Video extraction method, device, device and medium based on inflection point detection

Abstract

The present invention relates to the field of artificial intelligence, and provides a video extraction method, device, device and medium based on inflection point detection, which can divide the video to be processed according to each sudden increase inflection point, obtain at least one target segment, and improve the extracted high-hot video The interpretability of fragments effectively solves the inaccuracy caused by simple analysis of macro data, and at the same time eliminates the adverse effects of external factors on hot video extraction, and compares the video class of each target fragment with the knowledge points of each target fragment Mapping, get the video extraction result of each target segment, realize the automatic extraction feedback of hot videos, at the same time, the feedback content has been clustered and refined, on the one hand, avoid wasting time on browsing duplicate information, and on the other hand, avoid missing in the manual screening process Important Feedback Content. In addition, the present invention also relates to blockchain technology, and video extraction results can be stored in blockchain nodes.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence, in particular to a video extraction method, device, equipment and medium based on inflection point detection. Background technique [0002] As more and more scenarios in corporate training are gradually migrating from offline to online, a cycle mechanism of course output, feedback, and optimization adapted to the online teaching mode is needed to improve teaching quality. [0003] In the existing technology, macroscopic data such as the total number of views of teaching videos and the total number of bullet screens are usually used to evaluate the overall level of video content and the points of interest of viewers. However, factors other than teaching quality have a greater impact on such macroscopic data, such as: video Promotion intensity, cycle, video-oriented group size, etc. [0004] Specifically, when looking for high-profile video clips through bullet chat data, i...

AI Technical Summary

Problems solved by technology

[0004] Specifically, when looking for high-profile video clips through bullet chat data, if you directly use the bullet chat release quantity index to extract time nodes, on the one hand, the difference in the number of viewers between different videos is greatly affected by external factors; on the other hand, almost all videos There is a natural attenuation phenomenon in the number of viewers. The number of viewers at the end of the video is naturally smaller than that at the beginning of the video. The base of the number of bullet chats has changed. It is difficult to formulate a reasonable threshold for the number of bullet chats to extract the time points of high-profile content.

However, if the threshold is directly determined by the proportion of the number of people posting bullet chatting, since the teacher style and class mode of different videos will have an impact on the proportion of the number of people posting bullet chatting, it is impossible to objectively find out the most popular videos with a unified ratio index. fragment

Images