Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Beyond Memorability: Visualization Recognition

196 views

Published on

Beyond Memorability: Visualization Recognition

Published in: Design
  • Be the first to comment

  • Be the first to like this

Beyond Memorability: Visualization Recognition

  1. 1. Beyond Memorability: Visualization Recognition If Lin Michelle A. Borkin et al., 2015
  2. 2. 1 秒 的 記 憶 實 驗 容易被記住的作品 不容易被記住的作品
  3. 3. 延伸探討 • 探討如何有效記憶 • 人們會注意視覺化作品中的那個元素幫助記憶 • 給予更多閱讀時間,對視覺作品的記憶會有什 麼差異? • 探討如何有效傳達訊息 • 會使用什麼資訊或元素辨識視覺化作品? • 會如何回憶描述一個視覺化作品?
  4. 4. 實驗用視覺化作品 隨機從中選出393篇作品
  5. 5. 實驗用視覺化作品 • 四大類,政府、科學期刊、資訊圖表、新聞 • 政府的作品,標示(label)元件最少。 • 政府作品幾乎沒有human recognizable objects (HRO) • 科學類在額外添加資訊(數據型或訊息型),幾乎沒 有。
  6. 6. 實驗設計 • 受試者33人(17女-16男),平均年紀22.9歲 (SD=4.2, 18-27)。無色盲。 • 隨機看393篇中的100篇。
  7. 7. 實驗設計THE EXPERIMENT “RECALL”“ENCODING” “RECOGNITION” Encoding Recall ON apple ~60 minutes EYE-TRACKING DATA Encoding Recall 10 seconds / image Recognition Recall2 seconds / image EYE-TRACKING DATA Recognition Recall20 minutes TEXT DESCRIPTIONS
  8. 8. 實驗設計 • Encoding階段,每張看10秒, 兩張中間間隔半秒。 • 共看一百張,每10張可選擇休 息。 • 眼球追蹤數據。 THE EXPERIMENT “RECALL”“ENCODING” “RECOGNITION” Encoding Recall ON apple ~60 minutes EYE-TRACKING DATA Encoding Recall 10 seconds / image Recognition Recall2 seconds / image EYE-TRACKING DATA Recognition Recall20 minutes TEXT DESCRIPTIONS
  9. 9. 實驗設計THE EXPERIMENT “RECALL”“ENCODING” “RECOGNITION” Encoding Recall ON apple ~60 minutes EYE-TRACKING DATA Encoding Recall 10 seconds / image Recognition Recall2 seconds / image EYE-TRACKING DATA Recognition Recall20 minutes TEXT DESCRIPTIONS • Recognition(辨識)階 段,同樣的100張,外加 其他100張不同的作品, 隨機交互出現,每張出現 2秒,間隔0.5秒。 • 如果受試者認出是曾出現 過的作品,就按下空白鍵 (HIT, Hit Rate)。 • 每20張後,受試者可以 選擇休息,而大多都沒休 息。 • 眼球追蹤數據。
  10. 10. 實驗設計THE EXPERIMENT “RECALL”“ENCODING” “RECOGNITION” Encoding Recall ON apple ~60 minutes EYE-TRACKING DATA Encoding Recall 10 seconds / image Recognition Recall2 seconds / image EYE-TRACKING DATA Recognition Recall20 minutes TEXT DESCRIPTIONS • Recall階段(再描述),給予受試 者再看上一階段他們正確辨識 出的作品。 • 這些作品會被隨機置放集合起 來。 • 給予20分鐘時間,以文字描 述,視覺化作品的內容。
  11. 11. 實驗設計THE EXPERIMENT “RECALL”“ENCODING” “RECOGNITION” Encoding Recall ON apple ~60 minutes EYE-TRACKING DATA Encoding Recall 10 seconds / image Recognition Recall2 seconds / image EYE-TRACKING DATA Recognition Recall20 minutes TEXT DESCRIPTIONS
  12. 12. 實驗設計-數據分析 每一份視覺化作品的內容分類 Table 1. The visualization labeling taxonomy used to annotate our target visualizations. The data subtypes taxonomy is taken from [8]. LABEL [OPTIONAL SUBTYPES] DESCRIPTION Annotation [Arrow] Outline of any visual elements annotating the data. A specific subtype of “arrow” was included to denote whether the annotation was in the form of an arrow. Axis [Time] Outline of where an axis is located including any tick marks and numeric values along the axis. A specific subtype of “time” was included to denote an axis involving time. Data Outline of where the full data plot area is located (e.g., the area between the x-axis and y-axis in a 2D plot). Data (type) [Area, Bar, Circle, Diagram, Distribution, Grid & Matrix, Line, Map, Point, Table, Text, Trees & Networks] Outline of where the actual data values are visually encoded (e.g., bars in a bar graph, points in a scatterplot, etc.). Graphical Element Outline of any visual elements that are not related to the visual representation or description of the data. Legend Outline of any legends or keys that explain the data’s visual encoding (e.g., color scales, symbol keys, map legends, etc.). Object [Photograph, Pictogram] Outline of any human recognizable objects (HRO) in the image. Objects are either realistic in representation (photograph) or abstract drawings (pictogram). Descriptions of each object were also recorded. Text [Axis Label, Header Row, Label, Paragraph, Source, Title] Outline of any text in the image. Subtypes cover all the common representations from prose to labels. GENDER EQUALITY IN LABOR FORCE PARTICIPATION ORIGINAL RATIOOFFEMALETO MALE 0% 25% 50% 75% 100% CHINA KOREA JAPAN INDIA Source: Gender Statistics 2013, World Bank GENDER EQUALITY IN LABOR FORCE PARTICIPATION DATA REDUNDANCY RATIOOFFEMALETO MALE 0% 25% 50% 75% 100% CHINA KOREA JAPAN INDIA Source: Gender Statistics 2013, World Bank GENDER EQUALITY IN LABOR FORCE PARTICIPATION CHINA LEADS IN FEMALE LABOR FORCE PARTICIPATION WHEREAS INDIA LAGS SIGNIFICANTLY BEHIND IN 2013 MESSAGE REDUNDANCY GENDER EQUALITY IN LABOR FORCE PARTICIPATION DATA & MESSAGE REDUNDANCY CHINA LEADS IN FEMALE LABOR FORCE PARTICIPATION WHEREAS INDIA LAGS SIGNIFICANTLY BEHIND IN 2013. CHINA INDIAJAPANKOREA RATIOOFFEMALETO MALE 0% 25% 50% 75% 100% CHINA KOREA JAPAN INDIA RATIOOFFEMALETO MALE 0% 25% 50% 75% 100% CHINA KOREA JAPAN INDIA CHINA INDIAJAPANKOREA Source: Gender Statistics 2013, World Bank Source: Gender Statistics 2013, World Bank Fig. 3. Illustrative examples of data redundancy (i.e., additional quanti- tative encodings of the data) and message redundancy (i.e., additional qualitative encodings of the main trend or message of the data). More examples are provided in the Supplemental Material. such as color, size, or opacity to represent a value already exhibited in a visualization such as the x- or y-axis values. In contrast, a visualiza- tion exhibits message redundancy if the main conclusion or message of the visualization is explicitly presented to the viewer in multiple 0% 25% 50% 75% 100% News media Government Science Infographic 64% 78% 54% 58% Percent of Visualization covered with Data PercentofVisualization Publication Venue Fig. 4. Percentage of visualization average pixel area covered by the data label. Scientific visualizations on average had the highest percent- age of image area devoted to data presentation. sualization type, we see that diagrams, maps, and tables cover a larger percentage of the image area than other visualization types. These types of visualizations tend to have annotations and text labels incor- porated into the data representation itself, thus requiring less of the image area around the data plot for additional explanations. Another observation is the difference in the average total number of elements in a visualization across sources. Visualizations from government sources have on average 11.9 labeled elements per visu- alization, significantly fewer compared to other visualization sources (p < 0.001,t(177) = 4.79 when compared2 to the numerically closest visualization source, science). In contrast, visualizations from info-
  13. 13. 實驗設計-數據分析 • 註解(Annotation)、 • 圖軸線(Axis)、 • 數據圖(Data)、 • 數據圖種類(Data type, 依據前一篇的分 類)、 • 與數據無關的視覺元件(Graphical Element)、 • 圖例(Legend)、 • 可辨識的具體圖示(Object, 影像與圖示、 公司Logo、國旗等)、 • 文字(Text)。 每一份視覺化作品的內容分類
  14. 14. 實驗設計-眼球數據分析 • 每一份視覺化作品的內容分類 • 計算每類別眼球停留的次數,如果眼球停留位置 包含兩個以上類別,則每個類別都算一次。 • Encoding跟Recognition階段都加入計算。 • 眼睛停留時間超過15毫秒
  15. 15. 實驗設計-辨識率 • 計算Recognition階段的空白鍵Hit Rate, • 計算方程式:HR = Hits / (Hits+Misses) • 數 介於0到1
  16. 16. 實驗設計-文字描述分析 • Recall階段的文字描述品質分數 • 找三位文字專家審 全部共2249個描述 • 受試者如何再描述視覺化作品所傳達的內容與訊息,也代 表說作品傳達給讀者訊息的效率。 • 分數為0到3分 • 0代表不正確或語無倫次的描述、 • 3代表文字描述有談論的作品的主題與主要訊息、由提 到作品中得數據與資訊、以及特別的額外細節等。
  17. 17. 10秒後辨識結果與1秒結果比較 a) Visualizations sorted by “at-a-glance” HR 0 50 100 150 200 250 300 350 400 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Images sorted by AMT HR HR new HR AMT HR HR Raw HR for prolonged exposure study Standard deviation for raw HR points b) c) MEMORABILITY ACROSS DIFFERENT ENCODING DURATIONS LO QUA DESCR Visualizations sorted by “at-a-glance” HR 0 50 100 150 200 250 300 350 400 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Images sorted by AMT HR HR new HR AMT HR “at-a-glance” exposure semantic associations visual associations no distinct visual or semantic associations prolonged exposure HR newHR AMTHR Raw HR for “at-a-glance” study [8] Fitted HR for prolonged exposure study 記憶程度:閱讀時間越多,記得越好。
  18. 18. 10秒後辨識結果與1秒結果比較 • 10秒的記憶正確率增加,平均79.70%正確率,1秒 為 55.61%正確率 • 不同視覺化作品間的相對分數(HR)沒有變化。 • Spearman rank correlation 斯皮爾曼等級相關係 數 • 1秒與10秒結果間為顯著正相關0.62 (p < 0.001), 表示各視覺化作品,在兩場實驗中的結果是相似。 記憶程度:閱讀時間越多,記得越好。
  19. 19. 10秒後辨識結果與1秒結果比較 a) Visualizations sorted by “at-a-glance” HR 0 50 100 150 200 250 300 350 400 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Images sorted by AMT HR HR new HR AMT HR HR Raw HR for prolonged exposure study Standard deviation for raw HR points b) c) MEMORABILITY ACROSS DIFFERENT ENCODING DURATIONS LOW QUALITY DESCRIPTION M Visualizations sorted by “at-a-glance” HR 0 50 100 150 200 250 300 350 400 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Images sorted by AMT HR HR new HR AMT HR “at-a-glance” exposure semantic associations visual associations no distinct visual or semantic associations prolonged exposure HR newHR AMTHR Raw HR for “at-a-glance” study [8] Fitted HR for prolonged exposure study 所以會比較好記憶的作品,不論1或10秒都是容易記憶。 不好記得作品,就是不好記。
  20. 20. 那麼 最容易與不容易辨識(記憶)得作品 他們之間的差異是如何?
  21. 21. RECOGNITIONENCODING MOST RECOGNIZABLE LEAST RECOGNIZABLE the most and least recognizable visualizations from [8]. TOP: Eye-tracking fixation heat maps (i.e., averag om the encoding phase of the experiment in which each visualization was presented for 10 seconds. T exploration of the visualization. BOTTOM: Eye-tracking fixation heat maps from the recognition phase o
  22. 22. • Recognition階段中,最易辨識的作品,眼球追蹤 結果,大多集中在中間。這樣就足以提供 多的 內容供辨識。 • 最不易辨識的作品,Encode階段與Recognition 階段的眼球移動模式類似。 (2SEC) RECOGNITIONENCODING MOST RECOGNIZABLE LEAST RECOGNIZABLE
  23. 23. to quantify the difference more generally between the fixation patterns for the                        able and bottom 1/3 least recognizable visualizations “at­a­glance” [8], we calculated                      recognition fixation distance viewed away from the image center, the spatial variances                        ocations, and the number of distinct loci fixated during recognition.   calculated the average distance from the visualization center at which participant                      d the Euclidean distance between the center of the visualization and each of a participan                            on the visualization, and averaged over all of these distances. Participants on averag                        m the center for the least recognizable visualizations (208 pixels) than for the most                          RECOGNITIONENCODING MOST RECOGNIZABLE LEAST RECOGNIZABLE
  24. 24. • 最不易辨識的作品,眼球會做更多的移動探索幫 助回憶。這時視覺化物件已經不足以幫助回憶, 所以需要去探索文字訊息以幫助回憶。 • 所以不同視覺化物件,對記憶的幫助能力不同。 (2SEC) RECOGNITIONENCODING MOST RECOGNIZABLE LEAST RECOGNIZABLE
  25. 25. 那麼 那一類的視覺化物件 能幫助記憶
  26. 26. 兩種類物件連接記憶 Visual Associations Semantic Associations Human Recognizable Objects Visual annotations FOR EXAMPLE... FOR EXAMPLE... Titles Labels Visualization/plot (bars) 100% Percent Bins 79% (bars) TITLE Captions are helpful. Textual descriptions Axis labels Visual Associations 視覺連接 Semantic Association 語義連接
  27. 27. 兩種類物件連接記憶 • 容易辨識的前三分之一作品, 有74%有人可辨識的具體圖示 (HRO) • 相對地,不易辨識的前三分之 一作品只有8% Visual Associations 視覺連接 Semantic Association 語義連接 • Recognition辨識階段,眼球最 常會停留在標題(title),標題提 供語義連接回憶。 • 一般會先尋找視覺化物件幫助回 憶,當沒有足 幫助回憶的視覺 化物件,就會去尋找語義物件幫 助回憶
  28. 28. 兩種類物件連接記憶 • 1秒與10秒的結果,不易辨識的作品大多一樣。 • 不易辨識的作品,缺乏足 強度的視覺與語意的 連接回憶物件。 • 政府作品最多不易辨識。 • 政府作品太常用相同的模板與美學模式。所以容 易造成受試者在不同作品間混淆。(缺乏強度)
  29. 29. Recall - 再描述 • 資訊圖表類別作品被描述次數最多、描述的字詞也最多、 描述內容品質也最好。
 (代表訊息最成功傳達) • 資訊圖表類別作品也最容易被辨識
  30. 30. Recall - 再描述 • 政府作品最少被描述、描述字詞也少、描述內容 品質也不好,與科學類別相似。 • 政府作品也不容易被辨識
  31. 31. Recall - 再描述 • 描述品質好壞與容不容易辨識有相關。容易辨識 的作品,在後續描述的品質也最好。
 (容易辨識的作品,訊息也能傳達的最好)
  32. 32. HIGH QUALITY DESCRIPTION LOW QUALITY DESCRIPTION MEMORABLE FORGETTABLE
  33. 33. 描述品質好壞與容不容易辨識的關係? 視覺化作品內物件 如何影響記憶與理解
  34. 34. 眼球追蹤代表 人從那些物件取得記憶與理解 Recall再描述代表 人對視覺化作品理解程度
  35. 35. 標題 • 眼球追蹤停留的比例,標題最高 • 四類中資訊圖表類別的作品,比例更高 • 科學的作品,大多沒標題,
 這可能可以解釋為什麼科學類作品的再描述分數很低
  36. 36. 標題 • 有標題作品的再描述分數,平均是1.90, • 無標題作品是再描述分數,平均是1.30。 • 統計上有顯著差異。
  37. 37. 標題 • 標題放在作品上方位置,會比放在底部,有較多 的眼球停留數。 • 政府作品73%把標題放底部。   In Table S1 we include some statistics about where a title was most commo                          publication source categories. In Table S2 we calculate how frequently titles                    participants and all visualization in a specific category) and how often they wer                        where the title was located. In Table S3 we include the average title length across                            source category.     Table S1: Title Placement: most visualizations had the title locat                  visualization. Government visualizations had the most number of visua                the bottom.      Title on Top  Title on Bottom  Infographics  85.9%  4.4%  News  96.7%  1.6%  Government  82.0%  16.0%  Science  2.5%  0.0%    眼球停留數
  38. 38. 標題 • 標題最常被注視,且在頂部的標題較常被再描述。 • 所有文字元件,標題最重要。標題內容的好與壞 會影響Rcall的再描述品質。
 
 (也就是傳達效率的好壞)
  39. 39. 標題 than for one that didn’t, which points to the ease (and correspondingly, difficulty) of encoding and retrieving                                  these visualizations from memory.  However, not all titles are equally effective. Below we provide examples of visualizations that are at the                                  high/low extremes of recognition HR, and quality ratings. The examples are split by “good” (i.e., had a large                                    fraction of participants mention or use the title in their recall text description and received high quality scores)                                    and “bad” (i.e., unutilized titles and received low quality scores):    GOOD TITLE EXAMPLES:      Title: “Top 10 Most Read Books in the World”  Text: “Based on number of books printed and  sold over the last 50 years. Some titles may  have had more copies printed than some of  these books, but a vast number of those  books were not sold, so we'll assume that  they did not get read.”  Source:  http://visual.ly/top­10­most­read­books­world  “The most read books globally. I think it went Holy Bilbe                      something Harry Potter Lord of the Rings The Alchemist .... and                      The Diary of Anne Frank was last”    “Most popular/ widely read books for example the Holy Bible has                      the most worldwide readers followed by Harry Potter Lord of the                      Rings Twilight and other books (not necessarily in that order)”    “Most read books. Holy bible. Quotes of Mao Ze Tung. Harry                      Potter.”    “The most read books in the world. The bible is the first. The                          Lord of the Rings the Twighlight Saga the Davinci Code are also                        represented.” 
  40. 40. 標題 Source:  http://visual.ly/did­romney­outperform­obama ­debate­1    “66% of Americans believe Romney did better in the debate is                      what the quote on the left top reads. Democrats believed Obama                      did better with a small majority. Other groups were also                    considered around 4 total. Fox News viewers may have been one                      category. Obama pic on left lower Romney on right”      BAD TITLE EXAMPLES:    Title: “Cities”  Text: “2010, % change on previous year,  forecast”  Source:  http://www.economist.com/blogs/dailychart/2 010/11/global_cities    “Countries list employment and GDP?”    “ranking of cities by growth in employment? istanbul was at the                      top”    “Some countries have gotten better on this metric some worse.                    More better than worse though.”    “Increase in employment rate and GDP(?) of different cities                  internationally.”    “Something to do with the economies of developing nations and                    useable income vs taxed income”      15 
  41. 41. 數據圖表 Pictograms • 眼球停留第二多。 • 有圖表的Recall再描述品質,比沒圖表顯著好
  42. 42. 其他元件 • 其他眼球停留較多的元件 • 簡短文字、圖例、表的列標題 • 標題、標示、短訊息的內容,最常被Recall的描述 時被使用。 • 科學類作品的Recall描述標題,期刊本來就沒有, 所以受試者的描述則變成標示(註解)、軸線標示、 圖例為參考。
  43. 43. 額外標示 Legend Outline of any legends or keys that explain the data’s visual enc Object [Photograph, Pictogram] Outline of any human recognizable representation (photograph) or abstract drawings (pictogram). D Text [Axis Label, Header Row, Label, Paragraph, Source, Title] Outl representations from prose to labels. GENDER EQUALITY IN LABOR FORCE PARTICIPATION ORIGINAL RATIOOFFEMALETO MALE 0% 25% 50% 75% 100% CHINA KOREA JAPAN INDIA Source: Gender Statistics 2013, World Bank GENDER EQUALITY IN LABOR FORCE PARTICIPATION DATA REDUNDANCY RATIOOFFEMALETO MALE 0% 25% 50% 75% 100% CHINA KOREA JAPAN INDIA 34% 69%70% 82% Source: Gender Statistics 2013, World Bank GENDER EQUALITY IN LABOR FORCE PARTICIPATION CHINA LEADS IN FEMALE LABOR FORCE PARTICIPATION WHEREAS INDIA LAGS SIGNIFICANTLY BEHIND IN 2013 MESSAGE REDUNDANCY GENDER EQUALITY IN LABOR FORCE PARTICIPATION DATA & MESSAGE REDUNDANCY CHINA LEADS IN FEMALE LABOR FORCE PARTICIPATION WHEREAS INDIA LAGS SIGNIFICANTLY BEHIND IN 2013. CHINA INDIAJAPANKOREA RATIOOFFEMALETO MALE 0% 25% 50% 75% 100% CHINA KOREA JAPAN INDIA RATIOOFFEMALETO MALE 0% 25% 50% 75% 100% CHINA KOREA JAPAN INDIA 34% 69%70% 82% CHINA INDIAJAPANKOREA Source: Gender Statistics 2013, World Bank Source: Gender Statistics 2013, World Bank Fig. 4. Per data label. age of imag 數 據 式 額 外 標 示 訊 息 式 額 外 標 示 混 合 額 外 標 示
  44. 44. 額外標示 • 作品中有額外標示,在Recall時,再描述的品質較好。(傳達效果較好) • 再描述品質最好的前三分之一作品,
 57%有訊息額外標註,34%有數據額外標示。 • 再描述品質最差的前三分之一作品,
 22%有訊息額外標註,12%有數據額外標示。
  45. 45. 額外標示 • 資訊圖表類有較多的額外標示 • 額外標示,與作品容不容易辨識及再描述品質有關。 • 額外標示能幫助閱讀者抓住作品的主要訊息,且改善作 品的辨識與再描述品質。
  46. 46. 總結 • 容易辨識的視覺化作品,內容與細節也容易回想。
 描述內容也較好 • 標題的內容會顯著影響,人們接收作品的訊息的狀況, 以及再描述的狀況。 • 數據圖也能幫助腦中有效地回憶,也能幫助傳遞訊息 • 額外標示能使作品更容易被再描述與理解。
  47. 47. 對研究的評論與批判 • 每個受試者實驗100張作品,
 但不是100張都相同。 • 每個受試者的背景,學經歷不同。
 (記憶與再描述能力不同) • 隨機取樣(X) 
 系統隨機取樣(?)
  48. 48. HIGH QUALITY DESCRIPTION LOW QUALITY DESCRIPTION MEMORABLE FORGETTABLE ? ? 每個受試者的背景,學經歷不同。(記憶與再描述能力不同)
  49. 49. 生態演化領域專業者 與 非該領域者 理解與描述能力不同
  50. 50. 「謝謝。」

×