| In recent years,the domestic Internet content field,Toutiao,Netease,Yidianzixun information,Bai Jiahao and other platforms have risen.These content distribution platforms make use of multi-channel content sources,and the number of content has surged,among which short video content occupies a large proportion.Therefore,it is very important to set appropriate short video recommendation schemes for these platforms.Compared with previous videos,short video is easy to share,spread quickly,and its contents are short and accurate,and caters to users' reading habits of fastconsuming content.Based on the background of the explosive growth of the number of short video content on the content distribution platform,taking into account the characteristics of the short video itself,this paper designed a recommended architecture platform suitable for short video,and candidate generation,filtering,and ranking the short video recommendation process,convergence in an architecture platform.Some solutions to the problems that may occur in the early stage of the platform,such as cold start,content pool construction,and classification label system construction,have been proposed.At the same time,for the ranking link in the short video recommendation framework,this article introduces the Text Convolutional Neural Network(TextCNN)algorithm that shines in the field of text classification to process short video text features.In the video scoring phase,the depth neural network(DNN)algorithm performs score prediction,and finally ranks each user's rating for short videos,and sorts them in descending order according to the predicted score results,and finally recommends the short video list to the user.This article also conducts data experiments on the algorithm of the sorting stage in the video recommendation framework,and the data is taken from a content distribution platform App.Based on the TextCNN+DNN algorithm proposed in this paper,the RMSE value is 0.892,and the results are compared with the RMSE results of UserCF,ItemCF and the implicit semantic model under the same data.Experiments show that the TextCNN+DNN algorithm has a better performance in predicting results and model generalization ability under the same experimental data. |
