Applying Neural Networks to Person's Values Structure Prediction

With the development of deep learning, artificial neural network has become a popular and versatile architecture for modeling based on large multidimensional data. As the humans related big data appears, the potential of ANNs usage within sociological context increases due to cases when large number of predictors with complicated relations to target variable so that manual search of these connections becomes impossible. While traditional methods of sociological prediction may either not provide non-linear relationships option, as in the case of regression, or may lack the noise resistance, as in the case of decision trees, ANNs can presumably provide the flexibility and stability needed by such a wide class of situations. The person’s values structure type ​​prediction is an attractive possibility since it can be used to predict behavior, preferences, especially political behavior and choice, because of the Schwartz’s values and behavior connection. The prediction of the person’s values structure type based on the large set of non-value factors presumably represents the type of tasks where neural networks has high potential compared to traditional models used in the sociological analysis because of the need to approximate nonlinear connection, which can be done by generating intermediate representations of the data between the factors and the target variable - the key principle of neural network functioning. In my research I try to simulate the situation of relatively large amount of data available based on ESS data, using it for the prediction of person’s values structure type based on Schwarz values. Being one of the most fundamental and therefore difficult to the prediction phenomenon, with many useful applications when known, it, hopefully, allows to simulate the big-data-related tasks context. In this context I hope to demonstrate the advantage of neural networks as the classification model in comparison to more traditional methods of empirical sociological research and derive recommendations to neural network hyperparameters tuning applicable beyond my particular problem setting.