This study proposed an integration model that utilizes vector space features of multimodal data to address deficiencies in knowledge integration and logic within Science and Technology linkage research, and facilitate it in agricultural field. The research constructs a Vector Spaces Model (VSM) from three aspects: spatial direction, spatial matrix, and spatial topics. First, based on the social structure of multimodal data, an interactive merging method was used to integrate the linear model and the chain model to construct an Innovation Ecological Chain (IEC) from scientific research to the market. Second, a distance matrix was constructed based on classification of data processing depth and technology maturity to measure the semantic distance on the ecological chain. Third, combined metadata and ontology with Latent Dirichlet Allocation (LDA) to construct the spatial topic model, furthermore, the Large Language Model (LLM) utilizes in multimode data topic extraction. The experiments have shown that, 12 types of data from 1953 to 2023 in the Chinese highland barley can be obtained. Under the landscape of enterprise knowledge discovery, the 65 records topics in the corporate annual reports (2011–2022) spans across 5 sectors in VSM, 6 types of data could integrate in VSM and distribute in 6 sectors of it, and 17 key stakeholders identified from 5 types of data, and the Fund is important to distinguish the identity. Overall, the work verified that multimodal data could integrate under the VSM, while metadata and ontology can quantify the spatial vectors in semantics.