Systems and methods for use in processing unstructured data into relevant recommendations

A non-transitory computer-readable storage medium comprising executable instructions, which when executed by at least one processor, cause the at least one processor to:

• access transaction data representative of a plurality of historical transactions, the transaction data including unstructured data describing ones of the plurality of historical transactions and structured data indicative of the historical transactions, at least a portion of the historical transactions involving a user, the unstructured data including description text for product(s) involved in the ones of the plurality of historical transactions;
• compile, using a convolution neural network (CNN) to process the unstructured data from the transaction data and a recurrent neural network (RNN) to process the structured data of the transaction data, and a multi-layer perceptron (MLP), one or more feature vectors indicative of the user, by
• convolving, by the CNN, sequences of word embeddings converted from the description text for the product(s) from the structured data into first features;
• computing, by the RNN, second temporal features of the transaction data from the structured data, wherein the RNN includes a Long Short-Term Memory (LSTM) network;
• combining, by the MLP, the first features and the second temporal features into the one or more feature vectors;
• and
• generate a recommendation, based on the one or more feature vectors for the user, the recommendation including at least one action relative to one or more of the product(s);
• and
• publish the recommendation to a party associated with the user and/or the user.

A computer-implemented method for generating one or more recommendations, the method comprising:

• accessing, by a computing device, demographic data and transaction data, the transaction data including unstructured transaction data representative of a plurality of historical transactions, at least a portion of the transaction data and the demographic data involving a user;
• extracting, by the computing device, word embeddings from the unstructured data, based on one or more natural language processing techniques;
• padding, by the computing device, the extracted word embeddings to include a fixed length;
• processing, by the computing device, using a convolutional neural network (CNN), the padded word embeddings into a sequence of feature vectors to extract first features from the demographic data;
• modeling second temporal features of the transaction data using a recurrent neural network (RNN);
• combining the first features from the demographic data and the second temporal features from the transaction data, using a Multi-Layer Perceptron (MLP) approach, into a profile feature vector;
• utilizing a Reinforcement Learning approach, which includes Q-Learning, to learn user preferences based on the profile feature vector;
• generating personalized recommendations based on preferences from the Q-Learning and/or the transaction data specific to the user;
• and
• publishing the personalized recommendations to the user at a communication device associated with the user.