Bullet chat topic extraction method, medium, equipment and system based on n-gram model

Abstract

The invention discloses a barrage topic extraction method, medium, equipment and system based on an N-gram model, and relates to the field of live broadcasting. The method includes the following steps: extracting barrage data; extracting features corresponding to words expressing a specific intention and adding them to a custom lexicon; adding words without practical meaning to a custom stop lexicon; data preprocessing: removing " The data in which the "Bullet Chat Content" field is empty; remove the punctuation marks in the "Bullet Chat Content" field; the preprocessed bullet chat content is represented by the N-gram model, which represents the occurrence of a certain word in the sentence The probability is related to the previous N-1 words, and N is a positive integer; divide each barrage content into a set of word vectors, and divide each barrage content according to the word formation rules in the custom lexicon. Disable thesaurus to filter useless words. The present invention can accurately extract barrage topics.

Description

technical field [0001] The present invention relates to the field of live broadcasting, in particular to a method, medium, device and system for extracting the subject of a bullet screen based on an N-gram model. Background technique [0002] The main text content of the live broadcast platform is generally expressed as a bullet screen. In order to count the bullet screen content, it is necessary to extract the bullet screen text information of the live broadcast platform. At present, most of the traditional bullet screen text extraction solutions in the live broadcast industry use manual tagging, which consumes a lot of manpower and material costs. The method is obviously inefficient. Moreover, the existing bullet screen text is simply represented based on the bag-of-words model, ignoring the relationship between a single word and the context, which makes the extraction of bullet screen topics inaccurate. SUMMARY OF THE INVENTION [0003] The purpose of the present inve...

AI Technical Summary

Problems solved by technology

At present, most of the traditional barrage text extraction solutions in the live broadcast industry use manual labeling. This method consumes a lot of manpower and material costs. The method is obviously inefficient

Moreover, the existing bullet chat text is represented purely based on the bag of words model, ignoring the relationship between a single word and the context, making the extraction of bullet chat topics inaccurate

Images