Files in this item

FilesDescriptionFormat

application/pdf

application/pdfZou_Jing.pdf (208kB)
(no description provided)PDF

Description

Title:News and financial market
Author(s):Zou, Jing
Advisor(s):Zhai, ChengXiang
Department / Program:Computer Science
Discipline:Computer Science
Degree Granting Institution:University of Illinois at Urbana-Champaign
Degree:M.S.
Genre:Thesis
Subject(s):News
Adjusted Document Frequency-Inverse Document Frequency (ADFIDF)
Market prediction
Topic model
Abstract:News can influent the market. It has been proven that using text mining techniques, financial news can be used to predict market trend and volatility. In this thesis, we study some existing prediction algorithms that are based on Naive Bayes Classifier and Adjusted Document Frequency-Inverse Document Frequency(ADFIDF) weighting. Although occurrence of features selected by ADFIDF weighting can usually represent volatility bursts in financial market, it has been unclear whether it is also effective for market trend or trading volume. We conduct experiments that test the effectiveness of ADFIDF feature selection algorithm in finding correlation between news, market volume and trend. Our experiement result shows that a thin positive correlation exists between ADFIDF features occurrence and market volume. However, features occurrence and market trend are not directly correlated. We also propose a novel algorithm of finding correlation between news and financial market. The proposed algorithm is based on topic model and adjust TF-IDF weighting. It allows us to identify a few factors that could influence the performance of a prediction algorithm, such as number of topics of a model and adjustment. of IDF value. Our experiment results also show that grouping stocks together does not necessarily improve the performance of a prediction algorithm.
Issue Date:2012-09-18
URI:http://hdl.handle.net/2142/34447
Rights Information:Copyright 2012 Jing Zou
Date Available in IDEALS:2012-09-18
2014-09-18
Date Deposited:2012-08


This item appears in the following Collection(s)

Item Statistics