Abstract: In general, methods and system for modeling the incremental value of a response to different types of treatment are disclosed. Some examples include modeling discount sensitivity for specific guests. One aspect is a method for modeling incremental sales for a retail enterprise which includes generating a proxy campaign. In some embodiments, the proxy campaign is used to train an uplift model to predict an uplift score for each guest in response to a proposed campaign. In some embodiments, the uplift score is used to select guests for the proposed campaign.

Abstract: Methods and systems for analyzing and evaluating item recommendations presented on a web site are disclosed. In one aspect, a user interface is generated for display on a website. Item recommendations personalized for an individual user are displayed within in item recommendation region of the website. Impression, clickstream, and sales data are received from user activity on the website. The impression, clickstream, and sales data are displayed on a dashboard of an administrator user interface. An administrator user can select visualizations to display on the dashboard using selectors for impressions, clicks, sales, users, and time. Two or more sets of data can be overlaid to illustrate relationships between user interactions with recommended items and actual sales of those items. The dashboard is dynamically updated in response to new data received from the website in real-time.

Abstract: Multitask learning is applied to predict a customer's propensity to purchase an item within a particular category of items. Then, the network is tuned using transfer learning for a specific promotional campaign. Retail revenue and promotional revenue are jointly optimized, conditioned on customer trust. Accordingly, a particular promotional program may be selected that is specific to the user.

Abstract: Methods and systems for generating link predictions are provided. In one aspect, a method includes initializing a graph including a plurality of nodes representing selections of items in a training dataset to a multivariate normal distribution having a predetermined mean and a predetermined initial variance, the items in the training dataset comprising items in an item collection. The method includes, for each node in the graph, modeling embeddings for the node as embeddings of each neighboring node having a shared edge, with each being updated based at least in part on a transition probability and a variance. A predetermined number of iterations of updating are executed each iteration including an updated variance based on a learning rate. Based on receipt of an identification of an item from among the item collection, a plurality of predicted selections of items are identified using the embeddings for a node corresponding to the item.

Abstract: Methods and systems for analyzing and evaluating item recommendations presented on a web site are disclosed. In one aspect, a user interface is generated for display on a website. Item recommendations personalized for an individual user are displayed within in item recommendation region of the website. Impression, clickstream, and sales data are received from user activity on the website. The impression, clickstream, and sales data are displayed on a dashboard of an administrator user interface. An administrator user can select visualizations to display on the dashboard using selectors for impressions, clicks, sales, users, and time. Two or more sets of data can be overlaid to illustrate relationships between user interactions with recommended items and actual sales of those items. The dashboard is dynamically updated in response to new data received from the website in real-time.

Abstract: Methods and systems for recommending items to an unknown user of a website are disclosed. In one aspect, browsing activity of known users is analyzed to determine which items or categories of items are most popular in a given context (time, place, device, etc.). The known user browsing activity (clickstream data) is used to generate a multi-dimensional attribute matrix. Matrix factorization and clustering are used to generate affinity scores for items based on user context. These scores are used to generate item recommendations for an unknown user in a particular browsing context. In some embodiments, personalized item recommendations are updated based on interactions made between the unknown user and content of the website.

Abstract: Methods and systems for analyzing and evaluating item recommendations presented on a web site are disclosed. In one aspect, a user interface is generated for display on a website. Item recommendations personalized for an individual user are displayed within in item recommendation region of the website. Impression, clickstream, and sales data are received from user activity on the website. The impression, clickstream, and sales data are displayed on a dashboard of an administrator user interface. An administrator user can select visualizations to display on the dashboard using selectors for impressions, clicks, sales, users, and time. Two or more sets of data can be overlaid to illustrate relationships between user interactions with recommended items and actual sales of those items. The dashboard is dynamically updated in response to new data received from the website in real-time.

Abstract: Multitask learning is applied to predict a customer's propensity to purchase an item within a particular category of items. Then, the network is tuned using transfer learning for a specific promotional campaign. Retail revenue and promotional revenue are jointly optimized, conditioned on customer trust. Accordingly, a particular promotional program may be selected that is specific to the user.

Abstract: Methods and systems for generating link predictions are provided. In one aspect, a method includes initializing a graph including a plurality of nodes representing selections of items in a training dataset to a multivariate normal distribution having a predetermined mean and a predetermined initial variance, the items in the training dataset comprising items in an item collection. The method includes, for each node in the graph, modeling embeddings for the node as embeddings of each neighboring node having a shared edge, with each being updated based at least in part on a transition probability and a variance. A predetermined number of iterations of updating are executed each iteration including an updated variance based on a learning rate. Based on receipt of an identification of an item from among the item collection, a plurality of predicted selections of items are identified using the embeddings for a node corresponding to the item.

Abstract: Methods and systems for recommending items to an unknown user of a website are disclosed. In one aspect, browsing activity of known users is analyzed to determine which items or categories of items are most popular in a given context (time, place, device, etc.). The known user browsing activity (clickstream data) is used to generate a multi-dimensional attribute matrix. Matrix factorization and clustering are used to generate affinity scores for items based on user context. These scores are used to generate item recommendations for an unknown user in a particular browsing context. In some embodiments, personalized item recommendations are updated based on interactions made between the unknown user and content of the website.

Abstract: Methods and systems for analyzing and evaluating item recommendations presented on a web site are disclosed. In one aspect, a user interface is generated for display on a website. Item recommendations personalized for an individual user are displayed within in item recommendation region of the website. Impression, clickstream, and sales data are received from user activity on the website. The impression, clickstream, and sales data are displayed on a dashboard of an administrator user interface. An administrator user can select visualizations to display on the dashboard using selectors for impressions, clicks, sales, users, and time. Two or more sets of data can be overlaid to illustrate relationships between user interactions with recommended items and actual sales of those items. The dashboard is dynamically updated in response to new data received from the website in real-time.

Abstract: A system includes a subsystem interaction module device, which includes at least one interface configured to receive input signals associated with multiple components of a system. The subsystem interaction module device also includes at least one processing unit configured to identify a potential fault in one or more of the components using the input signals and to provide an indicator identifying the potential fault. The at least one processing unit is configured to identify the potential fault by: identifying conflicting frequencies that are associated with different faults in the components of the system; and determining a confidence level associated with the potential fault based on the conflicting frequencies.

Abstract: A method includes receiving one or more vibration signals representing physical vibrations generated by a shaft of a machine during operation. The method also includes determining a rotational speed of the shaft. The method further includes identifying one or more sets of frequency components of the vibration signals based on the speed of the shaft, where each set of frequency components is associated with an operational characteristic of the shaft. In addition, the method includes determining one or more health indicators of the shaft based on relative energy levels of the frequency components in each set of frequency components.

Abstract: A system includes a plurality of sensors configured to measure one or more characteristics of a gearbox. The system also includes a subsystem interaction module device, which includes at least one interface configured to receive input signals associated with multiple components of a system. The subsystem interaction module device also includes at least one processing unit configured to identify a potential fault in one or more of the components using the input signals and to provide an indicator identifying the potential fault. The at least one processing unit is configured to identify the potential fault by: identifying conflicting frequencies that are associated with different faults in the components of the system; and determining a confidence level associated with the potential fault based on the conflicting frequencies.