*Result*: Multimodal Deep Learning Models for Unstructured Data Integration in Enterprise Analytics.

*The necessity for sophisticated processing systems that can effectively extract valuable insights has increased due to the growth of multi-modal unstructured data. Despite the potential, current studies show significant limitations, such as poor multi-modal data combining, insufficient big data volume or variation management, and a lack of thorough taxonomies for classifying AIbased ISs. In order to provide a more thorough understanding and forecast of consumer behaviour, this study proposes a multimodal framework for learning that combines multiple information sources, including user location, behaviour data, and product attributes. It does this by integrating multimodal data analysis or big data technology. Big data analytics alone are not enough to handle a threat landscape that is expanding quickly and getting more complex every day. Unstructured big data can take any form, including text, audio, photos, and video, and has no set organisation or structure. The issues of emotions and sentiment modelling resulting from unstructured large data with many modalities are discussed in this work. First, we provide a current overview of emotion and sentiment modelling, encompassing the most advanced methods. Next, we provide a novel architecture for large data sentiment and multimodal emotion modelling. Data collection, multimodal data aggregation, multimodal data feature extraction, fusion and decision, and application are the five key components of the suggested architecture. The multimodal data feature extraction module in the design is suggested to use two new feature extraction methods: divide-and-conquer linear discriminant analysis (DivConLDA) and divide-and-conquer principal component analysis (Div-ConPCA). To verify the effectiveness of the suggested methods, tests are conducted on a multicore computer architecture. [ABSTRACT FROM AUTHOR]*