Political Ideology Analysis of Hong Kong Media

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering

Final Year Thesis Oral Defense

Title: "Political Ideology Analysis of Hong Kong Media"

by

TSE Ho Nam

Abstract:

Media cannot be 100% politically neutral. Almost all media report on news 
selectively and by using specific word choices. This reality can be 
observed in recent events on the Anti-extradition Bill movement in Hong 
Kong, where on the same incident, different media would report it using 
different descriptions. An example can be seen in reports of a 
demonstration in the Hong Kong Airport, where one pro-democratic newspaper 
used 'citizens' and 'airport stress-test', and one pro-Beijing newspaper 
used 'rioters' and 'unlawful assembly'.

However, this phenomenon is not ideal. Humans are susceptible to biases, 
and it is difficult for both writers and readers to counter this. In view 
of this, in our thesis, we built an interpretable machine learning model 
using BERT and Hierarchical Attention Network (HAN) on political ideology 
detection in Hong Kong Chinese newspapers. In particular, we trained a 
supervised learning model that predicts the source of an article from its 
content. We then leveraged the design of HANs to capture how the machine 
learning model makes its judgement by showing us which words or phrases in 
the article are believed to be indicative of the political stance. We 
argue the model is robust from both in terms of numerical accuracy and 
linguistic analysis of the outputs of the model.


Date            : 14 May 2020 (Thursday)

Time            : 11:50 - 12:30

Zoom Meeting    : https://hkust.zoom.us/j/592856162

Advisor         : Dr. SONG Yangqiu

2nd Reader      : Prof. LIN Fangzhen