Methods and systems for graph-based cross-domain restaurant recommendation

A non-transitory computer readable medium comprising computer readable instructions which when executed, cause a processor to at least:

• access by the processor, payment transaction records from a transaction database of a payment processor, the payment transaction records related to restaurants associated with a plurality of geographically separate locations;
• determine, by the processor, from the payment transaction records a number of co-visitors shared by each of the restaurants associated with the plurality of geographically separate locations above a predetermined threshold;
• generate, by the processor, a graphical representation comprising a restaurant-restaurant based graph of the restaurants, wherein a plurality of nodes represent the restaurants and edges between the nodes represent the number of the co-visitors shared by the restaurants with the number of co-visitors above the predetermined threshold and the distance between the restaurants with co-visitors, wherein the graphical representation is generated without explicit feedback;
• transform, by the processor, the graphical representation into restaurant embeddings reflecting a mutual information metric;
• generate, by a neural network model, restaurant preferences based on the restaurant embeddings;
• computing a ranked list of the restaurants based on the restaurant preferences;
• and
• transmit, by the processor, the ranked list of the restaurants over the network to the device of the user,
• wherein the mutual information metric is used to determine restaurant similarity based on a user's behavior without relying upon a collection of user profiles or restaurant profiles, wherein the restaurant similarity is determined by either a number of common visitors between a pair of restaurants or by both the number of common visitors and a geographical distance between the pair of restaurants.

A device, comprising:

• one or more processor components;
• one or more data storage components, at least one of the one or more data storage components including instructions that when executed cause at least one of the one or more processor components to:
• access transaction records from a transaction database of a payment processor, the transaction records related to restaurants associated with a plurality of geographically separate locations;
• determine from the payment transaction records a number of co-visitors shared by each of the restaurants associated with the plurality of geographically separate locations above a predetermined threshold;
• generate a graphical representation comprising a restaurant-restaurant based graph of the restaurants, wherein a plurality of nodes represent the restaurants and edges between the nodes represent the number of the co-visitors shared by the restaurants with the number of co-visitors above the predetermined threshold and the distance between the restaurants with the co-visitors, wherein the graphical representation is generated without explicit feedback;
• and
• transform the graphical representation into restaurant embeddings reflecting a mutual information metric;
• generate, by a neural network model, restaurant preferences based on the restaurant embeddings;
• compute a ranked list of the restaurants based on the restaurant preferences;
• and
• transmit, by the computer system, the ranked list of the restaurants over the network to the device of the user,
• wherein the mutual information metric is used to determine restaurant similarity based on a user's behavior without relying upon a collection of user profiles or restaurant profiles, wherein the restaurant similarity is determined by either a number of common visitors between a pair of restaurants or by both the number of common visitors and a geographical distance between the pair of restaurants.