Abstract: Methods, systems, and non-transitory computer readable storage media are disclosed for training a machine-learning model utilizing batchwise weighted loss functions and scaled padding based on source density. For example, the disclosed systems can determine a density of words or phrases in digital content from a digital content source that indicate an affinity towards one or more content classes. In some embodiments, the disclosed systems can use the determined source density to split digital content from the source into segments and pad the segments with padding characters based on the source density. The disclosed systems can also generate document embeddings using the padded segments and then train the machine-learning model using the document embeddings. Furthermore, the disclosed system can use batchwise weighted cross entropy loss for applying different class weightings on a per-batch basis during training of the machine-learning model.

Abstract: Systems, methods, and non-transitory computer-readable media are disclosed for utilizing focused aggregation of classification model outputs to classify variable length documents. For instance, the disclosed systems can utilize a classification model to determine category scores for segments from an electronic document. Furthermore, the disclosed systems can identify positive trigger segments from the segments by comparing the category scores to a threshold category score. Moreover, the disclosed systems can determine a positive trigger ratio for the target category based on the positive trigger segments and the segments. Additionally, the disclosed systems can generate an aggregated category score for the electronic document from the positive trigger segments (when the positive trigger ratio satisfies a threshold positive trigger ratio) and distribute the electronic documents to client devices based on the aggregated category score.

Abstract: Methods, systems, and non-transitory computer readable storage media are disclosed for training a machine-learning model utilizing batchwise weighted loss functions and scaled padding based on source density. For example, the disclosed systems can determine a density of words or phrases in digital content from a digital content source that indicate an affinity towards one or more content classes. In some embodiments, the disclosed systems can use the determined source density to split digital content from the source into segments and pad the segments with padding characters based on the source density. The disclosed systems can also generate document embeddings using the padded segments and then train the machine-learning model using the document embeddings. Furthermore, the disclosed system can use batchwise weighted cross entropy loss for applying different class weightings on a per-batch basis during training of the machine-learning model.

Abstract: Systems, methods, and non-transitory computer-readable media are disclosed for utilizing focused aggregation of classification model outputs to classify variable length documents. For instance, the disclosed systems can utilize a classification model to determine category scores for segments from an electronic document. Furthermore, the disclosed systems can identify positive trigger segments from the segments by comparing the category scores to a threshold category score. Moreover, the disclosed systems can determine a positive trigger ratio for the target category based on the positive trigger segments and the segments. Additionally, the disclosed systems can generate an aggregated category score for the electronic document from the positive trigger segments (when the positive trigger ratio satisfies a threshold positive trigger ratio) and distribute the electronic documents to client devices based on the aggregated category score.