Abstract: The present disclosure is directed toward systems, methods, and computer readable media for training and utilizing an item-level importance sampling model to evaluate and execute digital content selection policies. For example, systems described herein include training and utilizing an item-level importance sampling model that accurately and efficiently predicts a performance value that indicates a probability that a target user will interact with ranked lists of digital content items provided in accordance with a target digital content selection policy. Specifically, systems described herein can perform an offline evaluation of a target policy in light of historical user interactions corresponding to a training digital content selection policy to determine item-level importance weights that account for differences in digital content item distributions between the training policy and the target policy.

Abstract: Generating and providing a content feed to a user that surfaces information items that are determined to be interesting or relevant to the user including content that is determined to be “distant” to the user is provided. Explicit user actions are used to discover peers who are not colleagues of the user (e.g., peers with whom the user does not share a close organizational relationship, peers with whom the user does not regularly communicate, etc.), but who the user indicates an interest in via his/her actions. These peers are categorized as elevated peers of the user, and information items associated with and trending around the elevated peers are surfaced to the user in a content feed.

Abstract: Systems and methods for image processing are described. One or more embodiments of the present disclosure identify an image including a plurality of objects, generate a scene graph of the image including a node representing an object and an edge representing a relationship between two of the objects, generate a node vector for the node, wherein the node vector represents semantic information of the object, generate an edge vector for the edge, wherein the edge vector represents semantic information of the relationship, generate a scene graph embedding based on the node vector and the edge vector using a graph convolutional network (GCN), and assign metadata to the image based on the scene graph embedding.

Abstract: This disclosure involves predictive modeling with entity representations computed from neural network models simultaneously trained on multiple tasks. For example, a method includes a processing device performing operations including accessing input data for an entity and transforming the input data into a dense vector entity representation representing the entity. Transforming the input data includes applying, to the input data, a neural network including simultaneously trained propensity models. Each propensity model predicts a different task based on the input data. Transforming the input data also includes extracting the dense vector entity representation from a common layer of the neural network to which the propensity models are connected.

Abstract: The present disclosure describes systems and methods for information retrieval. Embodiments of the disclosure provide a retrieval network that leverages external knowledge to provide reformulated search query suggestions, enabling more efficient network searching and information retrieval. For example, a search query from a user (e.g., a query mention of a knowledge graph entity that is included in a search query from a user) may be added to a knowledge graph as a surrogate entity via entity linking. Embedding techniques are then invoked on the updated knowledge graph (e.g., the knowledge graph that includes additional edges between surrogate entities and other entities of the original knowledge graph), and entities neighboring the surrogate entity are retrieved based on the embedding (e.g., based on a computed distance between the surrogate entity and candidate entities in the embedding space). Search results can then be ranked and displayed based on relevance to the neighboring entity.

Abstract: The disclosed techniques include at least one computer-implemented method performed by a system. The system can receive a textual query and process query features of the textual query to identify a color profile indicative of a color intent of the query. The system can identify candidate images that at least partially match the desired content and color intent of the query. The system can further order candidate images based in part on a similarity of a candidate color profile for each candidate image with the identified color profile of the query, and output image data indicative of the ordered set of candidate images.

Abstract: Systems and methods for color representation are described. Embodiments of the inventive concept are configured to receive an attribute-object pair including a first term comprising an attribute label and a second term comprising an object label, encode the attribute-object pair to produce encoded features using a neural network that orders the first term and the second term based on the attribute label and the object label, and generate a color profile for the attribute-object pair based on the encoded features, wherein the color profile is based on a compositional relationship between the first term and the second term.

Abstract: In implementations of systems for generating occurrence contexts for objects in digital content collections, a computing device implements a context system to receive context request data describing an object that is depicted with additional objects in digital images of a digital content collection. The context system generates relationship embeddings for the object and each of the additional objects using a representation learning model trained to predict relationships for objects. A relationship graph is formed for the object that includes a vertex for each relationship between the object and the additional objects indicated by the relationship embeddings. The context system clusters the vertices of the relationship graph into contextual clusters that each represent an occurrence context of the object in the digital images of the digital content collection.

Abstract: Digital image text editing techniques as implemented by an image processing system are described that support increased user interaction in the creation and editing of digital images through understanding a content creator's intent as expressed using text. In one example, a text user input is received by a text input module. The text user input describes a visual object and a visual attribute, in which the visual object specifies a visual context of the visual attribute. A feature representation generated by a text-to-feature system using a machine-learning module based on the text user input. The feature representation is passed to an image editing system to edit the digital object in the digital image, e.g., by applying a texture to an outline of the digital object within the digital image.

Abstract: The present disclosure relates applying a survival analysis to model when a particular recipient will view an electronic message. For example, one or more embodiments train a survivor function to model the time that will elapse, on a continuous scale, before a recipient will open an electronic message. For example, one or more embodiments involve accessing analytics training data and extracting a first set of features affecting the time that elapsed before past recipients opened an electronic message and a second set of features affecting whether the recipients opened the electronic message at all. The system then generates a mixture model modified survivor function and determines the effect of each feature set on its corresponding outcome to learn parameters for the mixture model modified survivor function.

Abstract: A system for analyzing risk using machine learning models may be trained using a data set to generate a risk assessment model that is optimized for metrics commonly used in for financial risk evaluation. The metrics may include Gini and CaptureRate, for example. The system may receive a request for a financial service, and generate a risk assessment by applying the risk assessment model to factors associated with the request. The system may also decide on the request in response to the risk assessment.

Abstract: In implementations of systems for generating sequential supporting answer reports, a computing device implements a report system to receive a user input defining a question with respect to a visual representation of analytics data rendered in a user interface. The report system determines a final answer to the question by processing a semantic representation of the question using a machine learning model. A sequence of reports is generated and the sequence defines an order of progression from a first supporting answer to the final answer. Each report of the sequence of reports includes a visual representation of a supporting answer to the question. The report system displays a dashboard in the user interface including a first report of the sequence of reports, the first report depicting a visual representation of the first supporting answer to the question.

Abstract: A fact replacement and style consistency tool is described. Rather than rely heavily on human involvement to replace facts and maintain consistent styles across multiple digital documents, the described change management system identifies factual and stylistic inconsistencies between these documents, in part, using natural language processing techniques. Once these inconsistencies are identified, the change management system generates a user interface that includes indications of the inconsistencies and information describing them, e.g., an indication noting not only a type of inconsistency but also presenting a first portion and at least a second portion of the multiple documents that are factually inconsistent.

Abstract: The disclosed techniques include at least one computer-implemented method performed by a system. The system can receive a textual query and process query features of the textual query to identify a color profile indicative of a color intent of the query. The system can identify candidate images that at least partially match the desired content and color intent of the query. The system can further order candidate images based in part on a similarity of a candidate color profile for each candidate image with the identified color profile of the query, and output image data indicative of the ordered set of candidate images.

Abstract: Generating and providing a content feed to a user that surfaces information items that are determined to be interesting or relevant to the user including content that is determined to be “distant” to the user is provided. Explicit user actions are used to discover peers who are not colleagues of the user (e.g., peers with whom the user does not share a close organizational relationship, peers with whom the user does not regularly communicate, etc.), but who the user indicates an interest in via his/her actions. These peers are categorized as elevated peers of the user, and information items associated with and trending around the elevated peers are surfaced to the user in a content feed.

Abstract: A sound association system identifies one or more aurally active words in digital text. Aurally active words refer to words that denote particular sounds. Context-based sounds corresponding to the one or more aurally active words are also identified. Each context-based sound is anchored to or associated with the corresponding one or more aurally active words and is played back when the digital text is played back or read, providing context-based background sounds associated with the one or more aurally active words. For example, a context-based sound can be played back at a higher volume when the one or more aurally active words are played back or read, and at a lower volume when other words of the digital text are played back or read.

Abstract: Generating and providing a content feed to a user that surfaces information items that are determined to be interesting or relevant to the user including content that is determined to be “distant” to the user is provided. Explicit user actions are used to discover peers who are not colleagues of the user (e.g., peers with whom the user does not share a close organizational relationship, peers with whom the user does not regularly communicate, etc.), but who the user indicates an interest in via his/her actions. These peers are categorized as elevated peers of the user, and information items associated with and trending around the elevated peers are surfaced to the user in a content feed.

Abstract: The present disclosure includes systems, methods, and non-transitory computer readable media that utilize a genetic framework to generate enhanced digital layouts from digital content fragments. In particular, in one or more embodiments, the disclosed systems iteratively generate a layout chromosome of digital content fragments, determine a fitness level of the layout chromosome, and mutate the layout chromosome until converging to an improved fitness level. The disclosed systems can efficiently utilize computing resources to generate a digital layout from a layout chromosome that is optimized to specified platforms, distribution audiences, and target optimization goals.

Abstract: The present disclosure includes systems, methods, and non-transitory computer readable media that utilize a genetic framework to generate enhanced digital layouts from digital content fragments. In particular, in one or more embodiments, the disclosed systems iteratively generate a layout chromosome of digital content fragments, determine a fitness level of the layout chromosome, and mutate the layout chromosome until converging to an improved fitness level. The disclosed systems can efficiently utilize computing resources to generate a digital layout from a layout chromosome that is optimized to specified platforms, distribution audiences, and target optimization goals.

Abstract: Online learning of click-through rates on search result blocks from one or more federated sources may be provided. Click-through feedback for the search result blocks may be received from the one or more federated sources in response to a query. Weights may be assigned to each of the search result blocks based on the received click-through feedback. The search result blocks may then be ranked based on the assigned weights. Finally, a search results page may be generated for displaying the ranked search results blocks to a user.