Abstract
Social media advertising contains textual and visual information that can express sentiment in different and sometimes inconsistent ways. This study investigates multimodal sentiment analysis by combining sentiment-dictionary information, OCC-inspired emotional rules, and collaborative attention for text-image feature integration. Therefore, the SA of social media advertising texts is proposed. The sentiment word set is constructed by combining the multi-source sentiment dictionary, and the sentiment polarity is determined through voting rules. Text classification is carried out utilizing the BP neural network. A fusion strategy derived from modal contribution is proposed. The appropriate fusion method is selected by calculating the contribution of each modal to the whole. The OCC model is applied to the SA of social media advertising texts. From the perspectives of event results and object behaviors, simplified emotional rules were established. By using dictionary resources such as HowNet and SentiWordNet, a calculation method of sentiment tendency intensity values derived from sets of meanings and synonyms was proposed. SA is achieved by simultaneously learning the attention distribution of text and images through the collaborative attention mechanism. The reported experiments indicate promising sentiment-classification performance under the evaluated settings. However, generalization across different datasets, platforms, languages, time periods, and user groups has not been established. When processing 20,000 sample documents, it only takes 45.89 seconds. Under the condition that the word vector is 50 dimensions, the accuracy rate of advertising text evaluation reaches 89%.
Recommended Citation
Songtao, Guan
(2026)
"Multimodal Sentiment Analysis for Social Media Advertising: An OCC-Inspired Framework with Collaborative Attention,"
Iraqi Journal for Computer Science and Mathematics: Vol. 7:
Iss.
4, Article 2.
DOI: https://doi.org/10.52866/2788-7421.1425
Available at:
https://ijcsm.researchcommons.org/ijcsm/vol7/iss4/2

