Geo-SPS: bipartite graph representation for GeoSpatial prenatal survey data.

Obstetric studies had long revealed that the human female mental state, although subjective, has a nonlinear relation to the gestation, which could eventually leads to eugenics characteristics. Due to the difference of regions, there are differences between the data, people want to analyze the correlation between the survey data of different regions. Traditional obstetric studies explore health information on understanding and predicting this psychological state. However, most traditional research is based on statistical methods that exploring the correlation between numbers. This type of method lacks an understanding of the natural semantics of obstetric data, so it is prone to problems such as deviation or missing data (such as missing geospatial information) in data analysis. To tackle this problem, we study the use of a generic graph representation on gynecology and obstetric surveys with geospatial features, and propose a bipartite approach, or Geo-SPS, to mine the semantic relationship between low quality health information data. Our solution highlights our unconventional adaptation of novel graph theory from computational physics into biomedical ontology, that carefully maps semantic objects into a bipartite graph. With this tool, Geo-SPS provides a unique approach to semantic similarity metrics. The method supports fast and understandable processing of mixed textual data under different geographic spaces in a graph convolutional neural network. We further evaluate and validate the feasibility of Geo-SPS using a case study on obstetric surveys and health information data from over 3000 pregnant women in three different places from China. Results show that Geo-SPS can effectively represent prenatal mental state from this mixed data set, with an accurate classification on birth defects. [ABSTRACT FROM AUTHOR]