Abstract: A device may receive purchase data identifying purchases by users of user devices and identifying non-temporal data associated with the users, and may preprocess the purchase data to generate sequences of multivariate and multimodal symbols. The device may process the sequences of multivariate and multimodal symbols, with a long short-term memory based encoder-decoder model, to generate sequence embeddings, and may process the non-temporal data associated with the users, with a knowledge graph, to determine knowledge graph embeddings capturing the non-temporal data. The device may process the sequence embeddings and the knowledge graph embeddings, with a knowledge graph embedding model, to generate modified sequence embeddings, and may process the modified sequence embeddings, with a clustering model, to determine clusters of the users in relation to products or services purchased by the users. The device may perform one or more actions based on the clusters of the users.

Abstract: The present disclosure describes methods and systems for quantifying certainty for a prediction based on a knowledge graph. The method includes receiving a target triple and a knowledge graph comprising a set of structured data and a set of certainty scores for the structured data; converting the target triple to an embeddings space according to neighborhood sampling by a neural network, wherein the embeddings space includes a set of point coordinates; generating a plausibility prediction for the target triple using a scoring function; repeating converting the target triple to the embedding space and generating another plausibility prediction for the target triple N times with dropouts to obtain N plausibility scores for the target triple, wherein N is an integer larger than one; generating a predicted plausibility score and a certainty score for the target triple; and outputting the predicted plausibility score and the certainty score.

Abstract: A device may train a KGE model based on an initial knowledge graph, and may generate an initial embedding matrix based on training the KGE model. The device may receive a new KG representing new information, and may convert the new KG to new KG triples. The device may generate corruption data based on the new KG triples, and may generate embeddings based on the new embedding matrix and the corruption data. The device may process the embeddings, with the KGE model, to generate scores and a loss, and may regularize the embeddings and the scores/loss for seen and unseen concepts. The device may calculate a regularized loss based on regularizing the embeddings, the scores, and the loss, and may calculate an incremental learning loss based on the loss and the regularized loss. The device may train the KGE model based on the scores and the incremental learning loss.

Abstract: This application relates generally to intelligent and explainable link prediction in knowledge graph systems that automatically incorporate user feedback. In one aspect, this application discloses an iterative process for predicting a link set as a group of links in a knowledge graph in an embedding space by expanding the knowledge graph with predicted and validated single links in each iteration such that a final set of links are predicted with each one being added to the set depending on previously added predicted links. In another aspect, this application also discloses automatically extracting rules from user feedback of link predictions and generating a user feedback knowledge graph from the extracted rules, which in combination with an original knowledge graph are used for the generation of the link predictions.

Abstract: This application relates generally to intelligent and explainable link prediction in knowledge graph systems that automatically incorporate user feedback. In one aspect, this application discloses an iterative process for predicting a link set as a group of links in a knowledge graph in an embedding space by expanding the knowledge graph with predicted and validated single links in each iteration such that a final set of links are predicted with each one being added to the set depending on previously added predicted links. In another aspect, this application also discloses automatically extracting rules from user feedback of link predictions and generating a user feedback knowledge graph from the extracted rules, which in combination with an original knowledge graph are used for the generation of the link predictions.

Abstract: The present disclosure describes methods and systems for generating an approximated embedding of an out-of-knowledge-graph entity based on a knowledge graph. The method includes: receiving a target entity, a dataset associated with the target entity, and an embeddings space of a knowledge graph comprising a set of structured data, wherein the target entity is out of the knowledge graph and the embeddings space includes a set of vectors representing the set of structured data in the embeddings space; selecting a set of elements from the knowledge graph, each element being related to the target entity according to the dataset associated with the target entity; constructing a set of descriptory triples based on the target entity and the set of elements; obtaining an embedding matrix based on the descriptory triples and the embeddings space; and generating an approximated embedding for the target entity based on the embedding matrix.

Abstract: The present disclosure describes methods and systems to predict predicate metadata parameters in knowledge graphs via neural networks. The method includes receiving a knowledge graph based on a knowledge base including a graph-based dataset. The knowledge graph includes a predicate between two nodes and a set of predicate metadata.

Abstract: This disclosure provides methods and systems for predicting missing links and previously unknown numerals in a knowledge graph. A jointly trained multi-task machine learning model is disclosed for integrating a symbolic pipeline for predicting missing links and a regression numerical pipeline for predicting numerals with prediction uncertainty. The two prediction pipelines share a jointly trained embedding space of entities and relationships of the knowledge graph. The numerical pipeline additionally includes a second-layer multi-task regression neural network containing multiple regression neural networks for parallel numerical prediction tasks with a cross stich network allowing for information/model parameter sharing between the various parallel numerical prediction tasks.

Abstract: The present disclosure describes methods and systems for generating an explanation for a prediction based on a knowledge graph. The method includes receiving a target triple and a knowledge graph including a set of structured data; converting the knowledge graph to an embeddings space and outputting a plausibility prediction for the target triple; sampling a set of neighbors of elements of the target triple in the embeddings space; obtaining a set of example triples based on the set of neighbors according to the target triple; obtaining a prototype graph based on the set of the example triples according to the target triple; generating an explanation graph based on the prototype graph, the set of example triples, and the target triple; and generating an explanation for the plausibility prediction based on the explanation graph.

Abstract: A device may receive purchase data identifying purchases by users of user devices and identifying non-temporal data associated with the users, and may preprocess the purchase data to generate sequences of multivariate and multimodal symbols. The device may process the sequences of multivariate and multimodal symbols, with a long short-term memory based encoder-decoder model, to generate sequence embeddings, and may process the non-temporal data associated with the users, with a knowledge graph, to determine knowledge graph embeddings capturing the non-temporal data. The device may process the sequence embeddings and the knowledge graph embeddings, with a knowledge graph embedding model, to generate modified sequence embeddings, and may process the modified sequence embeddings, with a clustering model, to determine clusters of the users in relation to products or services purchased by the users. The device may perform one or more actions based on the clusters of the users.

Abstract: A hypothesis generation system may determine sets of link types that are respectively associated with a plurality of nodes included in an incomplete knowledge graph to determine a plurality of intersection-over-union scores. The hypothesis generation system may determine, based on a plurality of vectors of an embedding space representation associated with the incomplete knowledge graph, a plurality of similarity scores and may determine, based on the plurality of intersection-over-union scores and the plurality of similarity scores, a plurality of affinity scores. The hypothesis generation system may determine, based on the plurality of affinity scores and the plurality of nodes, one or more node pairs; may generate, for a node pair, of the one or more node pairs, one or more triplet hypothesis candidate templates; and may generate, for a triplet hypothesis candidate template, of the one or more triplet hypothesis candidate templates, a plurality of triplet hypothesis candidates.

Abstract: This disclosure provides methods and systems for predicting missing links and previously unknown numerals in a knowledge graph. A jointly trained multi-task machine learning model is disclosed for integrating a symbolic pipeline for predicting missing links and a regression numerical pipeline for predicting numerals with prediction uncertainty. The two prediction pipelines share a jointly trained embedding space of entities and relationships of the knowledge graph. The numerical pipeline additionally includes a second-layer multi-task regression neural network containing multiple regression neural networks for parallel numerical prediction tasks with a cross stich network allowing for information/model parameter sharing between the various parallel numerical prediction tasks.

Abstract: The present disclosure describes methods and systems to predict predicate metadata parameters in knowledge graphs via neural networks. The method includes receiving a knowledge graph based on a knowledge base including a graph-based dataset. The knowledge graph includes a predicate between two nodes and a set of predicate metadata.

Abstract: Systems and methods enable the discovery of new relationships between diseases and genes by prioritizing the selection of gene targets for a disease using an embedding space generated from a knowledge graph by mapping datasets collected from various data sources using a graph schema, modeling disease and gene associations with link weightings, analyzing the data with several machine learning models, and scoring predictions.

Abstract: The systems and methods described herein may generate multi-modal embeddings with sub-symbolic features and symbolic features. The sub-symbolic embeddings may be generated with computer vision processing. The symbolic features may include mathematical representations of image content, which are enriched with information from background knowledge sources. The system may aggregate the sub-symbolic and symbolic features using aggregation techniques such as concatenation, averaging, summing, and/or maxing. The multi-modal embeddings may be included in a multi-modal embedding model and trained via supervised learning. Once the multi-modal embeddings are trained, the system may generate inferences based on linear algebra operations involving the multi-modal embeddings that are relevant to an inference response to the natural language question and input image.

Abstract: The systems and methods described herein may generate multi-modal embeddings with sub-symbolic features and symbolic features. The sub-symbolic embeddings may be generated with computer vision processing. The symbolic features may include mathematical representations of image content, which are enriched with information from background knowledge sources. The system may aggregate the sub-symbolic and symbolic features using aggregation techniques such as concatenation, averaging, summing, and/or maxing. The multi-modal embeddings may be included in a multi-modal embedding model and trained via supervised learning. Once the multi-modal embeddings are trained, the system may generate inferences based on linear algebra operations involving the multi-modal embeddings that are relevant to an inference response to the natural language question and input image.