Abstract: A recommender system can include a defender computing device that is configured to obtain customer interaction data characterizing customer interactions with an ecommerce marketplace. The defender computing device can also be configured to determine an item recommendation based on the customer interaction data using a trained differentially private recommendation model and send the item recommendation to the customer. The trained differentially private recommendation model is more likely to determine the same item recommendation after poisoned data is injected into the customer interaction data than a recommendation model that is not privately trained.

Abstract: This application relates to apparatus and methods for automatically detecting attacks to advertisement systems. In some examples, a computing device trains a machine learning process based on a training dataset. The training dataset may be an identified portion of a website session dataset that includes a lower percentage of malicious data caused by attacks than other portions, or may include no malicious data. Once trained, the computing device generates features from a website session dataset for a customer, and applies the trained machine learning process to the generated features to detect malicious data within the website session dataset for the customer. Further, the computing device may filter the website session data to remove the detected malicious data, and may store the filtered website session data within a data repository. The computing device may provide the filtered website session data to a recommendation system to generate item recommendations for the customer.

Abstract: Systems and methods for generating top-k recommendation using latent space representations generated by deep joint variational autoencoder processes are disclosed. A user identifier is received and a set of prior interactions associated with the user identifier is obtained. A set of latent space representations of the set of prior interactions is generated using a trained inference model. The trained inference model includes a joint variational autoencoder model. A set of k-recommended items is generated based on a comparison of the set of latent space representations of the set of prior interactions and a set of latent space representations of one or more items. A user interface including the set of k-recommended items is generated.

Abstract: Systems and methods for attribute recommendation are disclosed. Transaction data related a user is received and attribute recommendations for the user are generated based on the transaction data. The attribute recommendations are generated by a variational inference model configured using a transaction matrix and a loyalty matrix. A set of N recommendations is generated by ranking the generated attribute recommendations based on a combined transaction score and loyalty score and a user interface is generated including the set of N recommendations.

Abstract: A system including one or more processors and one or more non-transitory computer-readable media storing computing instructions configured to run on the one or more processors and perform creating a standardized flavor list from flavors associated with items in a catalog; determining a respective score indicating a degree of affinity for a user for each flavor of each of one or more item types associated with the items; creating a respective ordered list of flavors for each of the one or more item types for the user in which the respective ordered list of flavors are ranked by the respective score for the each flavor in the ordered list; and re-ranking a non-personalized list of recommended items associated with an anchor item of the items using the respective ordered list of flavors for the user. Other embodiments are disclosed.

Abstract: A system including one or more processors and one or more non-transitory computer-readable media storing computing instructions configured to run on the one or more processors and perform generating one or more item relational graphs for one or more items based on historical user purchases; transforming, using spectral filtering, the one or more item relational graphs into one or more frequency signals to remove noise from the one or more frequency signals; constructing, using a machine learning model, one or more item pair label classifications for one or more item pairs of the one or more items; generating a respective similarity score for each of the one or more item pairs; outputting a top k results for the one or more item pairs ranked by the respective similarity scores; and re-ranking, using a re-ranking algorithm, the top k results of the one or more item pairs based on a user preference for display on a user interface of an electronic device of a user. Other embodiments are disclosed.

Abstract: A privacy system includes a computing device configured to obtain user transactional data characterizing at least one transaction of a user on an ecommerce marketplace and to determine a privacy vulnerability score of the user by comparing the transactional data to a user vulnerability distribution. The computing device is also configured to send the privacy vulnerability score to a personalization engine.

Abstract: A system including one or more processors and one or more non-transitory computer-readable media storing computing instructions configured to run on the one or more processors and perform creating a standardized flavor list from flavors associated with items in a catalog; determining a respective score indicating a degree of affinity for a user for each flavor of each of one or more item types associated with the items; creating a respective ordered list of flavors for each of the one or more item types for the user in which the respective ordered list of flavors are ranked by the respective score for the each flavor in the ordered list; and re-ranking a non-personalized list of recommended items associated with an anchor item of the items using the respective ordered list of flavors for the user. Other embodiments are disclosed.

Abstract: A recommender system can include a defender computing device that is configured to obtain customer interaction data characterizing customer interactions with an ecommerce marketplace. The defender computing device can also be configured to determine an item recommendation based on the customer interaction data using a trained differentially private recommendation model and send the item recommendation to the customer. The trained differentially private recommendation model is more likely to determine the same item recommendation after poisoned data is injected into the customer interaction data than a recommendation model that is not privately trained.

Abstract: This application relates to apparatus and methods for automatically detecting attacks to advertisement systems. In some examples, a computing device trains a machine learning process based on a training dataset. The training dataset may be an identified portion of a website session dataset that includes a lower percentage of malicious data caused by attacks than other portions, or may include no malicious data. Once trained, the computing device generates features from a website session dataset for a customer, and applies the trained machine learning process to the generated features to detect malicious data within the website session dataset for the customer. Further, the computing device may filter the website session data to remove the detected malicious data, and may store the filtered website session data within a data repository. The computing device may provide the filtered website session data to a recommendation system to generate item recommendations for the customer.