Abstract: Some techniques described herein relate to generating a hierarchical topic model (HTM), which can be used to generate custom content. In one example, a method includes determining first-level topics in a topic hierarchy related to a corpus of documents. A first-level topic of the first-level topics includes multiple words. The multiple words are grouped into clusters based on word embeddings of the multiple words. The multiple words are then subdivided into second-level topics as subtopics of the first-level topic, such that the number of second-level topics equals the number of clusters. A document of the corpus of documents is assigned to the first-level topic and to a second-level topic of the second-level topics, and an indication is received of access by a user to the document. Custom content is generated for the user based on one or more other documents assigned to the first-level topic and the second-level topic.

Abstract: Systems and methods for video processing are described. Embodiments of the present disclosure generate a plurality of image feature vectors corresponding to a plurality of frames of a video; generate a plurality of low-level event representation vectors based on the plurality of image feature vectors, wherein a number of the low-level event representation vectors is less than a number of the image feature vectors; generate a plurality of high-level event representation vectors based on the plurality of low-level event representation vectors, wherein a number of the high-level event representation vectors is less than the number of the low-level event representation vectors; and identify a plurality of high-level events occurring in the video based on the plurality of high-level event representation vectors.

Abstract: Digital image selection techniques are described that employ machine learning to select a digital image of an object from a plurality of digital images of the object. The plurality of digital images each capture the object for inclusion as part of generating digital content, e.g., a webpage, a thumbnail to represent a digital video, and so on. In one example, digital image selection techniques are described that employ machine learning to select a digital image of an object from a plurality of digital images of the object. As a result, the service provider system may select a digital image of an object from a plurality of digital images of the object that has an increased likelihood of achieving a desired outcome and may address the multitude of different ways in which an object may be presented to a user.

Abstract: This disclosure describes one or more implementations of systems, non-transitory computer-readable media, and methods that regularize learning targets for a student network by leveraging past state outputs of the student network with outputs of a teacher network to determine a retrospective knowledge distillation loss. For example, the disclosed systems utilize past outputs from a past state of a student network with outputs of a teacher network to compose student-regularized teacher outputs that regularize training targets by making the training targets similar to student outputs while preserving semantics from the teacher training targets. Additionally, the disclosed systems utilize the student-regularized teacher outputs with student outputs of the present states to generate retrospective knowledge distillation losses.

Abstract: Content creation techniques are described that leverage content analytics to provide insight and guidance as part of content creation. To do so, content features are extracted by a content analytics system from a plurality of content and used by the content analytics system as a basis to generate a content dataset. Event data is also collected by the content analytics system from an event data source. Event data describes user interaction with respective items of content, including subsequent activities in both online and physical environments. The event data is then used to generate an event dataset. An analytics user interface is then generated by the content analytics system using the content dataset and the event dataset and is usable to guide subsequent content creation and editing.

Abstract: A data classification system is trained to classify input data into multiple classes. The system is initially trained by adjusting weights within the system based on a set of training data that includes multiple tuples, each being a training instance and corresponding training label. Two training instances, one from a minority class and one from a majority class, are selected from the set of training data based on entropies for the training instances. A synthetic training instance is generated by combining the two selected training instances and a corresponding training label is generated. A tuple including the synthetic training instance and the synthetic training label is added to the set of training data, resulting in an augmented training data set. One or more such synthetic training instances can be added to the augmented training data set and the system is then re-trained on the augmented training data set.

Abstract: An image search system uses a multi-modal model to determine relevance of images to a spoken query. The multi-modal model includes a spoken language model that extracts features from spoken query and a language processing model that extract features from an image. The multi-model model determines a relevance score for the image and the spoken query based on the extracted features. The multi-modal model is trained using a curriculum approach that includes training the spoken language model using audio data. Subsequently, a training dataset comprising a plurality of spoken queries and one or more images associated with each spoken query is used to jointly train the spoken language model and an image processing model to provide a trained multi-modal model.

Abstract: In some embodiments, techniques for extracting high-value sequential patterns are provided. For example, a process may involve training a machine learning model to learn a state-action map that contains high-utility sequential patterns; extracting at least one high-utility sequential pattern from the trained machine learning model; and causing a user interface of a computing environment to be modified based on information from the at least one high-utility sequential pattern.

Abstract: Some techniques described herein relate to generating a hierarchical topic model (HTM), which can be used to generate custom content. In one example, a method includes determining first-level topics in a topic hierarchy related to a corpus of documents. A first-level topic of the first-level topics includes multiple words. The multiple words are grouped into clusters based on word embeddings of the multiple words. The multiple words are then subdivided into second-level topics as subtopics of the first-level topic, such that the number of second-level topics equals the number of clusters. A document of the corpus of documents is assigned to the first-level topic and to a second-level topic of the second-level topics, and an indication is received of access by a user to the document. Custom content is generated for the user based on one or more other documents assigned to the first-level topic and the second-level topic.

Abstract: Methods and systems are provided for generating and providing insights associated with a journey. In embodiments described herein, journey data associated with a journey is obtained. A journey can include journey paths indicating workflows through which audience members can traverse. The journey data can include audience member attributes (e.g., demographics) and labels indicating journey paths traversed by audience members. A set of audience segments are determined that describe a set of audience members traversing a particular journey path. The set of audience segments can be determined using the journey data to train a segmentation model and, thereafter, analyzing the segmentation model to identify patterns that indicate audience segments associated with the particular journey path. An indication of the set of audience segments that describe the set of audience members traversing the particular journey path can be provided for display.

Abstract: Machine-learning based multi-step engagement strategy modification is described. Rather than rely heavily on human involvement to manage content delivery over the course of a campaign, the described learning-based engagement system modifies a multi-step engagement strategy, originally created by an engagement-system user, by leveraging machine-learning models. In particular, these leveraged machine-learning models are trained using data describing user interactions with delivered content as those interactions occur over the course of the campaign. Initially, the learning-based engagement system obtains a multi-step engagement strategy created by an engagement-system user. As the multi-step engagement strategy is deployed, the learning-based engagement system randomly adjusts aspects of the sequence of deliveries for some users.

Abstract: A data classification system is trained to classify input data into multiple classes. The system is initially trained by adjusting weights within the system based on a set of training data that includes multiple tuples, each being a training instance and corresponding training label. Two training instances, one from a minority class and one from a majority class, are selected from the set of training data based on entropies for the training instances. A synthetic training instance is generated by combining the two selected training instances and a corresponding training label is generated. A tuple including the synthetic training instance and the synthetic training label is added to the set of training data, resulting in an augmented training data set. One or more such synthetic training instances can be added to the augmented training data set and the system is then re-trained on the augmented training data set.

Abstract: Systems and methods for topic modeling are described. The systems and methods include encoding words of a document using an embedding matrix to obtain word embeddings for the document. The words of the document comprise a subset of words in a vocabulary, and the embedding matrix is trained as part of a topic attention network based on a plurality of topics. The systems and methods further include encoding a topic-word distribution matrix using the embedding matrix to obtain a topic embedding matrix. The topic-word distribution matrix represents relationships between the plurality of topics and the words of the vocabulary. The systems and methods further include computing a topic context matrix based on the topic embedding matrix and the word embeddings and identifying a topic for the document based on the topic context matrix.

Abstract: Systems and methods for video processing are described. Embodiments of the present disclosure generate a plurality of image feature vectors corresponding to a plurality of frames of a video; generate a plurality of low-level event representation vectors based on the plurality of image feature vectors, wherein a number of the low-level event representation vectors is less than a number of the image feature vectors; generate a plurality of high-level event representation vectors based on the plurality of low-level event representation vectors, wherein a number of the high-level event representation vectors is less than the number of the low-level event representation vectors; and identify a plurality of high-level events occurring in the video based on the plurality of high-level event representation vectors.

Abstract: A method of finding online relevant conversing posts, comprises receiving, by a web server serving an online forum, a query post from an inquirer using the online forum, computing a contextual similarity score between each conversing post of a set of conversing posts with a query post, wherein the contextual similarity score is computed between the body of each of conversing posts and of the query post, wherein N1 conversing posts with a highest contextual similarity score are selected; computing a fine grained similarity score between the subject of the query post and of each of the N1 conversing posts, wherein N2 conversing posts with a highest fine grained similarity score are selected; and boosting the fine grained similarity score of the N2 conversing posts based on relevance metrics, wherein N3 highest ranked conversing posts are selected as a list of conversing posts most relevant to the query post.

Abstract: A technique is disclosed for generating class level rules that globally explain the behavior of a machine learning model, such as a model that has been used to solve a classification problem. Each class level rule represents a logical conditional statement that, when the statement holds true for one or more instances of a particular class, predicts that the respective instances are members of the particular class. Collectively, these rules represent the pattern followed by the machine learning model. The techniques are model agnostic, and explain model behavior in a relatively easy to understand manner by outputting a set of logical rules that can be readily parsed. Although the techniques can be applied to any number of applications, in some embodiments, the techniques are suitable for interpreting models that perform the task of classification. Other machine learning model applications can equally benefit.

Abstract: Systems, methods, and non-transitory computer-readable media are disclosed for generating combined feature embeddings for minority class upsampling in training machine learning models with imbalanced training samples. For example, the disclosed systems can select training sample values from a set of training samples and a combination ratio value from a continuous probability distribution. Additionally, the disclosed systems can generate a combined synthetic training sample value by modifying the selected training sample values using the combination ratio value and combining the modified training sample values. Moreover, the disclosed systems can generate a combined synthetic ground truth label based on the combination ratio value. In addition, the disclosed systems can utilize the combined synthetic training sample value and the combined synthetic ground truth label to generate a combined synthetic training sample and utilize the combined synthetic training sample to train a machine learning model.

Abstract: Methods, computer systems, computer-storage media, and graphical user interfaces are provided for determining user affinities by tracking historical user interactions with tagged digital content and using the user affinities in content generation applications. Accordingly, the system may track user interactions with published digital content in order to generate user interaction reports whenever a user engages with the digital content. The system may aggregate the interaction reports to generate an affinity profile for a user or audience of users. A marketer may then generate digital content for a target user or audience of users and the system may process the digital content to generate a set of tags for the digital content. Based on the set of tags, the system may then evaluate the digital content in view of the affinity profile for the target user/audience to determine similarities or differences between the digital content and the affinity profile.

Abstract: Collaborative-filtered content recommendations with justification in real-time is described. A recommendation system determines these recommendations, in part, by identifying digital content items of a catalog that are associated with a single attribute used to describe digital content. The attribute used for the identification is based on affinity scores computed for a client device user to which the recommendations are being provided. These affinity scores indicate the client device user's affinity for different attributes used to describe the digital content. Once the digital content items are identified based on the one attribute, the recommendation system is then limited to ranking and selecting from the identified digital content items to provide the recommendations. The recommendation system does not process the entire catalog of digital content items at once to rank and select the items.

Abstract: Techniques are described for machine learning-based generation of target segments is leveraged in a digital medium environment. A segment targeting system generates training data to train a machine learning model to predict strength of correlation between a set of users and a defined demographic. Further, a machine learning model is trained with visit statistics for the users to predict the likelihood that the users will visit a particular digital content platform. Those users with the highest predicted correlation with the defined demographic and the highest likelihood to visit the digital content platform can be selected and placed within a target segment, and digital content targeted to the defined demographic can be delivered to users in the target segment.