Xiong, Wenting
(2015)
Helpfulness Guided Review Summarization.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
![[img]](http://d-scholarship.pitt.edu/style/images/fileicons/application_pdf.png) ![](http://d-scholarship.pitt.edu/22509/1.haspreviewThumbnailVersion/WentingXiong_ETD_12042014_final.pdf) Preview |
|
PDF (Final proof-read version on 12/04/2014)
Primary Text
Download (2MB)
|
Abstract
User-generated online reviews are an important information resource in people's everyday life. As the review volume grows explosively, the ability to automatically identify and summarize useful information from reviews becomes essential in providing analytic services in many review-based applications. While prior work on review summarization focused on different review perspectives (e.g. topics, opinions, sentiment, etc.), the helpfulness of reviews is an important informativeness indicator that has been less frequently explored. In this thesis, we investigate automatic review helpfulness prediction and exploit review helpfulness for review summarization in distinct review domains.
We explore two paths for predicting review helpfulness in a general setting: one is by tailoring existing helpfulness prediction techniques to a new review domain; the other is by using a general representation of review content that reflects review helpfulness across domains. For the first one, we explore educational peer reviews and show how peer-review domain knowledge can be introduced to a helpfulness model developed for product reviews to improve prediction performance. For the second one, we characterize review language usage, content diversity and helpfulness-related topics with respect to different content sources using computational linguistic features.
For review summarization, we propose to leverage user-provided helpfulness assessment during content selection in two ways: 1) using the review-level helpfulness ratings directly to filter out unhelpful reviews, 2) developing sentence-level helpfulness features via supervised topic modeling for sentence selection. As a demonstration, we implement our methods based on an extractive multi-document summarization framework and evaluate them in three user studies. Results show that our helpfulness-guided summarizers outperform the baseline in both human and automated evaluation for camera reviews and movie reviews. While for educational peer reviews, the preference for helpfulness depends on student writing performance and prior teaching experience.
Share
Citation/Export: |
|
Social Networking: |
|
Details
Item Type: |
University of Pittsburgh ETD
|
Status: |
Unpublished |
Creators/Authors: |
|
ETD Committee: |
|
Date: |
14 January 2015 |
Date Type: |
Publication |
Defense Date: |
5 August 2014 |
Approval Date: |
14 January 2015 |
Submission Date: |
24 July 2014 |
Access Restriction: |
No restriction; Release the ETD for access worldwide immediately. |
Number of Pages: |
168 |
Institution: |
University of Pittsburgh |
Schools and Programs: |
Dietrich School of Arts and Sciences > Computer Science |
Degree: |
PhD - Doctor of Philosophy |
Thesis Type: |
Doctoral Dissertation |
Refereed: |
Yes |
Uncontrolled Keywords: |
Text mining, Review helpfulness assessment, Peer review analysis, Review summarization |
Date Deposited: |
14 Jan 2015 19:19 |
Last Modified: |
15 Nov 2016 14:22 |
URI: |
http://d-scholarship.pitt.edu/id/eprint/22509 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
![View Item View Item](/style/images/action_view.png) |
View Item |