Answers are grounded in this paper and cite their sources.
Due to the huge amount of multimedia data and the backwardness of traditional data management technology for multimedia data management, it is difficult for people to get the data of interest from the huge amount of multimedia data. Therefore, the problems of "data overload" and "more multimedia data but less knowledge" have emerged. Recommendation system is born to solve such problems. Today, recommendation systems are becoming the core technology of many Internet products. The core problem that recommendation algorithms solve is to evaluate goods or services that users have not consumed themselves, and to recommend goods or services that may be of interest to users. Traditional video recommendation based on the metadata of video (text) or image content of video (image) as video content analysis only considers the content analysis of individual video modalities. However, video data contains information of multiple modalities of text, image and audio, and the multimodal content and the correlations in it are not explored. Therefore, existing video recommendation systems cannot achieve multi-view recommendation of videos. This paper addresses this problem by innovatively converting the multimodal information fusion process in knowledge graphs into an information propagation process on graphs and designing a new model. An intermediate semantic based metric is used. Based on the unimodal labeling, multimodal labeling is proposed to describe the multimodal semantic information in the video. Further, a multicore learning architecture is used to combine the semantics of multiple modalities so as to map the underlying multimodal features to a unified intermediate semantic space and learn the similarity metric function between videos based on the intermediate semantic space. Extensive experiments are conducted on real datasets. The experimental results show that our model outperforms the current mainstream collaborative filtering-based recommendation algorithms, knowledge graph-based recommendation algorithms, and multimodal learning-based recommendation algorithms.
No takes yet. Share an insight, caveat, or question.
Mark et al. (2022) studied this question.