Data used?

• for case Study
  • Amazon reviews (  5 diffrent products )
• Documents, sentences / phrase
• Diffrent formats ( .doc, .xml, ... )
• designed for a varietz of genres ( IN-SPIRE  Approach)

Analysis Technique

• Lexical Approach
  • compare doc. with a affect lexicon
  • lexicon derived from General Inquirer (GI)
  • semi-supervised bootstrapping (expand dictionaries)

Visualization Technique

• Multiple Views (inspired by IN-SPIRE)
• Galaxy View
• Two diffrent types of Rose Plots
  • large petals encode high affect score
  • included box plot into petal
  • petals are group (two affect categories)
  • unit circle shows deviation of expected value
• Histogram beneath each Ros Plot (number of doc.)
• Correlation Tool ( show distribution of products)

Problems

• no good overview visualization
• no aid to maintain mental map
• not scalable

Data used?

•  Case Study
  • academic papers (InfoVis VAST)
  • Product reviews ( Tv show, Cars)
  • newsgroup data ( 20News, well clustered)
• Input for t-SNE
  • High Dimensional Vectors

Application Pipeline

SS-NMF

t-SNE

Visualization/
Rendering

• two diffrent threads for t-SNE & SS-NMF
• allows real-time interaction of visualization

Analysis Technique

• Nonnegative Matrix Factorization (NMF)
  • advantages
    • consistency over multilpe runs
    • emipirical convergence
• t-SNE (modified)
  • reveal implicit groupings

Visualization Technique

• node-link diagram
• Continous visualization
  • real time interaction with Algorithm Output
  • simulation like behavior

Data Used?

• hotel  cutomer review data ( TripAdvisor )
  • Hotel Profile
  • Customer Profile
  • Review data

Analysis Technique

• Feature Based Opinion Mining
  • Free Text vs. Ratings
• New Concept of Uncertainty
  • reviews have additional value to positive / negative
• Opinion Combination
  • based on Subjective Logic

Visualization Technique

• Multiple Visualizations
  • Opinion Wheel
    • Opinion circle (Radial Visualization)
    • Opinion Triangle (Triangular Scatterplot )
  • Tag clouds
    • organized in a table

Data used?

• Medical Reports (Google Health)
• Article Structure (Paragraphs containing facets)

Analysis Technique

• new entity- relational data model introduced
• data needs to be transformed into model
  • extract multifaced entities (name entity recognition)
  • construct similarity graph (consine / topic-level similarity)
  • create new search indices (lucene)

Visualization Approach

Overall Visuall design

• visualize global and local patterns
• integrate search into visualisation (visual query)
• simply visualize multifacet relationships

Visualization Technique

• Density Map + Multifacet Graph
  • visual encoding of data model
  • visually simplify data exploration (edge bundling / highlighting)
  • interactions to examine data from diffrent perspectives

Data Used?

• Two diffrent Types of Data
• missing limited topic coverage
  • Web Surveys (50,000 Surveys)
    • all possible complaints ( purchase, delivery, product )
    • comment on features in detailed way
    • Cutomer Web survey
  • Rss News Items (16,000 items)
    • real-time distribution, repostings
    • no specific feature to indentify
    • explore spread of Opinion and reasons
    • 50 Presidential Election feeds

Analysis Technique

• no domain-dependant sentiment word list
  • rely on very general sentiment words
• Linguistic Preprocessing
• Feature Extraction
• Sentiment Word Detection
• Sentiment-to-Feature Mapping (Hybrid approach)
  • manually defined syntactic reference patterns
  • paired with typed-dependency parsers
  • modified distance-based mapping
  • awareness of uncertainty (not precise vs. guide user)

Visualization Technique

• Pixel Map Calendars - ( scatterplot )
  • display overall data distribution
  • one pixel represents one document
  • pixel ploted in bins - color encodes overall sentiment
  • axes of bin can have different time unit

• Time Density Plots
  • Sequential Sentiment Track - ( rectangular bars )
    • preserve temporal order
    • whitespace reduction - no overplotting
    • certainty encode in bar height
  • Time Density Track - ( Area Graph )
    • height of curved determinated by distance of two time stamps
    • high curves indicates max. Level of Interesst

Data Used?

• Amazon Customer Reviews  (3.168 Reviews)
• Only 4-5 stars ond 1-2 stars ratings
• about size of Scientific Abstract
• Diffuse Opinions in diffrent Reviews
  • based on Da Vinci Code book review

Analysis Technique

• Linguistic Variation  (TermWatch, Term Variation graph)
  • TermWatch - Clustering (CPCL implemented with SLME)
• Predictive Text Analysis  (Decision Trees - SVM )

Visualization Technique

• Term Variation Graph
  • display cluster overview & detail
  • time series data ploted in graph

• unfolded View of one cluster (pink Background)

• Time series Data ploted in Graph
• timestamps ploted on edge (year-month)

• Positive Term alongside Negative Terms
• monthwise separation

• Coordinated View - ( multiple Visualizations, Improvise )
  • Lexichron - Arc Diagram
    • top half positive reviews - bottom half negative review
    • arcs connect months with common Terms
    • arc thickness encodes # of common Terms
    • Bar thickness encodes # of Term per Month
  • Graph View
  • multicolumn Tables

Data Used?

• Broad Number of Application
  • Theme park attraction, product surveys, hotel reservations, IT services, movies...
• Web Surveys
  • Historic Data (after cutomer Purchase)
  • Tested with 52,189 Reviews (96,987 sentences)
• Tweets
  • Real-time (reaction to Movie)
  • limited to 140 charachters
  • tend to have heavy use of abbriviations
  • open-ended Data Sources (text streams)

Analysis Technique

• Feature-based Sentiment Analysis (use own algorithm)
• Term Association with Hypothesis Tests (Likelihood Ratio Test)
  • Focus on geo-spatial information ( geo-based Term Association )
  • Covers almost whole spectrum of frequencies ( Likelihood Ratio Test)

Visualization Technique

• Geo Maps
  • Pixel Sentiment Geo Map
  • Key Term Geo Maps
• Pixel-cell Based Sentiment Calendar
  • similar approach as in Pixel Map Calendars
  • row encodes diffrent feature    
  • organized monthly-year
• Self-organizing Term Association Map ( SOM )

• Pixel Sentiment Geo Maps
  • overlapping and data density
  • clustering and use of glyphs
  • disapperance of original data
  • Pixel Placment Algorithm  (Bresham - Midpoint)
  • avoid random patterns place by priority (sentiment)
  • radial layout of pixel placement

• Pixel Placment Algorithm
• Radial Layout

• Self-organizing Term Association Map ( SOM )
  • enriched version of Word Cloud with semantic context
  • grouped into associations
  • color encodes sentiment
  • saturation encodes strength of sentiment

Key Term Geo Map

Visual sentiment analysis of customer feedback streams using geo-temporal term associations

Visualization Technique

• Key Term Geo Maps
  • Key Term Geo Map
  • Key Term Distribution Map

• Key Term Geo Map
  • diffrent levels of detail
  • map key terms to zip code areas - ( not always possible )
  • change mapping between text & geo hierachy
  • Color, Size encoded accordingly ( respect overlaping )

• Key Term Distribution Map
  •  separate by sentiment ( negative vs. positive )
  • plot heat map on geo map ( Gaussian bluring func. )
  • overplot two heat maps
  • overlaped results
  • merge colors