Integrating multiple conflicting sources by truth discovery and source quality estimation

Multiple descriptions about the same entity from different sources will inevitably result in data or information inconsistency. Among conflicting pieces of information, which one is the most trustworthy? How to detect the fraudulence of a rumor? Obviously, it is unrealistic to curate and validate the trustworthiness of every piece of information because of the high cost of human labeling and lack of experts. To find the truth of each entity, much research work has shown that considering the quality of information providers can improve the performance of data integration. Due to different quality of data sources, it is hard to find a general solution that works for every case. Therefore, we start from a general setting of truth analysis at first and narrow down to two basic problems in data integration. We first propose a general framework to deal with numerical data with flexibility of defining loss function. Source quality is represented by a vector to model the source credibility in different error interval. Then we propose a new method called No Truth Truth Model(NTTM) to deal with truth existence problem in low-quality data. Preliminary experiments on real stock data and slot filling data show promising results.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2014

Author dc:creator

• Zhi, Shi

Contributors dc:contributor

• Han, Jiawei

• Truth Discovery
• Data Integration
• Data Quality

Statement dc:rights

• Copyright 2014 Shi Zhi

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/50493

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/50493