US 10,891,427 B2
Machine learning techniques for generating document summaries targeted to affective tone
Kushal Chawla, Karnataka (IN); Balaji Vasan Srinivasan, Bangalore (IN); and Niyati Himanshu Chhaya, Telangana (IN)
Assigned to ADOBE INC., San Jose, CA (US)
Filed by Adobe Inc., San Jose, CA (US)
Filed on Feb. 7, 2019, as Appl. No. 16/270,191.
Prior Publication US 2020/0257757 A1, Aug. 13, 2020
Int. Cl. G06F 16/00 (2019.01); G06F 40/166 (2020.01); G06N 20/00 (2019.01); G06F 40/20 (2020.01); G06F 16/34 (2019.01)
CPC G06F 40/166 (2020.01) [G06F 16/345 (2019.01); G06F 40/20 (2020.01); G06N 20/00 (2019.01)] 20 Claims
OG exemplary drawing
1. A non-transitory computer-readable medium for generating affective summarizations of text documents, the non-transitory computer-readable medium embodying program code comprising instructions which, when executed by a processor, cause the processor to perform operations comprising:
receiving, by an affective summarization system, a summarization training dataset including an article word sequence and a summary word sequence;
receiving, from a predictor subnetwork included in the affective summarization system, a predicted affect level based on a normalized affect score of the summary word sequence;
training, by an embeddings generator included in the affective summarization system, an embeddings function based on the predicted affect level, wherein the embeddings function is trained to provide an embeddings sequence including either a vocabulary token or a one-hot vector, based on the predicted affect level and the article word sequence;
modifying the summarization training dataset by removing the article word sequence and adding the embeddings sequence; and
training a summarization subnetwork that is included in the affective summarization system to provide an affective text summary based on the modified summarization training dataset.