US 12,481,631 B2
Information segmentation methods, apparatuses, and electronic devices
Longyin Wen, Los Angeles, CA (US); Xinyao Wang, Los Angeles, CA (US); Dexiang Hong, Beijing (CN); and Congcong Li, Beijing (CN)
Assigned to LEMON INC., Grand Cayman (KY)
Appl. No. 18/572,606
Filed by LEMON INC., Grand Cayman (KY)
PCT Filed Jun. 21, 2022, PCT No. PCT/SG2022/050427
§ 371(c)(1), (2) Date Dec. 20, 2023,
PCT Pub. No. WO2022/271100, PCT Pub. Date Dec. 29, 2022.
Claims priority of application No. 202110716045.9 (CN), filed on Jun. 25, 2021.
Prior Publication US 2024/0289312 A1, Aug. 29, 2024
Int. Cl. G06F 16/30 (2019.01); G06F 16/22 (2019.01)
CPC G06F 16/2228 (2019.01) 19 Claims
OG exemplary drawing
 
1. A method for information segmentation, characterized in that the method comprises:
for a target information node in an information sequence and based on first node information of the target information node, determining a first demarcation point probability value, wherein the first demarcation point probability value indicates a probability that the target information node is a first type of demarcation point;
determining, based on the first demarcation point probability value and second node information of the target information node, a second demarcation point probability value, wherein the second demarcation point probability value indicates a probability that the target information node is a second type of demarcation point; and
determining, based on the first demarcation point probability value and the second demarcation point probability value, at least two segmentation modes for the information sequence, wherein segmentation granularities of different segmentation modes are different,
wherein determining the first demarcation point probability value comprises:
importing, into a first cascaded classifier, information of event advanced features for the target information node, wherein the first cascaded classifier comprises at least two first classifiers, wherein advanced features comprise timing features and/or attention features; and
generating the first demarcation point probability value based on confidences output by respective first classifiers in the first cascaded classifier.