City-state size prediction based on deep learning quantitative modeling
Download as PDF
DOI: 10.25236/iwmecs.2023.007
Corresponding Author
Jixuan Wang
Abstract
The Lu County Old City Site is a Western Han Dynasty site discovered during the pre-construction archaeological exploration of the urban sub-center in 2016.By studying the case of the Lu County Old City Site, we can explore and discover the factors that determine the size and development of ancient cities. In this paper, we will quantify the factors affecting the scale and development of ancient cities and construct a model of the important variables and their correlations that determine the scale of ancient city-states. The research methodology includes collecting data and quantitatively analyzing the city size and a variety of related variables (e.g., political system, social culture, economic level, agricultural level, foreign trade, and local climate, etc.), and interpreting the size of ancient city-states as a model accompanying the development of the important variables based on the modern theory of urban economy to illustrate how the size of the city was key variables such as political system, social culture, economic level, and agricultural level Correlation. We adopt the method of constructing a knowledge graph to build a Bert-BiLSTM-Attention-CRF model that incorporates the mechanism of attention, thus accurately and efficiently quantifying the multiple indicators associated with the prediction of city-state size. The model is divided into four layers: the first layer is the BERT layer, which pre-trains the input text to obtain word vectors; the second layer is the BiLSTM layer, which takes the word vectors obtained from the first layer as the input and performs bi-directional training to capture the contextual features of the text; the third layer is the Attention layer, which assigns weights to the different contextual information, and extracts the features that are crucial for knowledge recognition; the fourth layer is the CRF layer, which decodes and annotates the output of the previous layer to obtain the global optimal sequence. Finally, the accuracy of the proposed quantization model and the feasibility of the city-state size prediction model are demonstrated through several sets of experiments.
Keywords
Deep learning; LSTM; quantization models; city-state size prediction