閱讀全文 | |
篇名 |
Low-rank Multimodal Fusion Algorithm Based on Context Modeling
|
---|---|
並列篇名 | Low-rank Multimodal Fusion Algorithm Based on Context Modeling |
作者 | Zongwen Bai、Xiaohuan Chen、Meili Zhou、Tingting Yi、Wei-Che Chien |
英文摘要 | As an important part of human daily life, video contains rich emotion information. Therefore, it is a current research trend to find efficient approaches to conducting emotional analysis on videos. Based on tensor fusion, we propose a low-rank multimodal fusion context modeling. At the beginning, modality information is preprocessed by GRU (Gate Recurrent Unit) in Recurrent Neural Network. We construct semantic dependencies to convey contextual information in the context of the video. The proposed model improves performance of applied emotion classification. Additionally, LMF (Low-rank Tensor Multimodal Fusion) with the advantage of end-toend learning is implemented as a fusion mechanism to improve classification efficiency. We implemented the experiments on CMU-MOSI, POM, and IEMOCAP of multi-modal sentiment analysis, speaker traits and emotion recognition. And results show that our method improved the performance by a margin of 2.9%, 1.3%, and 12.2% respectively contrast with TFN (Tensor Fusion Network). |
起訖頁 | 913-921 |
關鍵詞 | Neural architecture search、Sequence regression models、Performance prediction、Network structure feature |
刊名 | 網際網路技術學刊 |
期數 | 202107 (22:4期) |
出版單位 | 台灣學術網路管理委員會 |
DOI |
|
QR Code | |
該期刊 上一篇
| Cluster-based Deep One-Class Classification Model for Anomaly Detection |
該期刊 下一篇
| Hash Forest Structure Assisted Bi-auditing Protocol with Multiuser Modification in E-health Systems |