Word Co-Occurrence Networks from Ancient Greek Texts and the Cultural Evolution of Prosocial Religions
KAŠE, V. Word Co-Occurrence Networks from Ancient Greek Texts and the Cultural Evolution of Prosocial Religions. Brno, 2018.
|Anglický název:||Word Co-Occurrence Networks from Ancient Greek Texts and the Cultural Evolution of Prosocial Religions|
|Autoři:||Mgr. Vojtěch Kaše|
|Abstrakt EN:||Word co-occurrence networks represent a popular method with many natural language processing and information retrieval applications, like keywords extraction or automatic text summarization. In network science, these networks are widely studied, as they reveal interesting network features, e.g., the so called small-world effect. Here I focus on construction, analysis, visualization and further usage of a specific subgroup of these networks - namely weighted undirected word-adjacency networks. The network is constructed as follows: whenever two words appear next to each other in a document, there emerges an edge between the two, having weight equal to 1; whenever the neighbourhood re- appear, the weight of the edge is increased by 1. For my research purposes, here I work with a corpus of pro-processed Greek texts with words in their lemmatized form, coupled with part of speech tags (N = 25,522,507). From this corpus I extract only words tagged as nouns or adjectives (N = 6,957,624). I further divide this reduced corpus into four sub-corpora: (a) ancient Greek texts estimated to be written before the 4th c. BC, (b) texts from the 4th c. BC to the 1st c. BC, (c) texts from the 1st c. CE to the 4th c. CE and (d) texts from the 5th c. CE on. From these sub-corpora, I generate 4 networks. Having them, I analyze neighbourhood of certain nodes in the network, namely the Greek term ‘theos’, God, especially how the neighbourhood changes from one period to another. For that purpose I apply Dijsktra’s shortest path algorithm, with the distance edge attribute equal to inverse edge weight. The texts are pre-processed using Python 3, networks are generated and analyzed with NetworkX and visualized using Plotly. I suggest that this methodology can help us to explore certain hypothesis from the cultural evolutionary study of religion concerning the emergence and spread of prosocial religions in human history.|