Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Modelling News Sentiment Flow Using Spatial Hawkes Processes: Dependencies Between Topics and Countries

Brishammar, Carl (2018) FMS820 20181
Mathematical Statistics
Abstract
This Master Thesis proposes to use a Spatial Hawkes Process to model the news
flow in the world. The spatial part is divided in both geographical areas as well
as different topics. Therefore, in the Hawkes model every news article corresponds
to a point in space and time. In a certain region and topic the intensity of the
released articles will be modelled with the Hawkes Process. This can be of interest
for various applications depending on the topic and region chosen. The data in this
project comes from the company RavenPack which has labelled every news article
with a topic and a region. The area specifically examined in this report will be the
relations between the different areas in the news flow. A comparison will also be
... (More)
This Master Thesis proposes to use a Spatial Hawkes Process to model the news
flow in the world. The spatial part is divided in both geographical areas as well
as different topics. Therefore, in the Hawkes model every news article corresponds
to a point in space and time. In a certain region and topic the intensity of the
released articles will be modelled with the Hawkes Process. This can be of interest
for various applications depending on the topic and region chosen. The data in this
project comes from the company RavenPack which has labelled every news article
with a topic and a region. The area specifically examined in this report will be the
relations between the different areas in the news flow. A comparison will also be
done between some different spatial divisions to see if different behaviour can be
captured with a more complex model, with more regions and topics.
The model is compared to a Poisson Process model. It seems that the Hawkes
model works better than the Poisson Process to model the intensity of the different
parts of the news flow in all cases. The results also indicates that a very flexible
model will be able to capture more cases that are known from history. The complex
model sees connections that increase the intensity during hectic times in the recent
past news flow, for example during Brexit and the Arab Spring. However it seems
that with more regions the model is prone to overfit and a simpler model may be
preferable for out of sample uses (Less)
Please use this url to cite or link to this publication:
author
Brishammar, Carl
supervisor
organization
course
FMS820 20181
year
type
H2 - Master's Degree (Two Years)
subject
language
English
id
8952220
date added to LUP
2018-06-21 13:23:51
date last changed
2018-06-21 13:23:51
@misc{8952220,
  abstract     = {{This Master Thesis proposes to use a Spatial Hawkes Process to model the news
flow in the world. The spatial part is divided in both geographical areas as well
as different topics. Therefore, in the Hawkes model every news article corresponds
to a point in space and time. In a certain region and topic the intensity of the
released articles will be modelled with the Hawkes Process. This can be of interest
for various applications depending on the topic and region chosen. The data in this
project comes from the company RavenPack which has labelled every news article
with a topic and a region. The area specifically examined in this report will be the
relations between the different areas in the news flow. A comparison will also be
done between some different spatial divisions to see if different behaviour can be
captured with a more complex model, with more regions and topics.
The model is compared to a Poisson Process model. It seems that the Hawkes
model works better than the Poisson Process to model the intensity of the different
parts of the news flow in all cases. The results also indicates that a very flexible
model will be able to capture more cases that are known from history. The complex
model sees connections that increase the intensity during hectic times in the recent
past news flow, for example during Brexit and the Arab Spring. However it seems
that with more regions the model is prone to overfit and a simpler model may be
preferable for out of sample uses}},
  author       = {{Brishammar, Carl}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Modelling News Sentiment Flow Using Spatial Hawkes Processes: Dependencies Between Topics and Countries}},
  year         = {{2018}},
}