Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

Make your likes visible on Facebook?

Connect your Facebook account to Prezi and let your likes appear on your timeline.
You can change this under Settings & Account at any time.

No, thanks

SUR-Miner

No description
by

xiaodong gu

on 24 December 2015

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of SUR-Miner

"What Parts of Your Apps are Loved by Users?"
Xiaodong Gu and Sunhun Kim
Product Managers
are trying to listen to

users
System Overview
Evaluation
Split raw reviews into review sentences
Typo correcting
Tokenization, lowercase converting, stemming, etc

Step1 : Preprocessing
Step2: Classification
Classify reviews into different categories and select reviews for aspect evaluation
Review Categories
Classifier - Max Entropy
Feature Selection
- Lexicon based features
- Character N-gram
- Trunk word
- Structure based features
- POS tags
- Parsing tree
- Semantic dependence graph

Step 3 – Aspect-Opinion Extraction
Parse Review Sentences and Extract Aspect-Opinion Pairs
Pattern-based Parsing
Semantic Templates
Step 5 – Aspect Clustering and Summarization
Group aspect-opinion pairs with the same aspects
Summarize sentiments and typical opinions for each aspect group.
Aspect Heat Map
Aspect Trend Map
RQ1
(effectiveness): How effectively can SUR-Miner classify reviews, extract aspects and opinions, and analyze sentiments for app reviews?

RQ2
(comparison): How does the SUR-Miner compare
to state-of-the-art techniques for app review summarization?

RQ3
(usefulness): How the summaries by SUR-Miner
useful for developers?
Developer Web Survey
Comparison
Quantitative
Qualitative
Effectiveness

Challenge 1: Large review volume
~9000/day
~6000~/day
Challenge 2: Multiple review categories [Pagano RE'13]
Existing Review Mining Tools
LDA Topic model
Bags-of-words assumption
- Cannot distinguish review categories
- Cannot distinguish different aspects and corresponding opinions
- No sentiment analysis for each aspect
Step 4 – Aspect Sentiment Analysis
Analyze Sentiment for Each Aspect
The UI is nice, but the sound sucks
< UI , nice >
< sound , sucks >
THANK
YOU
But your users

are already talking
about you
Aspect - Opinions - Rating
Aspect - Opinions - Rating
Aspect - Opinions - Rating
.....
xguaa@cse.ust.hk
Xiaodong Gu
Sunghun Kim
hunkim@cse.ust.hk
The Problem?
There are a lot of them!
And they are talking about different purposes.
reviews
Parsing Tree
Semantic Dependence Graph
....
145 Questions most asked by developers [Begal ICSE'14]
SUR-Miner
< UI , nice >
< prediction , accurate >
< UI , not elegant >
< prediction , useful >
UI
Prediction
nice
not elegant
accurate
useful
17 apps, 16 categories, 2,000 review sentences per app
The user interface is not elegant
< user interface , not elegant >
topics by LDA
aspects & opinions by SUR-Miner
Classification
Aspect-Opinion Extraction
& Sentiment Analysis
[1]
Conclusion
A Review Summarization Framework
Review Classification
Pattern-Based Review Parsing
Sentiment Analysis for Aspects
Future Work
Mining other information from app reviews (e.g., feature request, bug report)
Mining from other text (e.g., bug reports, queries)
[1] Andrew Begel, Tomas Zimmerman. Analyze this! 145 Questions for data scientists in software engineering. ICSE'14
awsm
UI ! I
jst
wish more toolkits.
thx
!
Awesome UI!
I just wish more toolkits!
Thanks!
Online Demo
: http://www.cse.ust.hk/~xguaa/srminer
Online Demo:
http://www.cse.ust.hk/~xguaa/srminer
Apply Deeply Moving [2] to each review sentence.
[2] Deeply Moving: Deep learning for sentiment analysis. http://nlp.stanford.edu/sentiment/
Full transcript