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: Unstructured texts associated with a product is received, where the unstructured texts include, for example, a title of the product, one or more reviews of the product, questions and/or answers associated with the product. A phrase in an unstructured text is identified. A first knowledge base is searched, to identify that the phrase is a feature value that is associated with a feature. For example, the first knowledge base lists the feature value to be an instance of the feature. Accordingly, a tuple is generated, where the tuple includes the product as a subject, the feature as a predicate, and the feature value comprising the phrase as an object. A second knowledge base is updated with the tuple. The second knowledge base is usable for processing queries about the product. For example, the second knowledge base is used to generate a result of a query about the product.

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: Methods and systems are provided for identifying subparts of a text. A neural network system can receive a set of sentences that includes context sentences and target sentences that indicate a decision point in a text. The neural network system can generate context vector sentences and target sentence vectors by encoding context from the set of sentences. These context sentence vectors can be weighted to focus on relevant information. The weighted context sentence vectors and the target sentence vectors can then be used to output a label for the decision point in the text.

Abstract: Network-based probabilistic device linking techniques are described that link multiple devices associated with a common entity. In one example, log records are received from service providers including a device identifier and an IP address associated with a computing device that uses the service providers to access resources. The received log records are filtered and analyzed to identify connection frequencies between each device identifier and various IP addresses. Connection frequencies are scored and used to identify a subset of connections for computing linked devices belonging to a common entity, such as a single user, a household of users, users in a specific location, and so on. Linked devices are computed from the subset of selected connections and combined into linked device clusters. These linked device clusters can then be output so that market analysis can be performed on the linked device cluster rather than data pertaining to a single device.

Abstract: An analytics server receives data characterizing consumer interactions that are observed by a cross-section of data providers, which may include, for example, website administrators, campaign managers, application developers, and the like. Such observational data includes device and login identifiers for a particular interaction, and optionally, timestamp information indicating when the interaction occurred. A statistical device graph model is generated based on this observational data. The statistical device graph model allows inferences to be drawn with respect to whether a given device is a private device, a shared device, or a public device. This, in turn, allows private devices which are “owned” by a single consumer to be identified. Depending on the type of observational data collected by the data providers, a wide range of additional insights can be drawn from the statistical device graph model, including for example, device usage patterns and confidence levels.

Abstract: In some embodiments, a content analysis system accesses input content associated with a user of an online platform. The content analysis system extracts entity tags for entities contained in the input content and links the identities to standard entities in a knowledge base to generate linked entities. The content analysis system further generates a knowledge graph to include the linked entities and other standard entities connected to the linked entities as nodes and edges connecting these nodes. Based on the knowledge graph, the content analysis system identifies related entities that are similar to the linked entities and cause the online platform to be modified based on the related entities.

Abstract: The present disclosure relates to systems, methods, and non-transitory computer readable media for generating coachmarks and concise instructions based on operation descriptions for performing application operations. For example, the disclosed systems can utilize a multi-task summarization neural network to analyze an operation description and generate a coachmark and a concise instruction corresponding to the operation description. In addition, the disclosed systems can provide a coachmark and a concise instruction for display within a user interface to, directly within a client application, guide a user to perform an operation by interacting with a particular user interface element.

Abstract: Systems and methods are disclosed for clustering multiple devices that are associated with particular users by utilizing both probabilistic and deterministic data derived from analytics information on the users. An analytics computing system generates at least one deterministic device cluster that groups a first set of devices associated with a first user. The first set of devices share deterministic user identifiers specific to the first user. The analytics computing system also identifies a probabilistic link between a device in the first set of devices and additional devices. The probabilistic link indicates common usage patterns between two devices. Based on the probabilistic link, the analytics computing system generates a data structure that includes the deterministic device cluster and the additional devices.

Abstract: Sales Forecasting using Browsing Ratios and Browsing Durations is described. In one or more implementations, browsing ratios representative of how much users visited webpages associated with a product or service, and browsing durations representing how much time users spent visiting the webpages associated with the product or service are determined. Based on the determined browsing ratios and browsing durations, a sales forecast of the product or service can be accurately determined.

Abstract: The present disclosure relates to utilizing a neural network to flexibly generate label distributions for modifying a segment of text to emphasize one or more words that accurately communicate the meaning of the segment of text. For example, the disclosed systems can utilize a neural network having a long short-term memory neural network architecture to analyze a segment of text and generate a plurality of label distributions corresponding to the words included therein. The label distribution for a given word can include probabilities across a plurality of labels from a text emphasis labeling scheme where a given probability represents the degree to which the corresponding label describes the word. The disclosed systems can modify the segment of text to emphasize one or more of the included words based on the generated label distributions.

Abstract: A method for clustering geolocations using geo-point density includes receiving a user log of geolocation data extracted from user interactions with at least one electronic device. A density is determined relative to other geo-points for each geo-point in a set of geo-points extracted from the user log. Lower density geo-points in the set are merged into higher density geo-points in the set to result in a merged set of geo-points, and clusters of geo-points are identified from the merged set. Merging the geo-points tends to preserve frequently occurring geo-points while reducing those that constitute noise, which improves the reliability of identifying the clusters. Core geo-points of the user log are selected from the clusters. The core geo-points of the user log can be compared to core geo-points of other use logs to identify associations between the user logs.

Abstract: Various embodiments describe a segmentation application that uses a predictive model to segment content from instruction manuals. In an example, the segmentation application receives training data including training labels and steps available from instruction manuals. The segmentation application trains a predictive model based on the training data and a loss function. The training includes generating, by the predictive model, a prediction of whether the sub-step is the start of the step and minimizing the loss function based on comparison of the prediction to the training label. Upon completion of the training, the segmentation application identifies electronic sources and a start indicator indicating a start of a specific step. The segmentation application provides, in response to a query about an object from a client device, a step generated from the start indicator.

Abstract: Methods and systems are described that automatically organize directory hierarchies and label individual directories systematically. Upon a number of files in a first directory exceeding a maximum number of files, a second directory is created. The files formerly disposed only in the first directory are organized into both of the first directory and the second directory so that the threshold number of files is not exceeded in either of the first or second directories. Organizing the files into the first and second directories uses vector representations of each of the files generated by the system so that, when organized, the first and second directories each include files with similar content. Labels are selected for each of the directories based on a comparison between a vector representation of the collective contents of each directory and vector representations of titles in a database.

Abstract: Methods and systems are provided for identifying subparts of a text. A neural network system can receive a set of sentences that includes context sentences and target sentences that indicate a decision point in a text. The neural network system can generate context vector sentences and target sentence vectors by encoding context from the set of sentences. These context sentence vectors can be weighted to focus on relevant information. The weighted context sentence vectors and the target sentence vectors can then be used to output a label for the decision point in the text.

Abstract: The present disclosure relates to systems, non-transitory computer-readable media, and methods that utilize recurrent neural networks to determine the existence of one or more open intents in a text input, and then extract the one or more open intents from the text input. In particular, in one or more embodiments, the disclosed systems utilize a trained intent existence neural network to determine the existence of an actionable intent within a text input. In response to verifying the existence of an actionable intent, the disclosed systems can apply a trained intent extraction neural network to extract the actionable intent from the text input. Furthermore, in one or more embodiments, the disclosed systems can generate a digital response based on the intent identified from the text input.

Abstract: Embodiments of the present invention relate to identifying website visitors. Initially, a predictor is trained with a set of data of known website visitors to identify a rule with the highest effectiveness score. To do so, each rule in a set of rules is applied to all cookies in the set of data. Based on a selected goal of identifying unknown website visitors, the rule with the highest effectiveness score is identified. To identify a cookie of an unknown website visitor, a cookie representation corresponding to the cookie is identified. The cookie representation represents the cookie as an n-dimensional vector of features and is computed using hit statistics for various aspects of the cookie. Utilizing the cookie representation, a cookie-stitching rule is selected and applied to the cookie. In this way, a website visitor associated with the cookie can be identified.

Abstract: Various embodiments describe a segmentation application that uses a predictive model to segment content from instruction manuals. In an example, the segmentation application receives training data including training labels and steps available from instruction manuals. The segmentation application trains a predictive model based on the training data and a loss function. The training includes generating, by the predictive model, a prediction of whether the sub-step is the start of the step and minimizing the loss function based on comparison of the prediction to the training label. Upon completion of the training, the segmentation application identifies electronic sources and a start indicator indicating a start of a specific step. The segmentation application provides, in response to a query about an object from a client device, a step generated from the start indicator.

Abstract: Systems and methods are disclosed for clustering multiple devices that are associated with particular users by utilizing both probabilistic and deterministic data derived from analytics information on the users. An analytics computing system generates at least one deterministic device cluster that groups a first set of devices associated with a first user. The first set of devices share deterministic user identifiers specific to the first user. The analytics computing system also identifies a probabilistic link between a device in the first set of devices and additional devices. The probabilistic link indicates common usage patterns between two devices. Based on the probabilistic link, the analytics computing system generates a data structure that includes the deterministic device cluster and the additional devices.

Abstract: A method for clustering geolocations using geo-point density includes receiving a user log of geolocation data extracted from user interactions with at least one electronic device. A density is determined relative to other geo-points for each geo-point in a set of geo-points extracted from the user log. Lower density geo-points in the set are merged into higher density geo-points in the set to result in a merged set of geo-points, and clusters of geo-points are identified from the merged set. Merging the geo-points tends to preserve frequently occurring geo-points while reducing those that constitute noise, which improves the reliability of identifying the clusters. Core geo-points of the user log are selected from the clusters. The core geo-points of the user log can be compared to core geo-points of other use logs to identify associations between the user logs.