In the current era of e-commerce, online reviews are regarded as crucial for evaluating products and have a significant impact on consumer choices. Nevertheless, the user experience is being negatively impacted by the exponential growth of unstructured review data and the volume of reviews, which presents obstacles to the development of accuracy and feature selection in useful review prediction models. Moreover, existing scholarly investigations predominantly concentrate on the extraction of features from the textual and contextual strata of reviews, frequently neglecting the image data that may be present within said reviews. Furthermore, the multimodal characteristics of textual, photographic, and contextual attributes require the implementation of multimodal fusion techniques in order to refine information. This paper develops a multi-layer feature set utilising KAM (Knowledge-Aware Multimodal) theory in light of the fact that both review text and images are regarded as influential factors affecting the utility of online reviews. In order to facilitate the interaction and fusion of cross-modal information, a Tri-Modal Collaborative Attention Mechanism-based Cognitive and Semantic Computing influence review prediction model (TMCAM) is proposed for the three modalities of data. The experimental findings substantiate the TMCAM model's superior performance by illustrating that when image and text information are combined, the outcomes are more favourable than when single-modal information is used alone; contextual features can aid in forecasting the usefulness of reviews; and the use of collaborative attention mechanisms to exchange cross-modal information improves the perception of review usefulness in comparison to simply concatenating modal features. © 2025 Author(s).