CHAPTER 5. PORT CANCER WEBSITE: DESIGN AND IMPLEMENTATION
5.3 Implementation of emotion-based personalisation on the PORT website
5.3.2 Emotion-based adaptation: Algorithm for cancer website emotion-based
The approach of mapping rules was adopted from Germanakos et al. [33]. They have used emotional parameters (e.g., anxiety level) as implications of the type of adaptation that would be applied to an e-learning environment. In their model, emotional processing parameters (i.e., high, medium, low) affected additional navigation support and aesthetics of a web page. Whereby, for example, high anxiety level implied triggering the adaptation rules that increased the size and weight of the font of the e-learning content. The rules for algorithmA3 (Figure 5.14) were extracted from the significant correlations found in Study 2-Study 4 (see Appendix M) and Study 4 classification rules (see Chapter 7, Section 7.6.5.1). The combination of these findings suggests that certain emotions experienced at the start or during website use predict a preference or dislike for specific website features. For example, shame was shown to be negatively correlated with the feature filtering recommended content on user profile page. The correlation is addressed in A3 with the following type of a rule:
IF shame is felt at login, THEN the user will be directed away from the filter on the profile page, by hiding it or making it less visible to the user.
ELSE, IF shame is not felt, THEN the feature is highlighted, to direct the user to it. The functions HighlightFeature() and HideFeature() apply the two types of adaptation, and are explained later in this section. The A3 algorithm utilises the emotions reported
both at login and during the website visit. Emotions reported at different stages of website use (starting, dominant and end emotion) have been explored in emotion-based contextual recommenders [186]. In this respect, the A3 algorithm is the most advanced of the three emotion-based algorithms, as it considers login and during use emotions, and individual and combinations of emotions.
Thereby, this algorithm represents the ultimate achievement of this research, incorporating emotions reported at any stage of the website use to trigger adaptation. Unlike the content recommendations algorithms (i.e., A1 and A2), in A3 if the user
134
chooses not to report their emotions (i.e., ignores or closes the Emotion Tool), no adaptation occurs and the default appearance of the website is presented (A3: step 3). Nevertheless, even if no emotions are reported, forms of website adaptation occur during website use, depending on user activities. For example, adaptability resulting from user customisation of text font, or adaptivity triggering the automatic change of website background colour after the user defined their favourite colour. On the other hand, if the user does report their emotions, the adaptation that is triggered is either hiding or highlighting features (explained in the next section).
A3 is triggered by user login. At login (A3: step 2) the algorithm tracks the number of reported emotions (step 3). If no emotions are reported, a default website is presented (step 3), and the system moves on to tracking every use of the Emotion Tool during that website session. Otherwise, if login emotions are reported, the algorithm checks each
discrete emotion and its intensity (A3: step 5), as well as the combinations of reported emotions (A3: step 11), to determine whether and which adaptation to apply (A3: step 7 and step 13). A list of the detailed adaptation rules showing the conditions that have to be met for the discrete emotions (or combinations of emotions) and the type of adaptation they trigger, is provided in Appendix F. This set of predefined adaptation rules is recorded in the system, and is based on the findings (correlations and classifications) of this research (see Chapter 7). Examples of the rules are presented later in this section. The algorithm checks for each discrete emotion reported at login (A3: step 5) whether a predefined adaptation rule exists for it (A3: step 6). If the function
AdaptationApplicable() returns a positive value, it indicates that the specific login emotion predicts a (dis)like of one or more of personalisation features. Depending on the intensity of that emotion, the ApplyAdaptation() function applies the highlighting or hiding type of adaptation to all the personalisation features which that emotion affects.
135
Figure 5.14. Adapting website features to user emotions (pseudo code)
The algorithm also checks if more than one emotion was reported at login (A3: step 10). The CombinationOfLoginEmotions() function (step 11) goes through the adaptation rules that address the conditions which satisfy more than one discrete emotion (six such adaptation rules are recorded in the system – see Appendix F). The function
CombinationReported() checks if all the emotions in the predefined rules have been reported by the user, and whether the reported emotions’ intensities satisfy the adaptation rule’s conditions (A3: step 12). If the condition is met, the ApplyAdaptation() function
A3. Algorithm for emotion-based adaptation
Input: user log in data, login emotions and intensities, emotions and intensities reported during website use.
Output: changes to website layout and features’ appearance
1. For each log in do
2. For each login_EmotionTool_use do
3. If reported_emotions == 0, present default website and go to step 19 4. Else If reported_emotions ≥ 1
5. For each login_emotion do
6. If AdaptationApplicable(login_emotion) == Yes;
7. ApplyAdaptation(login_emotion, emotion_intensity, personalisation_feature);
8. end if 9. end for
10. If reported_emotions ≥ 2
11. For each CombinationOfLoginEmotions do
12. If CombinationReported(emotions, intensities) == Yes;
13. ApplyAdaptation(login_emotion, emotion_intensity, personalisation_feature);
14. end if 15. end for 16. end if 17. end if 18. end for
19. For each DuringUse_EmotionTool_use do 20. If reported_emotions == 0, go to step 35 21. Else If reported_emotions ≥ 1
22. For each du_emotion do
23. If AdaptationApplicable(du_emotion) == Yes;
24. ApplyAdaptation(du_emotion, emotion_intensity, personalisation_feature);
25. end if 26. end for
27. If reported_emotions ≥ 2
28. For each CombinationOfEmotions do
29. If CombinationReported(emotions, intensities) == Yes;
30. ApplyAdaptation(du_emotion, du_intensity, login_emotion, emotion_intensity,
personalisation_feature); 31. end if 32. end for 33. end if 34. end if 35. end for 36. end for
136
(A3: step 13), triggers the adaptation of those website features which were found to be affected when feeling that specific combination of emotions. For example one of the rules listed in Appendix F is:
IF at login sadness is moderate (3) or greater AND surprise is moderate or less, THEN highlighting adaptation is applied to the articles’ related content.
As explained at the beginning of this section, emotions reported during website use also trigger adaptation (A3: step 19), every time the Emotion Tool is used. The same process occurs as at login, whereby both individual discrete emotions (A3: step 22) and combinations of discrete emotions (A3: step 28) are matched to predefined adaptation rules. The difference is that the rules for combinations of emotions (A3: step 27 - 30) can
comprise both login and during use emotions (as illustrated by the two rules below). If the during-use emotion (or a combination of during-use and login emotions) triggers adaptation, the ApplyAdaptation() functions (A3: step 24 and 30) adapt the website features affected by those emotions. There are six adaptation rules for the during-use emotions (see Appendix F), of which two are conditions with two or more emotions, and these are:
IF at login surprise ≤ 2 AND during use joy > 1, THEN adapt the feature adaptive storyline by highlighting the background in green.
IF at login surprise > 1 AND during use surprise ≥ 3, THEN adapt Forum discussion recommendations by highlighting them.
Types of adaptation: The feature adaptation approach used on the PORT website is self- devised, based on adaptive hypermedia literature, including Brusilovsky’s well-known taxonomy [79], and on the existing systems which incorporate adaptive hypermedia, e.g. GRAPPLE [88]. As previously explained, two types of adaptation are applied to the PORT website features:
Hiding features – users are guided away from features; features are hidden by not presenting them at all on the website (e.g., the F1 feature in Appendix G) or showing them in a less visible format (e.g., F3 in Appendix G).
137
Highlighting features – users are guided to features; highlighted features are made more visible, their presentation stands out by using different or brighter colours than the ones commonly used on the website.
In hiding adaptation, whether feature removing or hiding is applied, depends on the respective features, i.e. the way removing information would affect the surrounding material, as well as based on the ease with which the information can be 'hidden in plain sight' (see Appendix G for the types of hiding adaptation). On the other hand,
highlighting adaptation was exclusively implemented by applying green colour as the
background or the font colour. This is based on the traffic light metaphor, whereby green
symbolises “go”, i.e. directing the user to action. Green, moreover, complements other colours of the website (white and purple), but also stands out compared to them. In feature highlighting, alternatively red and yellow could have been used, however both had drawbacks. Red was avoided, since it mainly implies a ‘stop’ action, or a warning. Orange and yellow were also avoided, since yellow is the main colour of the Emotion Tool, and both are less visible, depending on how the website is displayed. Note that Study 5 interviews confirmed this reasoning; users complimented the use of green colour in highlighting, and moreover claimed that red would not have been the desired choice, as it would not stimulate them to action, but rather cause confusion (Appendix L).
Based on the findings of this research (Appendix H and Appendix M), 25 different features and content types on the PORT website can be adapted to user emotions; the remaining features are not affected by adaptation. Appendix G presents the adaptation applied to the individual features. For example, highlighting is applied to the feature
User profile customisation (F1), whereby the notifications to go to the user profile and edit it are changed into brighter colours and the background for the Edit profile link is coloured in green. Another example of feature highlighting is displaying green borders around the Knowledge base recommendations (on KB pages) to guide users to this content.
The features are also adapted to guide the user away from them, if the emotions experienced by the user recommend that. The type of adaptation referred to here is called feature hiding. Hiding on the PORT website is applied to fewer features. For example,
138
the hiding applied to the feature user profile customisation results in all the notifications that remind the user to go to the profile and edit it to be turned off. Thus, the orange triangle in the upper right hand corner of the website and the red rectangle on the user profile page are removed. The user could still edit the profile, by reaching it via menus, but it would have to be a purposeful action on their behalf.
The second type of hiding adaptation is that applied to, for example, filtering recommended content on the user’s profile. In this case, only a link to the filter (the word
Filter) is presented; if clicked, the full filter is opened, visible as in the version without hiding. The third type of hiding is applied to recommendations, for example, article recommendations (on Articles pages), wherebythe background of the recommendations on top of Articles related pages is coloured in grey to blend with the grey website background, instead of the default purple colour.
Next, the implementation of the personalisation algorithms on the PORT website is demonstrated.