Multiple-entity-based recommendation system

A system for generating recommendations based on relationships between entity instances of multiple types of entities, the recommendations being between a plurality of first items and a plurality of second items, wherein the first items are instances of a first complex entity type defined at least in part by a first subset of the multiple types of entities and the second items are instances of a second complex entity type defined at least in part by a second subset of the multiple types of entities at least one of the first subset or the second subset comprising at least two of the multiple types of entities, the system comprising:

• computer memory storing representations of bipartite graphs representing the relationships between the entity instances of the multiple types of entities;
• and
• a computer processor executing instructions stored in the computer memory to perform operations for scoring pairs of a first item and a second item according to relevance of the second item to the first item, the operations comprising:
• using computational models for the bipartite graphs to compute entity vector representations of the entity instances of the types of entities of the first and second subsets, wherein each entity instance has one or more associated entity vector representations each being computed from a respective bipartite graph including the entity instance and being a unique representation of the entity instance for that bipartite graph, and wherein the first and second items each have multiple associated entity vector representations corresponding to multiple respective bipartite graphs, the multiple entity vector representations for at least one of the first item or the second item comprising entity vector representations of entity instances of at least two of the multiple types of entities;
• generating item vector representations of the first items at least in part by combining, for each first item, the associated entity vector representations of the respective entity instances of the first subset and generating item representations of the second items at least in part by combining, for each second item, the associated entity vector representations of the respective entity instances of the second subset;
• using a classifier model to compute relevance scores for pairs of a first item and a second items from the respective item vector representations;
• and
• causing an output of recommendations of the second items to the first items based on the relevance scores.

One or more non-transitory computer-readable media for generating recommendations based on relationships between entity instances of multiple types of entities, the recommendations being between a plurality of first items and a plurality of second items, wherein the first items are instances of a first complex entity type defined at least in part by a first subset of the multiple types of entities and the second items are instances of a second complex entity type defined at least in part by a second subset of the multiple types of entities, at least one of the first subset or the second subset comprising at least two of the multiple types of entities, the one or more computer-readable media storing:

• representations of bipartite graphs representing the relationships between the entity instances of the multiple types of entities;
• and
• processor-executable instructions which, when executed by one or more computer processors, cause the one or more computer processors to perform operations for scoring pairs of a first item and a second item according to relevance of the second item to the first item, the operations comprising:
• using computational models for the bipartite graphs to compute entity vector representations of the entity instances of the types of entities of the first and second subsets, wherein each entity instance has one or more associated entity vector representations each being computed from a respective bipartite graph including the entity instance and being a unique representation of the entity instance for that bipartite graph, and wherein the first and second items each have multiple associated entity vector representations corresponding to multiple respective bipartite graphs, the multiple entity vector representations for at least one of the first item or the second item comprising entity vector representations of entity instances of at least two of the multiple types of entities;
• generating item vector representations of the first items at least in part by combining, for each first item, the associated entity vector representations of the respective entity instances of the first subset and generating item representations of the second items at least in part by combining, for each second item, the associated entity vector representations of the respective entity instances of the second subset;
• using a classifier model to compute relevance scores for pairs of a first item and a second item from the respective item vector representations;
• and
• causing an output of recommendations of the second items to the first items based on the relevance scores.