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 providing a unified serving platform that allows for the reusability of machine learning models across a plurality of websites to determine personalized content. For example, a computing device trains a machine learning model with session data identifying browsing events and transaction data identifying purchasing events for a plurality of users. The computing device receives and stores session data and transaction data associated with a first website for the customer. The computing device may then receive a request for content to display to the customer on a second website. The computing device generates label data based on the session data and transaction data associated with the first website, and executes the trained machine learning model with the label data. Based on execution of the trained machine learning model, the computing device generates content to display on the second website, and transmits the content.

Abstract: A method including training a logistic regression model to create a trained model to provide probabilities of users clicking on emails of one or more email campaigns within each of multiple different time periods. Input predictor variables of the logistic regression model include (i) user feature data including personal user features and online activity history for users in the multiple different time periods and (ii) email feature data including sent times and item category data for multiple different emails in the one or more email campaigns. Output dependent variables of the logistic regression model include responses by the users to the one or more email campaigns. The method also includes triggering sending a first email of the one or more email campaigns to a first user of the users at a selected time period based at least in part on the trained model. Other embodiments are disclosed.

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: A system includes a computing device configured to receive order data indicating an order placed by a customer on an e-commerce platform and route the order into a test group or a control group when the order data indicates that the order will be filled by a store participating in the comparison test. The computing device is further configured to apply test features to the order if the order was routed into the test group and apply control features to the order if the order was routed into the control group and determine recommended substitute items based on the test features or the control features. The recommended substitute items are intended to replace items ordered by the customer that are unavailable. The computing device is also configured to determine one or more performance metrics of the test group and the control group.

Abstract: In some implementations, a provider edge device associated with a link aggregation group (LAG) may maintain, according to a link aggregation control protocol (LACP), a set of links that connect the PE device to a consumer edge device. The provider edge device may determine that the provider edge device and another provider edge device associated with the LAG are not receiving link aggregation control protocol data units (LACPDUs) from the consumer edge device. The provider edge device may cause the set of links to have a maintain LAG status, which causes the provider edge device to keep up the set of links and to cease maintaining the set of links according to the LACP. The provider edge device may route, based on causing the set of links to have the maintain LAG status, one or more packets to or from the consumer edge device via the set of links.

Abstract: Systems and methods including one or more processors and one or more non-transitory storage devices storing computing instructions configured to run on the one or more processors and perform: determining a list of substitutes for an item; determining qualities for each substitute; determining a similarity score for each substitute; determining a historical substitution score for each substitute; determining a final score for each substitute using the similarity score for each substitute and the historical substitution score for each substitute; ranking each substitute based upon the final score for each substitute; facilitating a display, on a user interface of a user device, of a highest ranked substitute; receiving, from the user interface of the user device, a selection of the highest ranked substitute; and after receiving the selection of the highest ranked substitute, substituting the highest ranked substitute for the item of the list of items. Other embodiments are disclosed herein.

Abstract: Systems and methods for generating a set of personalized complementary recommendations is disclosed. A user identifier and an anchor item identifier are received. A set of personalized-weighted items and a set of complimentary-weighted items are each generated based on the user identifier and the anchor item identifier. The personalized-weighted items are generated by a trained supervised model. The complementary-weighted items are generated by a trained unsupervised model. A set of personalized complementary recommendations including a subset of the personalized-weighted items and a subset of the complementary-weighted items is generated.

Abstract: A system can include one or more processing modules and one or more non-transitory computer-readable media storing computing instructions configured to run on the one or more processing modules and perform receiving, from an electronic device, a search query from a user of a plurality of users; processing first data; and facilitating displaying a set of items. Processing the first data can comprise determining one or more keywords by capturing the one or more keywords during a time window; creating a feature set of second data associated with at least a portion of the plurality of users; determining a set of items of the item set as being based at least in part on an item vector representation and a keyword vector representation; determining a respective purchase probability associated with each item of the set of items of the item set; ranking the set of items.

Abstract: Techniques are described for managing a split-brain scenario in a multihomed environment by exchanging isolation information between a leaf device and two or more spine devices to which the leaf device is multihomed via a link aggregation group (LAG). The techniques include selecting one of the spine devices as a primary spine device and determining, based on the isolation information, whether the spine devices are isolated from each other. In the split-brain scenario in which all of the spine devices are isolated from each other, the primary spine device is configured to maintain the LAG with the leaf device while the other spine devices mark the LAG with the leaf device as down. In this way, in the split-brain scenario, the leaf device may continue to send traffic to other leaf devices in the leaf layer using the LAG to the primary spine device.

Abstract: An approach is disclosed for recommending complementary items based on customer shopping routines. The approach receives anchor item data. The approach identifies a routine that corresponds to the anchor item data. The routine is based on an item purchasing behavior of a customer. The approach determines categorical data within the identified routine by applying a ranking algorithm to the categorical data of the categories and the anchor item data. The categorical data is relevant to the anchor item data. The approach generates relevant item data from the categorical data by applying an item recommendation model to item data that corresponds to the categorical data.

Abstract: This invention is directed to compositions and methods for treating cancer in a patient.

Abstract: A method including tracking first data including responses by users to email campaigns. The method also can include generating a user feature set for the users, the user feature set including personal features of each of the users and online activity history for each of the users across a first set of time periods. The method additionally can include generating an email feature set for the emails in the email campaigns. The method further can include training a logistic regression model to create a trained model using the first data, the user feature set, and the email feature set to provide a probability of a user of the users clicking on an email of the emails within each time period of a second set of time periods. The method additionally can include triggering sending a first email of the email campaigns to a first user of the users at a selected time period of the second set of time periods based at least in part on the trained model. Other embodiments are disclosed.

Abstract: A network device may receive, from an endpoint device, a first message that includes first endpoint identification information. The network device may be connected to the endpoint device via a plurality of links. The network device may receive, from another network device, a second message that includes second endpoint identification information. The network device may determine whether the first endpoint identification information corresponds to the second endpoint identification information. The network device may cause, based on determining whether the first endpoint identification information corresponds to the second endpoint identification information, a state of the plurality of links to be maintained or changed.

Abstract: A system can include one or more processing modules and one or more non-transitory computer-readable media storing computing instructions configured to run on the one or more processing modules and perform receiving, from an electronic device, a search query from a user of a plurality of users; processing first data; and facilitating displaying a set of items. Processing the first data can comprise determining one or more keywords by capturing the one or more keywords during a time window; creating a feature set of second data associated with at least a portion of the plurality of users; determining a set of items of the item set as being based at least in part on an item vector representation and a keyword vector representation; determining a respective purchase probability associated with each item of the set of items of the item set; ranking the set of items.

Abstract: This invention is directed to compositions, methods and kits that can be used for reducing cell viability, inducing cell apoptosis, and inhibiting cell proliferation, such as for the treatment of cancer. The invention is also directed to compositions, methods and kits that can be used for treating inflammatory conditions, such as by modulating inflammatory cytokines.

Abstract: This application relates to apparatus and methods for automatically identifying substitute items. A computing device can generate matrix data that identifies connection values between a plurality of items. The matrix data may be generated based on the application of one or more machine learning algorithms to historical data identifying accepted or denied item substitutions. The computing device may then receive item data identifying at least one second item and at least one attribute of that second item. The computing device may generate a graph based on the matrix data and the item data to determine connection values between the second item and the plurality of first items. The computing device may then determine a substitute item (e.g., a replacement item) for the second item based on the connection values between the second item and the plurality of first items.

Abstract: A network device may receive, from an endpoint device, a first message that includes first endpoint identification information. The network device may be connected to the endpoint device via a plurality of links. The network device may receive, from another network device, a second message that includes second endpoint identification information. The network device may determine whether the first endpoint identification information corresponds to the second endpoint identification information. The network device may cause, based on determining whether the first endpoint identification information corresponds to the second endpoint identification information, a state of the plurality of links to be maintained or changed.

Abstract: A system includes a computing device configured to obtain item attribute data that corresponds to a characteristic of an item ordered by a customer on an e-commerce platform and a common characteristic of a plurality of substitution items. The computing device is also configured to obtain customer attribute data identifying preferences of the customer and to determine a preference score for each substitution item in the plurality of substitution items based on the item attribute data and the customer attribute data. The preference score indicates a likelihood that the customer will accept one of the plurality of substitution items as a replacement for the item ordered by the customer. The computing device is also configured to rank each substitution item in the plurality of substitution items based on the preference scores.

Abstract: A system includes a computing device configured to receive order data indicating an order placed by a customer on an e-commerce platform and route the order into a test group or a control group when the order data indicates that the order will be filled by a store participating in the comparison test. The computing device is further configured to apply test features to the order if the order was routed into the test group and apply control features to the order if the order was routed into the control group and determine recommended substitute items based on the test features or the control features. The recommended substitute items are intended to replace items ordered by the customer that are unavailable. The computing device is also configured to determine one or more performance metrics of the test group and the control group.