Device and method for machine learning

A device for grouping target entities into clusters and storing the target entities into a knowledge graph, the device comprising:

• a data memory for associations and inference rules;
• a grouping device configured to determine a base association in which a cluster is associated with each of the target entities, as a function of the target entities and as a function of an association for entities;
• a learning device configured to determine inference rules as a function of the association for entities and as a function of the base association, each of the inference rules defining an association of entities with one of the clusters, the learning device being configured to determine an altered association as a function of the association for entities and the inference rules, wherein the base association includes subjects and objects as entities, the subjects representing the target entities, the objects representing clusters, the association defining predicates, each of the predicates associating an object with a subject, each of the inference rules including a trunk and a head that is associated with the trunk, wherein the head is a triplet of form h(X, Y), where h defines a predicate, X defines a subject, and Y defines an object, the trunk being defined as a function of at least one predicate or as a function of a conjunction of least two predicates, the head defining, for a subject that represents one of the target entities, with which of the objects, which represent the clusters, the subject is associated;
• and
• a feedback device configured to check whether a difference between the base association and the altered association falls below a threshold value, and if the difference falls below the threshold value, to output or store an association of the target entities with the clusters, and that otherwise determines a feedback value as a function of the difference, and determines a replacement association that replaces the association, as a function of the association for entities and the feedback value.

A non-transitory machine-readable memory medium on which is stored a computer program for grouping target entities into clusters and storing the target entities into a knowledge graph, the computer program, when executed by a computer, causing the computer to perform:

• in a computation step for the target entities, determining a base association in which a cluster is associated with each of the target entities as a function of an association for entities;
• determining inference rules as a function of the association for entities and as a function of the base association, each of the inference rules defining an association of entities with one of the clusters, wherein the base association includes subjects and objects as entities, the subjects representing the target entities, the objects representing clusters, the association defining predicates, each of the predicates associating an object with a subject, each of the inference rules including a trunk and a head that is associated with the trunk, wherein the head is a triplet of form h(X, Y), where h defines a predicate, X defines a subject, and Y defines an object, the trunk being defined as a function of at least one predicate or as a function of a conjunction of least two predicates, the head defining, for a subject that represents one of the target entities, with which of the objects, which represent the clusters, the subject is associated;
• determining an altered association as a function of the association for entities and the inference rules;
• checking whether a difference between the base association and the altered association falls below a threshold value;
• based on the difference falling below the threshold value, outputting or storing an association of the target entities with the clusters;
• and
• based on the different not falling below the threshold, (i) determining a feedback value as a function of the difference, and (ii) determining a replacement association as a function of the association for entities and the feedback value, which replaces the association for a new execution of the computation step.