Video semantics labeling method and device based on bullet screen and electronic equipment

Abstract

The embodiment of the invention provides a video semantics labeling method based on a bullet screen. The method includes: obtaining all words in the bullet screen of a target video and corresponding time stamps; averagely dividing the target video into a preset number of time slices; generating an initial topic set, which contains topics corresponding to all the time slices, and an initial plot set, which contains plots corresponding to all the time slices, according to preset probability correspondence relationships of words and the topics and the plots; generating a dictionary vocabulary setand a vocabulary distribution matrix; calculating temporal a priori information of the dictionary vocabulary set; using a preset total probability formula of bullet-screen vocabulary to calculate probability that each piece of the dictionary vocabulary corresponds to each topic and each plot; generating plot-topic distribution matrices of the time slices; merging adjacent similar time slices intoone time slice; determining plots corresponding to all time slices; and labeling the target video. By applying the scheme provided by the embodiment of the invention for video semantics labeling, labeling on video semantics is enabled to be more accurate.

Description

technical field [0001] The present invention relates to the technical field of video tagging, in particular to a bullet chat-based video semantic tagging method, device and electronic equipment. Background technique [0002] Online video occupies a huge amount of traffic on the Internet, and hundreds of videos are uploaded to the global Internet video platform every day, so the management of online video is very important. Video labeling can improve network traffic resource utilization and online video management efficiency, and is a key link in online video management. [0003] Video websites that have emerged in recent years have added a "barrage" function, which enables users to comment in real time on the episodes they are currently watching. Based on this, we can use the rich comment data in the bullet chat to cluster the bullet chat according to the density of the bullet chat, and then get the time slice of this type of bullet chat according to the earliest and latest...

AI Technical Summary

Problems solved by technology

[0004] However, the existing methods of using bullet chatting to semantically tag videos only determine the time boundary of the plot according to the distribution of bullet chatting in time, ignoring the consideration of the interaction characteristics of bullet chatting, and not taking into account the current time. The content discussed in the barrage may be related to the plot corresponding to the adjacent time, but has nothing to do with the plot corresponding to the current time, so the division of the plot is not accurate enough, so that the semantic annotation of the video is not accurate enough

Images