A Semantic Orthogonal Mapping Method Through Deep-Learning for Semantic Computing

In order to realize an artificial intelligent system, a basic mechanism should be provided for expressing and processing the semantic. We have presented semantic computing models in which original data are mapped in to a semantic space and presented as points in semantic spaces. That is, we presented a method to process semantic information by calculating Euclidean distances of those points in the semantic spaces. In our continuous studies, we note that different mapping matrixes are required to map the original data in to the semantic space when this model is applied in different application areas. Therefore, it is an important research topic to develop methods to create the mapping matrixes applied in different areas. Many research works are presented on applying the model in the areas of semantic information retrieving, semantic information classifying, semantic information extracting, and semantic information analyzing on reason and results, etc. In these works, the mapping matrixes are created based on the analyzations in the application areas with human knowledge. In this paper, we present a new method to perform the semantic mapping through deep-learning computation. The most important feature of our method is that we implement semantic mapping through training data sets rather than the mapping matrix which is created based on the analyzations of human being. We first discuss five basic operations, the semantic space creation, semantic mapping, semantic mapping matrix, semantic space expansion and contraction. After that, we present our method. In order to present correlations of the semantic information correctly in Euclidean distances, the axes of a semantic space must be orthogonal to each other. Therefore, we also discuss how to implement semantic orthogonal mapping. We believe that our study will open new application areas on semantic computing and deep-learning.