Συντάχθηκε 20-02-2012 10:46
από Galateia Malandraki
Email συντάκτη: gmalandraki<στο>tuc.gr
Ενημερώθηκε:
-
Ιδιότητα: υπάλληλος ΑΡΜΗΧ.
ΠΟΛΥΤΕΧΝΕΙΟ ΚΡΗΤΗΣ
Τμήμα Ηλεκτρονικών Μηχανικών & Μηχανικών Υπολογιστών
ΠΑΡΟΥΣΙΑΣΗ ΜΕΤΑΠΤΥΧΙΑΚΗΣ ΔΙΑΤΡΙΒΗΣ
ΜΟΣΧΟΠΟΥΛΟΥ ΘΕΟΔΟΣΙΟΥ
με θέμα
Extraction of Policy Networks using
Web-based Features
Τρίτη 21 Φεβρουαρίου 2012, 13:30 μμ
Αίθουσα Ε3002, Πολυτεχνειούπολη
Εξεταστική Επιτροπή
Αν. Καθ. Αλέξανδρος Ποταμιάνος (επιβλέπων)
Καθ. Βασίλειος Διγαλάκης
Αν. Καθ. Ευριπίδης Πετράκης
Abstract
Policy making process in modern democratic systems is the outcome of the interrelations and interdependencies among political entities (i.e organizations, companies, groups or unions) from public or private sectors and from different levels of governance. Thus, the network perspective, namely policy network, is an efficient tool for political scientists to describe, analyze and explain various financial and social phenomena during the policy making process. A policy network can be described as a social graph with nodes the actors and edges the relations among them. The relations in a policy network serve as channels for communication, exchange of information, expertise, trust and other policy resources. Traditionally, policy networks are created manually after a series of arduous, time consuming steps including interviews and questionnaires. Furthermore, the manual creation of such networks is often a high budget procedure which requires high-level of expertise. Another problem is that manually created policy networks suffer from subjective biases such as the respondent’s will for participation, cultural and political issues even external factors such as the economical or political system.
In this work we propose a method for the automatic extraction of policy network. More specifically, given the actors of the network, our approach estimates the strength of relations among them. Our fundamental assumption is that the strength of such relations can be discovered automatically and in an unsupervised way through a variety of features that can be harvested from the web. Such features include webpage counts, outlinks and lexical information that is extracted from web queries, web documents or web snippets. In our work, we propose three types of metrics as well as their fusion i) page-count-based metrics that use the number of occurrences/co-occurrences of actors in web documents ii) text-based metrics that exploit the actors lexical context in web snippets iii) link-based metrics that use the outlinks cited in web documents where the actors exist. iv) the linear combination of the three types of metrics above. The proposed approach is automatic and does not require any external knowledge source, other than the specification of the word forms that correspond to the political actors. It is also language independent as it is not based on any knowledge about the language. Furthermore, the proposed approach reduces the biases emerged by the traditional methods (who depend on the answers of a small number of respondents) as it can integrate multiple points of view by exploiting the collective information of the web. Our approach is evaluated on two human-rated networks taken from the political science literature. The networks are located in Ireland and Greece and web queries are performed in English and Greek respectively. Furthermore, the extracted networks are visualized and qualitatively evaluated by political scientists. Based on the fact that relations in policy networks evolve through time, we apply our method to extract the networks for the years of a specific time period and visualize this evolution.
It is shown that our method can efficiently estimate the strength of relations that express cooperation (positive relations), while fail to estimate relations of antagonism (negative relations). Furthermore, our approach effectively identifies the most ‘active’ actors computing the degree of centrality which is a widely used measure in network analysis. Finally, the visualization of the policy networks as well as their evolution draw interesting results and conclusions from the perspective of political sciences.
Συνημμένα:
-
Master_Moschopoulos_anakoinwsh.doc
Μεταφορτώσεις: 179,
Μέγεθος: 41 KB application/octet-stream
