first entity and representing a relationship between the first connected entities. The program code is further executable by the processor and is configured to apply a first similarity function to the first connected entities of the first triples to provide respective first similarity scores, select a first subset of one or more first triples from the plurality of first triples based on the first similarity scores, and train an artificial neural network (ANN) using the selected first subset of the one or more first triples.

[0015] In yet another aspect, a method is provided for utilizing a first topic model on a first dataset derived from a knowledge graph (KG) including a plurality of first triples. The first triples include a first entity and a first edge connecting the first entity to represent a relationship between the first connected entities. A first similarity function is applied to the first connected entities of the first triples to provide respective first similarity scores. A first subset of one or more first triples is selected from the plurality of first triples based on the first similarity scores. An artificial neural network is trained using the selected first subset of the one or more first triples.

[0016] These and other features and advantages will become apparent from the following detailed description of illustrative embodiments taken in conjunction with the accompanying drawings. [Brief explanation of the drawings]

[0017] The drawings referenced herein form part of the specification, and unless otherwise specified, the features shown in the drawings are meant to illustrate only some embodiments, and not all embodiments.

[0018] [Figure 1] FIG. 1 shows a system diagram illustrating a schematic diagram of a system including a transformer-based model, according to one embodiment.

[0019] [Figure 2] 2 shows a block diagram illustrating the tools shown in FIG. 1 and their associated APIs, according to one embodiment.

[0020] [Figure 3] 1 shows a flowchart illustrating one embodiment of a method that includes using a first topic model to select a training set for training an artificial neural network (ANN).

[0021] [Figure 4] 10 shows a flowchart illustrating one embodiment of a method that includes using a trained ANN or a second topic model to select a candidate set for testing.

[0022] [Figure 5] 1 shows a flowchart illustrating one embodiment of a method including use of a trained ANN, for example, in connection with testing, validation, or prediction.

[0023] [Figure 6] FIG. 1 shows a block diagram illustrating an example of a computer system / server of a cloud-based support system for implementing the systems and processes described above with respect to FIGS. 1-5.

[0024] [Figure 7] 1 shows a block diagram illustrating a cloud computing environment.

[0025] [Figure 8] 1 shows a block diagram illustrating a set of functional abstraction model layers provided by a cloud computing environment. DETAILED DESCRIPTION OF THE INVENTION

[0026] It will be readily understood that the components of the present embodiments, as generally described and illustrated in the Figures herein, could be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the present apparatus, system, method and computer program product embodiments, as illustrated in the Figures, is not intended to limit the scope of the claimed embodiments, but rather represents selected embodiments.

[0027] Throughout this specification, references to "a select embodiment," "one embodiment," or "an embodiment" mean that a particular feature, structure, or characteristic described in connection with an embodiment is included in at least one embodiment. Thus, the appearances of the phrases "a select embodiment," "in one embodiment," or "in an embodiment" in various places throughout this specification are not necessarily referring to the same embodiment. It should be understood that various embodiments can be combined with each other, and that any one embodiment can be used to modify another embodiment.

[0028] The illustrated embodiments are best understood by referring to the drawings, in which like parts are designated by like numerals throughout. The following description is intended as an example only and merely illustrates certain selected embodiments of devices, systems and processes consistent with the embodiments as claimed herein.

[0029] As shown and described herein, a technical solution to this technical problem is provided by developing the systems, computer program products, methods, and other aspects described and / or illustrated herein.

[0030] Referring to Figure 1, a schematic diagram of a system 100 is shown. As shown, a server 110 is provided in communication with multiple computing devices 180, 182, 184, 186, 188, and 190 via a computer network (also referred to herein as a network connection) 105. The server 110 is configured to receive input content 102 from the computer network 105 and transmit output content 104 to the computer network 105. The server 110 is provided with a processor 112, also referred to herein as a processing unit, operatively coupled to a memory 116 via a bus 114. A tool in the form of an artificial intelligence (AI) platform 150, also referred to herein as a knowledge engine, is shown local to the server 110 and operatively coupled to the processing unit 112 and the memory 116. As shown, the AI ​​platform 150 includes one or more tools 152 and 154. The tools 152 and 154 provide management for enabling training and inference processing of Transformer-based artificial neural network (ANN) models, including KG-BERT, from one or more computing devices 180, 182, 184, 186, 188, and 190 via a computer network 105.

[0031] The computing devices 180, 182, 184, 186, 188, and 190 communicate with each other and with other devices or components via one or more wired and / or wireless data communication links, each of which may comprise one or more of electrical wires, routers, switches, transmitters or receivers, or the like. In this networked configuration, the server 110 and network connections 105 may enable the exchange of input content 102 and output content 104, e.g., model training and inference operations, to one or more content users. Other embodiments of the server 110 may be used with components, systems, subsystems, and / or devices other than those shown herein.

[0032] As shown and described herein, the tools of the AI ​​platform (150) include a topic model manager (152) and a machine learning model (MLM) manager (154). The AI ​​platform (150) may be configured to receive and / or send inputs from a variety of sources, including, but not limited to, inputs from a network (105) via input content (102) and output content (104), one or more knowledge graph (KG) datasets from a knowledge base or corpus structure (160), topic models (172), and artificial neural networks (ANNs) (174).

[0033] As shown, the knowledge base (160) includes a library of datasets, which in one embodiment includes a first dataset, Dataset0 (1620), and a second dataset, Dataset1 (1621). In an exemplary embodiment, the datasets, Dataset0 (1620) and Dataset1 (1621), represent different data derived from the same knowledge graph (KG). For example, data from a KG may be split into Dataset0 (1620), a subset of which is used for training, and Dataset1 (1621), a subset of which is used for testing or inference. In an exemplary embodiment, Dataset0 (1620) and Dataset1 (1621) are embodied as a KG. Alternatively, either or both of Dataset0 (1620) and Dataset1 (1621) may be embodied in another form, specifically as structured data derived from the KG, such as, for example, an embedded lookup table generated by a topic model (172). The amount of data set presented here should not be considered limiting.

[0034] As understood in the art, a KG is a structure (e.g., an ontology) and not simply a store of data. Specifically, Information Extraction (IE) models, examples of which are known in the art, extract data and one or more data relationships from unstructured data, create entries in a KG for the extracted data and data relationships, and store the data and data relationships in the created KG entries. IE models may be used to automatically extract triples from unstructured text. Alternatively, datasets may be compiled manually, such as by subject matter experts (SMEs).

[0035] In one embodiment, data elements in a KG are stored or represented in nodes, and a relationship between two data elements is represented as an edge connecting or linking the two nodes. Similarly, in one embodiment, each node has a node-level confidence or accuracy value, and each relationship has a relationship confidence or accuracy value, and the relationship accuracy value is calculated based on the accuracy values ​​of the two interconnected nodes.

[0036] In Figure 1, a first dataset, Dataset0 (1620), represents data derived from a knowledge graph (KG) that includes a first plurality of triples. Specifically, Dataset0 (1620) includes Triple0,0 (1640,0), Triple0,1 (1640,1), and Triple0, m (1640, m ), where m is an integer equal to or greater than 2. While three triples are shown in the first dataset, Dataset0 (1620), it should be understood that Dataset0 (1620) may include additional triples, including tens, hundreds, thousands, millions, or billions of triples, e.g., "m" may be equal to, for example, 10 to 1 billion. Each Triple0,0 (1640,0), Triple0,1 (1640,1), ..., and Triple0, m (1640, m ) contains subjects, relations, and objects. For example, as shown herein, Triple0,0 (1640,0) contains Subject0,0 (1660,0), Relation0,0 (1670,0), and Object0,0 (1680,0). Triple0,1 (1640,1) contains Subject0,1 (1660,1), Relation0,1 (1670,1), and Object0,1 (1680,1). Triple0, m (1640, m ) includes Subject0, m (1660, m ), Relation0, m (1670, m ), and Object0, m (1680,m ) is included.

[0037] The topic model manager (152) is configured to utilize a topic model (172) for a first dataset, Dataset0 (1620). Broadly speaking, topic modeling is a machine learning technique that automatically analyzes text data to determine cluster words for a set of documents. Because this technique does not require a predefined list of human-classified tags or training data, it is known as "unsupervised" machine learning. In machine learning and natural language processing, a topic model is a type of statistical model for discovering abstract "topics" occurring within a collection of documents. Examples of topic models include, but are not limited to, Latent Semantic Analysis (LSA) and Word2Vec. Topic modeling is a text mining tool frequently used to discover hidden semantic structures within bodies of text.

[0038] According to an exemplary embodiment, utilizing a topic model (172) for a first dataset, Dataset0 (1620), downsamples the amount of data used to train the ANN (174) by selecting a subset of the first dataset, Dataset0 (1620), for training. Training with the downsampled subset, Dataset0 (1620), also referred to as the training set, of the first dataset, rather than the entire first dataset, reduces the computational time associated with training the ANN (174). According to an exemplary embodiment, the first similarity function is a cosine similarity function applied to the subject and object of each triple to generate a first similarity score. In an exemplary embodiment, the cosine similarity function is applied by a topic model, e.g., LSA or Word2Vec. For example, the similarity score of Triple0,0(1640,0), e.g., SimilarityScore0,0, is the cosine similarity score calculated from Subject0,0(1660,0) and Object0,0(1680,0), and the similarity score of Triple0,1(1640,1), e.g., SimilarityScore0,1, is the cosine similarity score calculated for Subject0,1(1660,1) and Object0,1(1680,1). m (1640, m ) similarity score, e.g., SimilarityScore0, m Subject0, m (1660, m ) and Object0, m (1680, m ) is the cosine similarity score calculated for

[0039] Similarity scores SimilarityScore0,0, SimilarityScore0,1, ... and SimilarityScore0, mBased on this, the topic model (172) calculates the triples 0,0 (1640,0), 0,1 (1640,1), ..., and 0,1 (1640,2) of the first dataset, Dataset0 (1620). m (1640, m ), where the subset includes 1 to "m minus one" (m-1) triples. According to one embodiment, the triples Triple0,0(1640,0), Triple0,1(1640,1), ... and Triple0, m (1640, m ) are ranked by the topic model manager (152), and the topic model manager (152) selects a predetermined amount of the top-ranked triples. According to another embodiment, the topic model manager (152) selects Triple0,0 (1640,0), Triple0,1 (1640,1), ..., and Triple0,1 (1640,2), ..., from a first dataset, Dataset0 (1620), that have similarity scores that meet a first threshold, as a training set. m (1640, m ) to select a subset of triples. For example, the training set may contain Triple 0,0(1640,0) and Triple 0,1(1640,1) that satisfy the first threshold, but not Triple 0, m (1640, m ) may be omitted. In embodiments in which the similarity score reflects cosine similarity, the cosine similarity indicates a measure of how similar two data members are likely to be with respect to the subject matter. Generally, cosine similarity scores range from negative one (-1) to one (1), with a score of negative one representing opposites, a score of one representing identity, and a score of zero representing no correlation. In one embodiment, the first threshold is set between 0 and 1, such as 0.5, and according to this embodiment, all scores between 0.5 and 1 satisfy the first threshold.

[0040] The MLM manager (154) is configured to train the ANN (174) using one or more triples of a selected training set, e.g., a subset of the first dataset, Dataset0 (1620), Triple0,0 (1640,0) and Triple0,1 (1640,1). In an exemplary embodiment, the ANN (174) is trained using the training triples for triple classification, link prediction, relationship prediction, combinations thereof, and / or one or more additional operations and tasks.

[0041] During training in one embodiment, the MLM manager (154) uses the ANN (174) to assign or otherwise specify a reliability or accuracy value to the data in the training set.

[0042] The AI ​​platform (150) is further configured to receive a query (176). Various computing devices (180), (182), (184), (186), (188), and (190) in communication with the network (105) represent access points for transmitting the query (176). Alternatively, the query (176) may be accessed from the knowledge base (160), another knowledge base, user input, or another source. In one embodiment, the query (176) is subjected to natural language processing (NLP) to extract a first entity, Entity1 (177), and a relationship, Relation (178).

[0043] In the embodiment shown in FIG. 1, the query (176) is a query containing known entities. KNOWN (177) and Relation (178), but unknown Entity UNKNOWN (179) is not included. UNKNOWN (179) is Entity KNOWN In one embodiment, Entity UNKNOWN (179) is Entity KNOWN(177) and Relation (178). According to one embodiment, Query (176) is an answer to the question posed by Entity (177) and Relation (178). KNOWN Entity is an object that contains a Subject (177) and Relations (178) and represents the answer to a Query (176). UNKNOWN In another embodiment, the query (176) is KNOWN Entity, a subject that contains objects (177) and relations (178) and represents the answer to a query (176). UNKNOWN (179) is required.

[0044] The topic model manager (152) is configured to utilize, for inference or testing purposes, a topic model (172) trained on a second dataset, represented in Figure 1 as Dataset1 (1621). Additionally, while one topic model (172) is shown in Figure 1, it should be understood that more than one topic model may be accessible and available, e.g., a first topic model for downsampling a dataset before training the ANN (174) and a second topic model for sample limiting another dataset before testing with the trained ANN (174).

[0045] According to an exemplary embodiment, the second dataset, Dataset1 (1621), represents additional data derived from the same KG as the first dataset, Dataset0 (1620). In one embodiment, datasets Dataset0 (1620) and Dataset1 (1621) are different and are derived from the same KG, such as the embedding lookup table generated by the topic model (172). In another embodiment, Dataset0 (1620) and Dataset1 (1621) are mutually exclusive; that is, the data in Dataset1 (1621) was not used for training. Dataset1 (1611) has a second entity and a second edge linking or connecting the second entity. Dataset1 (1621) contains multiple triples, specifically, Triple1,0 (1641,0), Triple1,1 (1641,1), ..., and Triple1, n (1641, n ), where n is an integer equal to or greater than 2. While three triples are shown in Dataset1 (1621), it should be understood that Dataset1 (1621) may contain additional triples, including tens, hundreds, thousands, millions, or billions of triples. Triple1,0 (1641,0) contains Subject1,0 (1661,0), Relation1,0 (1671,0), and Object1,0 (1681,0). Triple1,1 (1641,1) contains Subject1,1 (1661,1), Relation1,1 (1671,1), and Object1,1 (1681,1). Triple1, n (1641, n ) includes Subject1, n (1661, n ), Relation1, n (1671, n ), and Object1, n (1681, n ) is included.

[0046] According to an exemplary embodiment, utilizing a topic model (172) for a second dataset, Dataset1 (1621), down-samples the amount of data used to test or use the ANN (174), for example, for question answering query resolution, by selecting a subset of the second dataset, Dataset1 (1621), as a candidate set. A candidate set that is a subset of the second dataset, Dataset1 (1621), rather than the entire second dataset, Dataset1 (1621), reduces the computational time associated with testing or using the ANN (174). According to one embodiment, the topic model (172) applies a second similarity function to the triples Triple1,0 (1641,0), Triple1,1 (1641,1), and Triple1, n (1641, n ) applies to

[0047] The second similarity function evaluated by the topic model (172) may be the same as or different from the first similarity function. According to an exemplary embodiment, the second similarity function is a cosine similarity function applied to the subject and object of each triple to generate a second similarity score. For example, the similarity score for Triple1,0 (1641,0), e.g., SimilarityScore1,0, is the cosine similarity score calculated from Subject1,0 (1661,0) and Object1,0 (1681,0), and the similarity score for Triple1,1 (1641,1), e.g., SimilarityScore1,1, is the cosine similarity score calculated for Subject1,1 (1661,1) and Object1,1 (1681,1 ... Triple1,1 (1641,1). n (1641, n ) similarity score, e.g., SimilarityScore1, n Subject1, n (1661, n ) and Object1, n (1681, n) is the cosine similarity score calculated for

[0048] Similarity scores SimilarityScore1,0, SimilarityScore1,1, ... and SimilarityScore1, n Based on this, the topic model manager (152), or in one embodiment the MLM manager (154), generates a second dataset, Dataset1 (1621), including Triple1,0 (1641,0), Triple1,1 (1641,1), ..., and Triple1, n (1641, n ), where the subset includes 1 to "n minus one" (n-1) triples. According to one embodiment, the triples Triple1,0 (1641,0), Triple1,1 (1641,1), ... and Triple1, n (1641, n ) are ranked by the topic model manager (152), and a predetermined number of the top-ranked triples are selected as a candidate set. According to another embodiment, the topic model (152) selects Triple1,0 (1641,0), Triple1,1 (1641,1), ..., and Triple1,... of a second dataset, Dataset1 (1621), that have similarity scores that meet a second threshold. n (1641, n ) as a candidate set. For example, the training set may include Triple1,0 (1641,0) and Triple1,1 (1641,1) that satisfy the second threshold, but Triple1, n (1641, n) may be omitted. In embodiments in which the similarity score reflects cosine similarity, the cosine similarity indicates a measure of how similar two data members are likely to be with respect to the subject matter. Generally, cosine similarity scores range from negative one (-1) to one (1), with a score of negative one representing opposites, a score of one representing identity, and a score of zero representing no correlation. In one embodiment, the second threshold is set between 0 and 1, such as 0.5, and according to this embodiment, all scores between 0.5 and 1 satisfy the second threshold.

[0049] The MLM manager (154) utilizes the trained ANN (174) to, for each candidate triple in the second subset, UNKNOWN In one or more exemplary embodiments, utilizing an ANN (174) trained on the second subset (or candidate triples) includes evaluating the confidence scores of the entities in the query (176) and the candidate triples corresponding to the query (179). UNKNOWN In one embodiment, the Entity of the Query (176) is evaluated or scored based on the confidence value of each candidate triple corresponding to the Query (179). KNOWN If (177) is the subject, then evaluation by the trained ANN (174) will determine the candidate set of objects, e.g., Object1,0 (1681,0), Object1,1 (1681,1), ... and / or Object1, n (1681, n ) is determined. KNOWN If (177) is the object, the evaluation will select each subject in the candidate set, e.g., Subject1,0 (1661,0), Subject1,1 (1661,1), ... and / or Subject1, n (1661, n ) is determined. At least one of the second entities in the second subset is determined to be an entity of the query (176) based on the confidence score. UNKNOWNIt is output as a candidate for (179).

[0050] Various computing devices 180, 182, 184, 186, 188, and 190 in communication with the network 105 represent access points for transmitting the query 176. Some of the computing devices may include the knowledge engine 150 and, in one embodiment, a database device for storing a corpus of data as a mass of information used by the tools 152 and 154. The computer network 105 may, in various embodiments, include local network connections and remote connections so that the knowledge engine 150 and embedded tools 152 and 154 may operate in any scale of environment, including local and global, e.g., the Internet.

[0051] The server (110) may be an IBM Watson® system available from International Business Machines Corporation of Armonk, NY, extended with the features of the exemplary embodiment described below. The IBM Watson® Knowledge Manager system processes multiple triples (e.g., Triple 0,0 (1640,0), Triple 0,1 (1640,1), ... Triple 0, m (1640, m ) into the topic model (172). Specifically, in one embodiment, the triples in the first dataset, Dataset0 (1620), are downsampled to a subset of triples, also referred to herein as training triples. The training triples are used to train the ANN (174).

[0052] Additionally, the IBM Watson® Knowledge Manager system can handle multiple triples (e.g., Triple1,0 (1641,0), Triple1,1 (1641,1), ... Triple1,n (1641, n A second dataset, Dataset0 (1621), having the following features, is imported into the topic model (172). Specifically, in one embodiment, the triples in the second dataset, Dataset1 (1621), are downsampled to a subset of triples, also referred to herein as candidate triples. A trained ANN (174) is utilized on the candidate triples and the query.

[0053] The server 110 utilizes an AI platform 150 and associated tools 152 and 154 to support the training and testing of the ANN 174. While the AI ​​platform 150 is shown embodied within or integrated with the server 110, it may be implemented within a separate computing system (e.g., 190) connected to the server 110 via a computer network 105. While shown local to the server 110, the tools 152 and 154 may be collectively or individually embedded in memory 114 or elsewhere.

[0054] The types of information handling systems that can utilize the system 110 range from small handheld devices, such as a handheld computer / cell phone 180, to large mainframe systems, such as a mainframe computer 182. Examples of handheld computers 180 include personal entertainment devices, such as personal digital assistants (PDAs), MP4 players, portable televisions, and compact disc players. Other examples of information handling systems include pen or tablet computers 184, laptop or notebook computers 186, personal computer systems 188, and servers 190. As shown, various information handling systems can be networked together using a computer network 105. Types of computer networks 105 that can be used to interconnect various information handling systems include local area networks (LANs), wireless local area networks (WLANs), the Internet, public switched telephone networks (PSTNs), other wireless networks, and any other network topology that can be used to interconnect information handling systems. Many information handling systems include non-volatile data stores, such as hard drives and / or non-volatile memory. Some of the information processing systems may use separate non-volatile data stores (e.g., server 190 utilizes non-volatile data store 190A, while mainframe computer 182 utilizes non-volatile data store 182a). Non-volatile data store 182a may be a component external to the various information processing systems or may be internal to one of the information processing systems.

[0055] Information handling systems may take a variety of forms, some of which are illustrated in Figure 1. For example, an information handling system may take the form of a desktop, server, portable laptop, notebook, or other form factor computer or data processing system. Additionally, an information handling system may take other form factors, such as a personal digital assistant (PDA), gaming device, ATM machine, portable telephone device, communications device, or other device that includes a processor and memory.

[0056] An application program interface (API) is understood in the art as a software intermediary between two or more applications. With respect to the system shown and described in FIG. 1, one or more APIs may be utilized to support one or more tools (152) and (154) and their associated functionality. Referring to FIG. 2, a block diagram (200) illustrating NL processing tools and their associated APIs is provided. As shown, multiple tools are embedded within the knowledge engine (205), including a topic model manager (252) associated with API0 (212) and an MLM manager (254) associated with API1 (222). Each of the APIs may be implemented in one or more languages ​​and interface specifications. API0 (212), associated with the topic model manager (252), provides for the use of one or more topic models for a dataset and the selection of a subset; API1 (222), associated with the MLM manager (254), provides for training an ANN on a selected first subset and utilizing the trained ANN on a selected second subset, e.g., for testing purposes. As shown, each of the APIs 212 and 222 is operatively coupled to an API orchestrator 260, also referred to as an orchestrator layer, which is understood in the art to function as an abstraction layer that threads transparently with the separate APIs. In one embodiment, the functionality of the separate APIs may be chained or combined. Thus, the organization of the APIs shown herein should not be considered limiting. Thus, the functionality of the tools may be embodied or supported by their respective APIs, as shown herein.

[0057] To provide additional details for a better understanding of selected embodiments of the present disclosure, reference is now made to Figure 3, which illustrates a flowchart (300) of a method for using a first topic model to downsample a dataset to a subset of training data for training an artificial neural network (ANN). Referring to Figure 3, in an exemplary embodiment, a first dataset having or derived from one or more knowledge graphs containing a plurality of triples is accessed (302). The total number of triples is quantified, M TOTAL A variable M representing the triples in the dataset is initialized (306). The first topic model is set as a triple, M In (310), the first similarity score, Score M But Triple M According to an exemplary embodiment, the first similarity score is evaluated for Triple M The variable M is then incremented (312) to determine whether each of the triples in the first data set has been evaluated, e.g., whether the incremented value of M is equal to the total number of triples in the data set, M TOTAL A determination is made (314) as to whether M is greater than M. If the answer is negative, the method returns to step (308) to utilize the first topic model for the triple using the incremented variable M. If the evaluation is answered in the affirmative at step (314), the triple, Triple1 to Triple2 is

number

[0058] 4 illustrates a flowchart (400) for one embodiment of a method for downsampling a second dataset using a second topic model into a subset of candidates for use, testing, or validation of a trained ANN. Referring to FIG. 4, in an exemplary embodiment, a second dataset is accessed (402) having or derived from one or more knowledge graphs containing a plurality of triples. The total number of triples is quantified, N Total A count variable N is initialized 404 to represent the triples in the second dataset. A second topic model, which may be the same as or different from the first topic model discussed above in connection with FIG. 3, is set as triples, N In (410), the second similarity score, Score N , Triple N According to an exemplary embodiment, the second similarity score is evaluated as Triple N The variable N is then incremented (412) and a determination is made as to whether each of the triples in the second data set has been evaluated, e.g., whether the incremented value of N is greater than or equal to N. TOTALA determination is made (414) as to whether Triple1 is greater than Triple2. A negative response in determination step (414) indicates that one or more additional triples remain for processing, and the method returns to step (408) to utilize the second topic model for the triples using the incremented variable N. A positive response in determination step (414) indicates that all of the triples have been processed, and the method proceeds to step (416).

number

[0059] Referring to Figure 5, a flowchart (500) is provided illustrating an embodiment method of using the trained ANN, for example, in connection with testing, validation, or prediction. A query is received (502) that includes known entities and relationships, but omits unknown entities that have relationships to the known entities. A second set of candidate triples, developed according to the method of flowchart (400) of Figure 4, is accessed (504). The total number of candidate triples is denoted by P TOTAL (506), and a count variable P is initialized (508). P For the Candidate Triple corresponding to the unknown entity in the query, P The entity is Candidate Entity P For example, if the query contains a subject as an unknown entity, the candidate triple P The subject of is Candidate Entity POn the other hand, if the query contains an object as an unknown entity, it is identified as a Candidate Triple. P The object is a Candidate Entity P is identified as

[0060] Using the trained ANN, Candidate Triple P Candidate Entity P Trust score of P At (514), a count variable P is incremented and a count is calculated to determine whether each of the candidate triples has been processed with respect to the evaluation of the confidence score, e.g., whether the incremented count variable P is equal to or greater than P TOTAL A determination is made whether ≡(x,y)≡(x,y) is greater than ≡(x,y) ...

[0061] Candidate Entity1 to Candidate

number

[0062] According to one embodiment, one or more additional edges and / or nodes associated with one or more identified candidate entities are selectively embedded in the KG 522. Thus, as shown herein, a topic model is utilized to effectively constrain candidates for training, testing, and / or processing by the ANN, and then identify candidate entities for selectively embedding one or more edges in the corresponding KG.

[0063] Aspects of using topic models to train, test, validate, and / or use ANNs are illustrated and described using the tools and APIs shown in Figures 1 and 2 and the processes shown in Figures 3, 4, and 5. Aspects of the functional tools 152 and 154 and their associated functionality may be embodied in a computer system / server at a single location, or, in one embodiment, configured in a cloud-based system that shares computing resources. Referring to Figure 6, a block diagram 600 is provided illustrating an example of a computer system / server 602, hereinafter referred to as a host 602, in communication with a cloud-based support system for implementing the processes described above with respect to Figures 3-5. The host 602 is operable in numerous other general-purpose or special-purpose computing system environments or configurations. Examples of well-known computing systems, environments, and / or configurations that may be suitable for use with the host (602) include, but are not limited to, personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputer systems, mainframe computer systems, and file systems (e.g., distributed storage environments and distributed cloud computing environments) that include any of the above systems, devices, and their equivalents.

[0064] The host 602 may be described in the general context of computer system-executable instructions, such as program modules, being executed by a computer system. Generally, program modules may include routines, programs, objects, components, logic, data structures, etc. that perform particular tasks or implement particular abstract data types. The host 602 may be implemented in a distributed cloud computing environment 610 where tasks are performed by remote processing devices linked through a communications network. In a distributed cloud computing environment, program modules may be located in both local and remote computer system storage media, including memory storage devices.

[0065] As shown in Figure 6, the host (602) is depicted in the form of a general-purpose computing device. Components of the host (602) may include, but are not limited to, one or more processors or processing units (604), such as a hardware processor, a system memory (606), and a bus (608) coupling various system components including the system memory (606) and the processing unit (604). The bus (608) represents one or more of any of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, and a processor or local bus using any of a variety of bus architectures. By way of example and not limitation, such architectures include an Industry Standard Architecture (ISA) bus, a MicroChannel Architecture (MCA) bus, an Enhanced ISA (EISA) bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus. The host (602) typically includes a variety of computer system-readable media. Such media can be any available media that is accessible by the host 602 and includes both volatile and non-volatile media, removable and non-removable media.

[0066] The memory 606 may include computer system-readable media in the form of volatile memory, such as random access memory (RAM) 630 and / or cache memory 632. By way of example only, the storage system 634 may be provided for reading from and writing to a non-removable, non-volatile magnetic medium (not shown, commonly referred to as a "hard drive"). Although not shown, a magnetic disk drive may be provided for reading from and writing to a removable, non-volatile magnetic disk (e.g., a "floppy disk"), and an optical disk drive may be provided for reading from or writing to a removable, non-volatile optical disk, such as a CD-ROM, DVD-ROM, or other optical medium. In such an example, each may be connected to the bus 608 by one or more data media interfaces.

[0067] A program / utility 640 having a set (at least one) of program modules 642 may be stored in memory 606, as well as, by way of example and not limitation, an operating system, one or more application programs, other program modules, and program data. Each of the operating system, one or more application programs, other program modules, and program data, or some combination thereof, may comprise an implementation of a network environment. The program modules 642 generally perform the functionality and / or methodology of embodiments to support and enable downsampling and ANN training and use. For example, the set of program modules 642 may include tools 152 and 154 described in FIG. 1.

[0068] The host (602) may also communicate with one or more external devices (614), such as a keyboard, pointing device, etc.; a display (624); one or more devices that allow a user to interact with the host (602); and / or any device (e.g., a network card, modem, etc.) that allows the host (602) to communicate with one or more other computing devices. Such communication may occur via an input / output (I / O) interface (622). Furthermore, the host (602) may communicate with one or more networks, such as a local area network (LAN), a general wide area network (WAN), and / or a public network (e.g., the Internet), via a network adapter (620). As shown, the network adapter (620) communicates with other components of the host (602) via a bus (608). In one embodiment, multiple nodes of a distributed file system (not shown) communicate with the host (602) via the I / O interface (622) or the network adapter (620). Although not shown, it is understood that other hardware and / or software components may be used in conjunction with host 602. Examples include, but are not limited to, microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, data archive storage systems, etc.

[0069] In this document, the terms "computer program medium," "computer usable medium," and "computer readable medium" are used generally to refer to media such as main memory (606), including RAM (630), cache (632), and storage systems (634), such as removable storage drives and hard disks installed in hard disk drives.

[0070] Computer programs (also called computer control logic) are stored in the system memory 606. Computer programs may also be received via a communications interface, such as a network adapter 620. When executed, such computer programs enable the computer system to perform the functions of the present embodiments as discussed herein. In particular, when executed, the computer programs enable the processing unit 604 to perform the functions of the computer system. Thus, such computer programs represent the controller of the computer system.

[0071] In one embodiment, the host (602) is a node in a cloud computing environment. As known in the art, cloud computing is a service delivery model that enables convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, network bandwidth, servers, processing, memory, storage, applications, virtual machines, and services) that can be rapidly provisioned and released with minimal administrative effort or interaction with a service provider. This cloud model may include at least five characteristics, at least three service models, and at least four deployment models. Examples of such characteristics are:

[0072] On-Demand Self-Service: Cloud consumers can unilaterally provision computing capacity, such as server time and network storage, automatically as needed, without the need for human interaction with the service provider.

[0073] Wide network access: Capabilities are available over the network and accessed through standard mechanisms that facilitate use by heterogeneous thin or thick client platforms (eg, cell phones, laptops, PDAs).

[0074] Resource Pooling: A provider's computing resources are pooled to serve multiple consumers using a multi-tenant model, with different physical and virtual resources dynamically allocated and reallocated according to demand. Consumers generally have no control or knowledge regarding the exact location of the provided resources, although there is a sense of location independence in that they may be able to specify location at a higher layer of abstraction (e.g., country, state, or data center).

[0075] Rapid Elasticity: Capacity can be provisioned quickly and elastically, sometimes automatically, quickly scaled out, quickly released, and quickly scaled in. To the consumer, the capacity available for provisioning often appears unlimited and can be purchased in any quantity at any time.

[0076] Measured Services: Cloud systems automatically control and optimize resource usage by utilizing metering capabilities at some abstraction layer appropriate to the type of service (e.g., storage, processing, bandwidth, and active user accounts). Resource usage can be monitored, controlled, and reported, providing transparency to both providers and consumers of the services used.

[0077] The service model is as follows:

[0078] Software as a Service (SaaS): The consumer is offered the ability to use a provider's applications running on a cloud infrastructure. The applications are accessible from a variety of client devices through a thin-client interface such as a web browser (e.g., web-based email). The consumer may have limited user-specific application configuration settings and does not manage or control the underlying cloud infrastructure, including networks, servers, operating systems, storage, or even individual application functions.

[0079] Platform as a Service (PaaS): The ability offered to consumers is to deploy applications they create or acquire, written using programming languages ​​and tools supported by the provider, onto a cloud infrastructure. The consumer does not manage or control the underlying cloud infrastructure, including the network, servers, operating systems, or storage, but does control the deployed applications and, in some cases, the configuration of the application hosting environment.

[0080] Infrastructure as a Service (IaaS): The ability offered to consumers is to provision processing, storage, network, and other basic computing resources on which they can deploy and run any software, which may include operating systems and applications. Consumers do not manage or control the underlying cloud infrastructure, but they do have control over the operating systems, storage, deployed applications, and in some cases, limited control over selected network components (e.g., host firewalls).

[0081] The deployment model is as follows:

[0082] Private Cloud: Cloud infrastructure is operated solely for the organization. It can be managed by the organization or a third party and can reside on-premise or off-premise.

[0083] Community Cloud: Cloud infrastructure is shared by multiple organizations to support a specific community with shared concerns (e.g., mission, security requirements, policies, compliance considerations). It may be managed by the organization or a third party and may reside on-premises or off-premises.

[0084] Public Cloud: Cloud infrastructure is made available to the general public or large industry organizations and is owned by an organization that sells cloud services.

[0085] Hybrid Cloud: A cloud infrastructure is a composition of two or more clouds (private, community, or public) that remain a unique entity but are joined by standardized or proprietary technologies (e.g., cloud bursting for load balancing between clouds) that allow for data and application portability.

[0086] Cloud computing environments are service-oriented with an emphasis on statelessness, low coupling, modularity, and semantic interoperability. At the heart of cloud computing is an infrastructure comprising a network of interconnected nodes.

[0087] Referring now to FIG. 7, a cloud computing network 700 is shown. As shown, the cloud computing network 700 includes a cloud computing environment 750 having one or more cloud computing nodes 710 with which local computing devices used by cloud consumers may communicate. Examples of these local computing devices include, but are not limited to, a personal digital assistant (PDA) or mobile phone 754A, a desktop computer 754B, a laptop computer 754C, and / or an automobile computer system 754N. Individual nodes within the cloud computing node 710 may further communicate with each other. They may be physically or virtually grouped (not shown) within one or more networks, such as a private cloud, a community cloud, a public cloud, or a hybrid cloud, or combinations thereof, as described above. This allows the cloud computing environment 700 to provide infrastructure, platforms, and / or software as a service without requiring cloud consumers to maintain resources on their local computing devices. It will be understood that the types of computing devices (754A-N) shown in FIG. 7 are intended to be illustrative only, and that the cloud computing environment (750) can communicate with any type of computerized device via any type of network and / or network-addressable connection (e.g., using a web browser).

[0088] Referring now to Figure 8, there is illustrated a set of functional abstraction layers (800) provided by the cloud computing network of Figure 7. It should be understood in advance that the components, layers, and functions illustrated in Figure 8 are intended to be exemplary only, and embodiments are not limited thereto. As shown, the following layers and corresponding functions are provided: a hardware and software layer (810), a virtualization layer (820), a management layer (830), and a workload layer (840).

[0089] The hardware and software layer (810) includes hardware and software components. Examples of hardware components include mainframes, such as the IBM® zSeries® system; servers based on RISC (reduced instruction set computer) architecture, such as the IBM pSeries® system, the IBM xSeries® system, and the IBM BladeCenter® system; storage devices; and networks and network components. Examples of software components include network application server software, such as IBM WebSphere® application server software; and database software, such as IBM DB2® database software. (IBM, zSeries, pSeries, xSeries, BladeCenter, WebSphere, and DB2 are trademarks of International Business Machines Corporation, registered in many jurisdictions worldwide.)

[0090] The virtualization layer (820) provides an abstraction layer over which the following examples of virtual entities can be provided: virtual servers; virtual storage; virtual networks, including virtual private networks; virtual applications and operating systems; and virtual clients.

[0091] In one example, the management layer (830) may provide the following functions: resource provisioning, metering and pricing, a user portal, service level management, and SLA planning and fulfillment. Resource provisioning provides dynamic procurement of computing and other resources utilized to perform tasks within the cloud computing environment. Metering and pricing provides cost tracking as resources are utilized within the cloud computing environment and billing or invoicing for the consumption of these resources. In one example, these resources may include application software licenses. Security provides identity verification for cloud consumers and tasks, as well as protection for data and other resources. The user portal provides consumers and system administrators with access to the cloud computing environment. Service level management provides allocation and management of cloud computing resources so that required service levels are met. Service level agreement (SLA) planning and fulfillment provides advance arrangements and procurement of cloud computing resources where future requirements are anticipated according to SLAs.

[0092] The workload layer (840) provides examples of functions for which a cloud computing environment can be utilized. Examples of workloads and functions that can be provided from this layer include, but are not limited to, mapping and navigation; software development and lifecycle management; virtual classroom instruction delivery; data analytics processing; transaction processing; and topic model downsampling and ANN training and use.

[0093] While particular embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that, based on the teachings herein, changes and modifications can be made without departing from the embodiments and their broader aspects. Accordingly, the appended claims are intended to encompass within their scope all such changes and modifications as fall within the true spirit and scope of the embodiments. It is further understood that the embodiments are defined solely by the appended claims. Where a specific number of introduced claim elements is intended, such intent will be expressly set forth in the claims; those skilled in the art will understand that, in the absence of such recitation, no such limitation exists. By way of non-limiting example and as an aid to understanding, the following appended claims include the use of the introductory phrases "at least one" and "one or more" to introduce claim elements. However, the use of such phrases should not be construed as meaning that the introduction of a claim element by the indefinite article "a" or "an" limits any particular claim containing such introduced claim element to embodiments containing only one such element, even if the same claim also contains the introductory phrase "one or more" or "at least one" and an indefinite article such as "a" or "an," and the same applies to the use of definite articles in a claim.

[0094] The present embodiments may be systems, methods, and / or computer program products. Moreover, selected aspects of the present embodiments may take the form of entirely hardware embodiments, entirely software (including firmware, resident software, microcode, etc.) embodiments, or embodiments combining software and / or hardware aspects, all of which may be generally referred to herein as "circuits," "modules," or "systems." Furthermore, aspects of the present embodiments may take the form of a computer program product embodied in a computer-readable storage medium (or media) having computer-readable program instructions thereon that cause a processor to perform aspects of the present embodiments. Thus, the disclosed systems, methods, and / or computer program products, as embodied therein, operate to provide improvements to transfer learning operations.

[0095] A computer-readable storage medium may be a tangible device that can hold and store instructions for use by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of computer-readable storage media includes portable computer diskettes, hard disks, dynamic or static random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), magnetic storage devices, portable compact disc read-only memory (CD-ROM), digital versatile discs (DVD), memory sticks, floppy disks, mechanically encoded devices such as punch cards or groove-embossed structures having instructions recorded thereon, and any suitable combination of the foregoing. Computer-readable storage medium and computer-readable storage device as used herein should not be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission medium (e.g., light pulses passing through a fiber optic cable), or electrical signals transmitted over electrical wires.

[0096] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device or to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network may include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in the respective computing / processing device.

[0097] The computer-readable program instructions for carrying out the operations of the present embodiments may be either assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state-setting data, or source or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and traditional procedural programming languages ​​such as the "C" programming language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partly on the user's computer, as a standalone software package, partly on the user's computer and partly on a remote computer, or entirely on a remote computer or server or cluster server. In the latter scenario, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA) may execute computer-readable program instructions by utilizing state information of the computer-readable program instructions to personalize the electronic circuitry to perform aspects of the present embodiments.

[0098] Aspects of the present embodiments are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to the embodiments. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0099] These computer-readable program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that the instructions, executed by the processor of the computer or other programmable data processing apparatus, create means for performing the functions / acts specified in one or more blocks of the flowcharts and / or block diagrams. These computer-readable program instructions may also be stored in a computer-readable storage medium that can direct a computer, programmable data processing apparatus, and / or other device to function in a particular manner, such that a computer-readable storage medium having instructions stored therein comprises a product containing instructions that implement aspects of the functions / acts specified in one or more blocks of the flowcharts and / or block diagrams.

[0100] The computer-readable program instructions may also be loaded into a computer, other programmable data processing apparatus, or other device and cause the computer, other programmable apparatus, or other device to perform a series of operational steps to generate a computer-implemented process, such that the instructions executing on the computer, other programmable apparatus, or other device implement the functions / acts specified in one or more blocks of the flowcharts and / or block diagrams.

[0101] The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of instructions, having one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions shown in the blocks may occur out of the order shown in the figures. For example, two blocks shown in succession may in fact be executed substantially simultaneously, or the blocks may be executed in the reverse order, depending on the functionality involved. Additional blocks not shown in the figures may be included, for example, before, after, or concurrently with one or more of the illustrated blocks. It should also be noted that each block of the block diagrams and / or flowchart diagrams, and combinations of blocks in the block diagrams and / or flowchart diagrams, can be implemented by a dedicated hardware-based system that performs the specified functions or actions, or executes a combination of dedicated hardware and computer instructions.

[0102] Although particular embodiments are described herein for illustrative purposes, it should be understood that various modifications may be made without departing from the spirit and scope of the embodiments. In particular, transfer learning operations may be performed by different computing platforms or across multiple devices. Furthermore, data storage and / or corpora may be localized, remote, or spread across multiple systems. Accordingly, the scope of protection of the embodiments is limited only by the following claims and their equivalents.

Claims

1. a processor operably coupled to the memory; and an artificial intelligence (AI) platform in communication with the processor and the memory; The AI ​​platform comprises: utilizing a first topic model on a first dataset, the first dataset being derived from a knowledge graph (KG) including a plurality of first triples, the first triples having first entities and first edges connecting the first entities and representing relationships between the connected first entities, the utilizing the first topic model comprising applying a first similarity function to the connected first entities of the first triples to provide respective first similarity scores; and utilizing the first topic model to select a first subset of one or more first triples from the plurality of first triples based on the first similarity score. a topic model manager configured to: a machine learning manager (MLM) configured to train an artificial neural network (ANN) using the selected first subset of the one or more first triples. have Computer system.

2. The computer system of claim 1 , wherein the first similarity function comprises a cosine similarity function.

3. The AI ​​platform is configured to receive a query including known entities and relationships, the query omitting unknown entities that have the relationship to the known entities; The topic model manager: utilizing a second topic model, which may be the same as or different from the first topic model, for a second dataset having a plurality of second triples, the second triples having second entities and second edges connecting the second entities to represent relationships between the connected second entities, the utilizing of the second topic model comprising applying a second similarity function, which may be the same as or different from the first similarity function, to the connected second entities of the second triples to provide respective second similarity scores; and selecting a second subset of one or more second triples from the plurality of second triples based on the second similarity score; further configured to perform The MLM is utilizing the trained ANN to evaluate a respective confidence score for each of the second entities of the second subset that correspond to the unknown entities of the query; and outputting at least one of the second entities of the second subset based on the confidence score. further configured to:

3. A computer system according to claim 1 or 2.

4. The computer system of claim 3 , wherein the second similarity function comprises a cosine similarity function.

5. 5. The computer system of claim 3, wherein the second data set is derived from the KG and is different from the first data set.

6. The computer system of claim 1 , wherein the ANN comprises a Bidirectional Encoder Representation of Knowledge Graphs from Transformers (KG-BERT) model.

7. 7. The computer system of claim 1, wherein the computer system is not configured to perform normalization of the first entity of the KG, and the computer system is configured to operate in an unsupervised manner.

8. The processor Utilizing a first topic model on a first dataset, the first dataset being derived from a knowledge graph (KG) having a plurality of first triples, the first triples having first entities and first edges connecting the first entities and representing relationships between the connected first entities; applying a first similarity function to the connected first entities of the first triples to provide respective first similarity scores; selecting a first subset of one or more first triples from the plurality of first triples based on the first similarity score; and training an artificial neural network (ANN) using the selected first subset of one or more first triples. A computer program for executing

9. The computer program product of claim 8 , wherein the first similarity function comprises a cosine similarity function.

10. the processor, receiving a query including known entities and relationships, the query omitting unknown entities that have the relationships to the known entities; utilizing a second topic model, which may be the same as or different from the first topic model, for a second dataset having a plurality of second triples, the second triples having second entities and second edges connecting the second entities to represent relationships between the connected second entities, the utilizing of the second topic model comprising applying a second similarity function, which may be the same as or different from the first similarity function, to the connected second entities of the second triples to provide respective second similarity scores; and selecting a second subset of one or more second triples from the plurality of second triples based on the second similarity score; and utilizing the trained ANN to evaluate a respective confidence score for each of the second entities of the second subset that correspond to the unknown entities of the query; and outputting at least one of the second entities of the second subset based on the confidence score.

10. The computer program according to claim 8 or 9, further comprising:

11. The computer program product of claim 10 , wherein the second similarity function comprises a cosine similarity function.

12. 12. The computer program of claim 10 or 11, wherein the second data set is derived from the KG and is different from the first data set.

13. 13. The computer program of claim 8, wherein the ANN comprises a Bidirectional Encoder Representation of Knowledge Graphs from Transformers (KG-BERT) model.

14. 14. The computer program of claim 8, wherein the computer program causes the processor to execute a procedure for operating unsupervised, without causing the processor to execute program code for performing normalization of the first entity of the KG.

15. utilizing a first topic model on a first dataset, the first dataset being derived from a knowledge graph (KG) having a plurality of first triples, the first triples having first entities and first edges connecting the first entities and representing relationships between the connected first entities; applying a first similarity function to the connected first entities of the first triples to provide respective first similarity scores; selecting a first subset of one or more first triples from the plurality of first triples based on the first similarity score; and training an artificial neural network (ANN) using the selected first subset of one or more first triples. A computer-implemented method comprising:

16. 16. The computer-implemented method of claim 15, wherein the first similarity function comprises a cosine similarity function.

17. receiving a query including known entities and relationships, the query omitting unknown entities having the relationships to the known entities; utilizing a second topic model, which may be the same as or different from the first topic model, for a second dataset having a plurality of second triples, the second triples having second entities and second edges connecting the second entities to represent relationships between the connected second entities, the utilizing the second topic model comprising applying a second similarity function, which may be the same as or different from the first similarity function, to the connected second entities of the second triples to provide respective second similarity scores; and selecting a second subset of one or more second triples from the plurality of second triples based on the second similarity score; and utilizing the trained ANN to evaluate a respective confidence score for each of the second entities of the second subset that correspond to the unknown entities of the query; and outputting at least one of the second entities of the second subset based on the confidence score.

17. The computer-implemented method of claim 15 or 16, further comprising:

18. 18. The computer-implemented method of claim 17, wherein the second similarity function comprises a cosine similarity function.

19. 19. The computer-implemented method of claim 17 or 18, wherein the second data set is derived from the KG and is different from the first data set.

20. 20. The computer-implemented method of any one of claims 15 to 19, wherein the ANN comprises a Bidirectional Encoder Representation of Knowledge Graphs from Transformers (KG-BERT) model.

Images