Creating a knowledge graph based on text-based knowledge corpora

A system comprising:

• a memory device storing a text-based knowledge corpus;
• and
• a knowledge graph generation system comprising:
• one or more processing devices configured for performing operations comprising:
• extracting a plurality of sentences from the text-based knowledge corpus;
• for each sentence of the plurality of sentences, determining relations between noun-phrases within the sentence based, at least in part, upon dependency paths of the sentence and relations predicted by a relation classifier configured to predict a relation between a pair of noun-phrases within the sentence, wherein the relation classifier is trained based on training data through semi-supervised training, wherein the training data comprises (i) a set of labeled training data that comprises subject entities and object entities and relations between the subject entities and the object entities and (ii) a set of unlabeled training data that comprises subject entities and object entities;
• generating a sub-graph for each sentence of the plurality of sentences based on (a) the noun-phrases in the sentence without using other types of phrases in the sentence and (b) the relations between the noun-phrases in the sentence, wherein the sub-graph comprises a plurality of entities representing the noun-phrases in the sentence and a plurality of edges connecting the plurality of entities, each edge of the plurality of edges representing the relation between two noun-phrases connected by the edge;
• generating a knowledge graph by merging the sub-graphs for the plurality of sentences;
• identifying one or more object entities from the knowledge graph in response to a query containing a subject entity and a relation;
• and
• a network interface device configured for transmitting the one or more object entities to an online platform, wherein the one or more object entities are usable for modifying the online platform.

A non-transitory computer-readable medium having program code that is stored thereon, the program code executable by one or more processing devices for performing operations comprising:

• extracting a plurality of sentences from a text-based knowledge corpus;
• for each sentence of the plurality of sentences, determining relations between noun-phrases within the sentence based, at least in part, upon dependency paths of the sentence and relations predicted by a relation classifier configured to predict a relation between a pair of noun-phrases within the sentence, wherein the relation classifier is trained based on training data through semi-supervised training, wherein the training data comprises (i) a set of labeled training data that comprises subject entities and object entities and relations between the subject entities and the object entities and (ii) a set of unlabeled training data that comprises subject entities and object entities;
• generating a sub-graph for each sentence of the plurality of sentences based on (a) the noun-phrases in the sentence without using other types of phrases in the sentence and (b) the relations between the noun-phrases in the sentence, wherein the sub-graph comprises a plurality of entities representing the noun-phrases in the sentence and a plurality of edges connecting the plurality of entities, each edge of the plurality of edges representing the relation between two noun-phrases connected by the edge;
• generating a knowledge graph by merging the sub-graphs for the plurality of sentences;
• identifying one or more relations from the knowledge graph in response to a query containing at least a subject entity and an object entity;
• and
• causing the one or more relations to be transmitted to an online platform, wherein the one or more relations are usable for modifying the online platform.