| CPC G06F 40/40 (2020.01) [G06F 40/232 (2020.01); G06F 40/30 (2020.01); G06N 3/02 (2013.01)] | 19 Claims |
|
1. A non-transitory machine-readable medium having executable instructions to cause one or more processing units to perform a method to generate a plurality of headlines, the method comprising:
receiving a strategy for generating the plurality of headlines, wherein each of the plurality of headlines is a document outline for content creation;
collecting a collection of content;
generating a plurality of content clusters using unsupervised machine learning to cluster the content collection with the received strategy, wherein the generating involves a feedforward neural network comprising a probabilistic language model and an encoder, the encoder acting as a conditional summary, the encoder comprising an encoder function implementing an attention-based model, the encoder function extending a bag of words model by using a learned soft alignment between an input and the conditional summary; and
summarizing each of the plurality of content clusters to generate the plurality of headlines, each of the plurality of headlines comprising a score, the score assigned in association with a type, the type comprising a set of non-text parameters in the plurality of content clusters, the type further comprising an estimate of a degree of emotionality evoked by the headline, wherein each of the plurality of headlines is scored using a Mixture of Experts saliency model for which each expert is a particular type attribute.
|