Abstract: A messaging system performs trend analysis on content produced by users of the messaging system. The messaging system is configured to extract modifications from content items received from client devices associated with users where the content items are modified using the modifications that comprises a text caption or a media overlay. The messaging system is further configured to determine one or more words from the content items and the extracted modifications and determine a frequency of the one or more words in the content items and the extracted modifications. The messaging system is further configured to determine whether the one or more words is a trend based on the frequency and an aggregate frequency. The messaging system is further configured to in response to the one or more words being determined as the trend, generating trend content associated with the one or more words, the trend content being a text, an image, or an augmentation content.

Abstract: A messaging system performs engagement analysis based on labels associated with content items produced by users of the messaging system. The messaging system is configured to process content items comprising images to identify elements in the images and determine labels for the images based on conditions indicating when to associate a label of the labels with an image of the images based on the elements in the image. The messaging system is further configured to associate the label with the content item, in response to determining to associate the label with the image, associating the label with the content item. The messaging system is further configured to determine engagement scores for the label based on interactions of users with the content items associated with label and adjust the engagement scores to determine trends in the labels to generate adjusted engagement scores.

Abstract: A caption of a multimodal message (e.g., social media post) can be identified as a named entity using an entity recognition system. The entity recognition system can use an attention-based mechanism that emphasis or de-emphasizes each data type (e.g., image, word, character) in the multimodal message based on each datatypes relevance. The output of the attention mechanism can be used to update a recurrent network to identify one or more words in the caption as being a named entity.

Abstract: Sentiment classification can be implemented by an entity-level multimodal sentiment classification neural network. The neural network can include left, right, and target entity subnetworks. The neural network can further include an image network that generates representation data that is combined and weighted with data output by the left, right, and target entity subnetworks to output a sentiment classification for an entity included in a network post.

Abstract: Sentiment classification can be implemented by an entity-level multimodal sentiment classification neural network. The neural network can include left, right, and target entity subnetworks. The neural network can further include an image network that generates representation data that is combined and weighted with data output by the left, right, and target entity subnetworks to output a sentiment classification for an entity included in a network post.

Abstract: A caption of a multimodal message (e.g., social media post) can be identified as a named entity using an entity recognition system. The entity recognition system can use a visual attention based mechanism to generate a visual context representation from an image and caption. The system can use the visual context representation to identify one or more terms of the caption as a named entity.

Abstract: A messaging system performs engagement analysis based on labels associated with content items produced by users of the messaging system. The messaging system is configured to process content items comprising images to identify elements in the images and determine labels for the images based on conditions indicating when to associate a label of the labels with an image of the images based on the elements in the image. The messaging system is further configured to associate the label with the content item, in response to determining to associate the label with the image, associating the label with the content item. The messaging system is further configured to determine engagement scores for the label based on interactions of users with the content items associated with label and adjust the engagement scores to determine trends in the labels to generate adjusted engagement scores.

Abstract: Systems, devices, media, and methods are presented for generating a language detection model of a language analysis system. The systems and methods access a set of messages including text elements and convert the set of messages into a set of training messages. The set of training messages are configured for training a language detection model. The systems and methods train a classifier based on the set of training messages. The classifier has a set of features representing word frequency, character frequency, and a character ratio. The systems and methods generate a language detection model based on the classifier and the set of features.

Abstract: A caption of a multimodal message (e.g., social media post) can be identified as a named entity using an entity recognition system. The entity recognition system can use an attention-based mechanism that emphasis or de-emphasizes each data type (e.g., image, word, character) in the multimodal message based on each datatypes relevance. The output of the attention mechanism can be used to update a recurrent network to identify one or more words in the caption as being a named entity.

Abstract: Systems, devices, media, and methods are presented for generating a language detection model of a language analysis system. The systems and methods access a set of messages including text elements and convert the set of messages into a set of training messages. The set of training messages are configured for training a language detection model. The systems and methods train a classifier based on the set of training messages. The classifier has a set of features representing word frequency, character frequency, and a character ratio. The systems and methods generate a language detection model based on the classifier and the set of features.

Abstract: A caption of a multimodal message (e.g., social media post) can be identified as a named entity using an entity recognition system. The entity recognition system can use a visual attention based mechanism to generate a visual context representation from an image and caption. The system can use the visual context representation to identify one or more terms of the caption as a named entity.

Abstract: Sentiment classification can be implemented by an entity-level multimodal sentiment classification neural network. The neural network can include left, right, and target entity subnetworks. The neural network can further include an image network that generates representation data that is combined and weighted with data output by the left, right, and target entity subnetworks to output a sentiment classification for an entity included in a network post.

Abstract: A messaging system performs trend analysis on content produced by users of the messaging system. The messaging system is configured to extract modifications from content items received from client devices associated with users where the content items are modified using the modifications that comprises a text caption or a media overlay. The messaging system is further configured to determine one or more words from the content items and the extracted modifications and determine a frequency of the one or more words in the content items and the extracted modifications. The messaging system is further configured to determine whether the one or more words is a trend based on the frequency and an aggregate frequency. The messaging system is further configured to in response to the one or more words being determined as the trend, generating trend content associated with the one or more words, the trend content being a text, an image, or an augmentation content.

Abstract: Symbol prediction can be implemented using a multi-task system trained for different tasks. The tasks may include a single symbol prediction, symbol category prediction, and symbol subcategory prediction. Categories of symbols can be generated by clustering sets of training data using a clustering scheme.

Abstract: A neural network system can select content based on user and item content embeddings in an approach that can be updated in real time on the user device without server support. Requests for content sent to the server can include an anonymous user embedding that includes data describing the user's inputs. The content that is nearest to the user embedding in a joint embedding space can be returned as suggested content.

Abstract: Systems and methods for determining how topics evolve in a messaging system extract at least one N-gram from data content (e.g., caption of messages) in the messaging system and detect anomalous behavior in N-gram frequencies over time. The anomalous behavior is used to select candidate N-grams for a determination of whether a topic of a candidate N-gram is evolving or fading. The candidate N-grams are clustered into cluster groups that are used to train at least one time series forecasting model to predict N-gram frequencies in a future time window. A time series of the N-gram frequency is divided into old and recent partitions and pattern recognition is applied to the predicted N-gram frequencies to identify an evolving or fading topic when the difference between a frequency of each anomaly and an average rolling median for each partition is greater for the most recent partition.

Abstract: A machine learning based system can identify an entity as the likely subject of a multimodal message (e.g., a social media post having a short text phrase overlaid on an image) by creating embeddings for an image of the multimodal message and one or more string embeddings from text of the multimodal message. The embeddings can be weighted to maximize information gain, then recombined and compared against a result embedding database to identify an entity as the subject of the multimodal message.

Abstract: Sentiment classification can be implemented by an entity-level multimodal sentiment classification neural network. The neural network can include left, right, and target entity subnetworks. The neural network can further include an image network that generates representation data that is combined and weighted with data output by the left, right, and target entity subnetworks to output a sentiment classification for an entity included in a network post.

Abstract: A machine learning based system can identify an entity as the likely subject of a multimodal message (e.g., a social media post having a short text phrase overlaid on an image) by creating embeddings for an image of the multimodal message and one or more string embeddings from text of the multimodal message. The embeddings can be weighted to maximize information gain, then recombined and compared against a result embedding database to identify an entity as the subject of the multimodal message.

Abstract: A messaging system performs trend analysis on content produced by users of the messaging system. The messaging system is configured to extract modifications from content items received from client devices associated with users where the content items are modified using the modifications that comprises a text caption or a media overlay. The messaging system is further configured to determine one or more words from the content items and the extracted modifications and determine a frequency of the one or more words in the content items and the extracted modifications. The messaging system is further configured to determine whether the one or more words is a trend based on the frequency and an aggregate frequency. The messaging system is further configured to in response to the one or more words being determined as the trend, generating trend content associated with the one or more words, the trend content being a text, an image, or an augmentation content.