Dynamic predictive similarity grouping based on vectorization of merchant data

Abstract

Provided are various mechanisms and processes for generating dynamic merchant similarity predictions. In one aspect, a system is configured for receiving historical datasets that include a series of merchants from historical browsing sessions generated by one or more users. The merchants are converted into corresponding vector representations for training a predictive model to output associated merchants based on a generated weighted vector space. Once sufficiently trained, data from a new browsing session may be received, which may include a target merchant. The target merchant is input into the predictive model as a vector to output one or more context merchants having vectors with the highest cosine similarity value to the target merchant vector. Selected context merchants may then be transmitted to the user device as targeted merchant suggestions in the new browsing session. The predictive models may be continuously trained using data received from subsequent browsing sessions.

What is claimed is: 
     
         1 . A server system comprising:
 an interface configured to receive:
 a plurality of historical datasets including a plurality of merchants corresponding to historical browsing sessions generated by one or more users; and 
 a new dataset corresponding to a new browsing session generated by a first user, wherein the new dataset includes a first merchant of the plurality of merchants; 
   memory configured to store:
 the historical datasets and 
 the new dataset; and 
   a processor associated with a neural network, wherein the processor is configured for:
 converting each merchant in the historical datasets into a corresponding vector representation; 
 inputting each corresponding vector representation into the neural network to generate weights corresponding to dimensions of the vector representations; 
 converting the first merchant in the new dataset into a corresponding first merchant vector; 
 inputting the first merchant vector into the neural network to determine a plurality of associated merchants with corresponding vector representations that include a desired probability of association with the first merchant vector based on the weights; 
 selecting a second merchant from the plurality of associated merchants based on the probability that the vector representation corresponding to the second merchant is associated with the first merchant vector; and 
 transmitting the second merchant to a first device corresponding to the first user. 
   
     
     
         2 . The server system of  claim 1 , wherein :
 the neural network is trained using skip-gram modelling of the corresponding vector representations based on the historical browsing sessions.   
     
     
         3 . The server system of  claim 1 , wherein the processor is further configured for:
 converting user data corresponding to the first user into a user vector;   inputting the user vector into the neural network to generate a probability that the user vector is associated with each vector representation in the historical datasets; and   selecting a third merchant from the plurality of merchants based on the probability that the vector representation corresponding to the third merchant is associated with the user vector.   
     
     
         4 . The server system of  claim 1 , wherein each historical dataset includes historical browsing sessions with a predetermined minimum number of merchants. 
     
     
         5 . The server system of  claim 1 , wherein the second merchant is selected based on having a corresponding vector representation with the highest cosine similarity score with the first merchant vector. 
     
     
         6 . The server system of  claim 5 , wherein a predetermined number of merchants with the highest cosine similarity scores are selected and presented in the new browsing session. 
     
     
         7 . The server system of  claim 1 ,
 wherein the vector representations are grouped based on the geographic location of the corresponding merchants in the historical datasets, and   wherein the first merchant and the second merchant are within the same geographic group.   
     
     
         8 . A programmable device configured for generating dynamic similarity predictions of merchants using a neural network,
 wherein the programmable device is configured to operate in a training mode to:
 receive a plurality of historical datasets including a plurality of merchants corresponding to historical browsing sessions generated by one or more users; 
 convert each merchant in the historical datasets into a corresponding vector representation; 
 input each corresponding vector representation into the neural network to generate weights corresponding to dimensions of the vector representations; and 
   wherein the programmable device is further configured to operate in an inference mode to:
 receive a new dataset corresponding to a new browsing session generated by a first user, wherein the new dataset includes a first merchant of the plurality of merchants; 
 convert the first merchant in the new dataset into a corresponding first merchant vector; 
 inputting the first merchant vector into the neural network to determine a plurality of associated merchants with corresponding vector representations that include a desired probability of association with the first merchant vector based on the weights; 
 select a second merchant from the plurality of associated merchants based on the probability that the vector representation corresponding to the second merchant is associated with the first merchant vector; 
 transmit the second merchant to a first device corresponding to the first user. 
   
     
     
         9 . The programmable device of  claim 8 , wherein:
 the neural network is trained using skip-gram modeling of the corresponding vector representations from the historical browsing sessions.   
     
     
         10 . The programmable device of  claim 8 , further being configured to:
 convert user data corresponding to the first user into a user vector;   input the user vector into the neural network to generate a probability that the user vector is associated with each vector representation in the historical datasets; and   select a third merchant from the plurality of merchants based on the probability that the vector representation corresponding to the third merchant is associated with the user vector.   
     
     
         11 . The programmable device of  claim 8 , wherein each historical dataset includes historical browsing sessions with a predetermined minimum number of merchants. 
     
     
         12 . The programmable device of  claim 8 , wherein the second merchant is selected based on having a corresponding vector representation with the highest cosine similarity score with the first merchant vector. 
     
     
         13 . The programmable device of  claim 12 , wherein a predetermined number of merchants with the highest cosine similarity scores are selected and presented in the new browsing session. 
     
     
         14 . The programmable device of  claim 8 ,
 wherein the vector representations are grouped based on the geographic location of the corresponding merchants in the historical datasets, and   wherein the first merchant and the second merchant are within the same geographic group.   
     
     
         15 . A method for generating dynamic similarity predictions of merchants using a neural network, the method comprising:
 in a training mode:
 receiving a plurality of historical datasets including a plurality of merchants corresponding to historical browsing sessions generated by one or more users; 
 converting each merchant in the historical datasets into a corresponding vector representation; 
 inputting each corresponding vector representation into the neural network to generate weights corresponding to dimensions of the vector representations; and 
   in an inference mode:
 receiving a new dataset corresponding to a new browsing session generated by a first user, wherein the new dataset includes a first merchant of the plurality of merchants; 
 converting the first merchant in the new dataset into a corresponding first merchant vector; 
 inputting the first merchant vector into the neural network to determine a plurality of associated merchants with corresponding vector representations that include a desired probability of association with the first merchant vector based on the weights; 
 selecting a second merchant from the plurality of associated merchants based on the probability that the vector representation corresponding to the second merchant is associated with the first merchant vector; 
 transmitting the second merchant to a first device corresponding to the first user. 
   
     
     
         16 . The method of  claim 15 , wherein:
 the neural network is trained using skip-gram modelling of the corresponding vector representations based on the historical browsing sessions.   
     
     
         17 . The method of  claim 15 , further comprising:
 converting user data corresponding to the first user into a user vector;   inputting the user vector into the neural network to generate a probability that the user vector is associated with each vector representation from the historical datasets; and   selecting a third merchant from the plurality of merchants based on the probability that the vector representation corresponding to the third merchant is associated with the user vector.   
     
     
         18 . The method of  claim 15 , wherein each historical dataset includes a historical browsing session with a predetermined minimum number of merchants. 
     
     
         19 . The method of  claim 15 , wherein the second merchant is selected based on having a corresponding vector representation with the highest cosine similarity score with the first merchant vector. 
     
     
         20 . The method of  claim 19 , wherein a predetermined number of merchants with corresponding vector representations having the highest cosine similarity scores are selected and presented in the new browsing session.