Abstract: Systems, methods, and non-transitory computer-readable media can train a sequence model to output respective captions, or portions of captions, for content items. A determination can be made that a user of the social networking system is posting a content item for publication through a social networking system. A set of captions, or portions of captions, can be determined for the content item being posted based at least in part on the sequence model. The set of captions, or portions of captions, can be provided as suggestions to the user for use in a caption describing the content item being posted.

Abstract: Exemplary embodiments relate to improvements to neural networks for translation and other sequence-to-sequence tasks. A convolutional neural network may include multiple blocks, each having a convolution layer and gated linear units; gating may determine what information passes through to the next block level. Residual connections, which add the input of a block back to its output, may be applied around each block. Further, an attention may be applied to determine which word is most relevant to translate next. By applying repeated passes of the attention to multiple layers of the decoder, the decoder is able to work on the entire structure of a sentence at once (with no temporal dependency). In addition to better accuracy, this configuration is better at capturing long-range dependencies, better models the hierarchical syntax structure of a sentence, and is highly parallelizable and thus faster to run on hardware.

Abstract: Exemplary embodiments relate to improvements to neural networks for translation and other sequence-to-sequence tasks. A convolutional neural network may include multiple blocks, each having a convolution layer and gated linear units; gating may determine what information passes through to the next block level. Residual connections, which add the input of a block back to its output, may be applied around each block. Further, an attention may be applied to determine which word is most relevant to translate next. By applying repeated passes of the attention to multiple layers of the decoder, the decoder is able to work on the entire structure of a sentence at once (with no temporal dependency). In addition to better accuracy, this configuration is better at capturing long-range dependencies, better models the hierarchical syntax structure of a sentence, and is highly parallelizable and thus faster to run on hardware.

Abstract: Systems, methods, and non-transitory computer-readable media can determine one or more chunks for a content item to be captioned. Each chunk can include one or more terms that describe at least a portion of the subject matter captured in the content item. One or more sentiments are determined based on the subject matter captured in the content item. One or more emotions are determined for the content item. At least one emoted caption is generated for the content item based at least in part on the one or more chunks, sentiments, and emotions. The emoted caption can include at least one term that conveys an emotion represented by the subject matter captured in the content item.

Abstract: Systems, methods, and non-transitory computer-readable media can train a sequence model to output respective captions, or portions of captions, for content items. A determination can be made that a user of the social networking system is posting a content item for publication through a social networking system. A set of captions, or portions of captions, can be determined for the content item being posted based at least in part on the sequence model. The set of captions, or portions of captions, can be provided as suggestions to the user for use in a caption describing the content item being posted.

Abstract: One or more aspects of the subject disclosure are directed towards performing a semantic parsing task, such as classifying text corresponding to a spoken utterance into a class. Feature data representative of input data is provided to a semantic parsing mechanism that uses a deep model trained at least in part via unsupervised learning using unlabeled data. For example, if used in a classification task, a classifier may use an associated deep neural network that is trained to have an embeddings layer corresponding to at least one of words, phrases, or sentences. The layers are learned from unlabeled data, such as query click log data.