Multimodal Sentiment Analysis of Online Product Marketing Information Based on Artificial Intelligence Neural Networks and Text Mining

With the rise of multimodal content (such as text and images) in online product marketing, sentiment analysis techniques face increasing demands for accuracy and versatility. However, existing approaches often struggle with modality coordination, deep emotional feature extraction, and semantic consistency, which hinder effective user sentiment recognition in complex marketing contexts. To address these challenges, this work proposes a hybrid fusion multimodal sentiment analysis (HF-MSA) model that combines artificial intelligence neural networks with text mining techniques. The focus is on efficiently integrating heterogeneous modality data to enhance the robustness and interpretability of sentiment analysis. A big-data-driven multimodal mining framework is developed, utilizing bidirectional encoder representations from transformers and bidirectional long short-term memory networks to extract textual features, while a gated graph convolutional network combined with a self-attention mechanism models syntactic dependencies. For the image modality, channel and spatial attention modules are incorporated to improve key region recognition. Supported by a dual-layer fusion strategy at both the feature and decision levels, the HF-MSA model achieves an accuracy of 77.43% and an F1 score of 72.84% on the Twitter17 dataset, reflecting improvements of 2.82% and 3.02%, respectively, over existing models. This work advances the theoretical development of multimodal sentiment analysis from static fusion to dynamic collaboration and provides a scalable tool for user sentiment insights, helping enterprises optimize online marketing strategies and enhance user engagement.