Neural network-based content title generation method, device, device, and computer-readable storage medium

Abstract

The present application discloses a content title generation method based on a neural network including an encoder-decoder network, the method includes: for at least two types of information of the content, respectively generating each of the at least two types of information A vector representation of one type of information; each encoder is used to process the vector representation of each type of information; a decoder is used to process the output of each encoder to generate one or more title words of the content, The decoder calculates the copy probability of the word by calculating the attention weight of the word in the text, thereby determining whether to copy from the word in the text or to generate the title word from the word representation generated after the vector representation of the text is processed by the encoder; and based on The one or more title words generate the content title. Corresponding devices and the like are also disclosed.

Description

technical field [0001] The present application relates to title generation, and more specifically, to a neural network-based content title generation method, device, device, and computer-readable storage medium. Background technique [0002] With the rapid development of the Internet, content data such as text, voice, image and video have entered the era of big explosion. In an age of such enormous amounts of data, the speed at which information can be absorbed from content becomes very important. Compared with text and images, audio and video data increase the time dimension, so it is more time-consuming to understand. Audio and video titles convert a video clip into a simple text description, effectively speeding up the understanding of the video. At the same time, through this method, the audio and video retrieval problems are converted into text retrieval problems, which also facilitates the retrieval of targets in massive data. [0003] Generally, long videos have ti...

AI Technical Summary

Problems solved by technology

This method of title creation has many disadvantages, for example, it requires manpower and time costs, and cannot adapt to the massive video update speed. For example, it has a high level of knowledge reserve, literary accomplishment, and user's title style preference for data labelers. requirements, making it difficult to create short video titles, which affects the production efficiency of short videos

Images