In recent years, mobile health recommendations are used in an increasing number of applications. Researchers have highlighted the importance of explaining these recommendations to lay users, with benefits such as increased trust and a higher tendency to follow up on these recommendations. However, a diferent explanation modality can impact the way users perceive the recommendation, either in a positive or negative way. This paper will explore and evaluate six diferent explanation designs through a qualitative user study, and give general design guidelines and considerations regarding explaining pain-related health recommendations to lay users.
As highlighted earlier, adding explanations to
recommendations can improve the overall efectiveness. These
make the system interpretable, which in turn can
improve trust towards the system [
the expertise of end users should be taken into account
when designing explanations. Ribera et al. [
Thus we see that interpretability through explanations
has multiple benefits and can result in an increased trust
towards the system. However, as previously mentioned,
the adoption of explanations in HRS is still low.
Furthermore, most health-related AI explanations are being
researched with AI and domain expert users in mind [
With these considerations in mind, we investigate the following research questions:
explaining health recommendations and why? RQ2 What design considerations are substantial when explaining health recommendations to lay users?
As mentioned in section 1.1, we will focus on designing diferent explanations that will explain why users are receiving specific recommendations for their pain flareups. Keeping the context and type of end users in mind, the following design guidelines have to be kept in mind for all variants of explanations: • Mobile-friendly: as the explanations will be ofered within the context of the mobile health app, the explanations have to be well-suited for display on a small mobile screen. • Summative: the explanations should possess the ability to summarise categorical data, as input consists of (semi-)unstructured user input. • Suited for non-experts: as the end users are non-experts, the explanations should not use any advanced and statistical concepts to explain why the recommendation is suggested.
Keeping these criteria in mind, we came up with the
following designs in Figure 1 based on well-known and
widely used explanation types:
• Text-based: briefly explain why the
recommendation is related to the most prevalent input. The
wording is based on the "communicating
healthrelated news to patients" guidelines described by
[
pool of 286 people who were already using the mobile health coaching application without the pain logbook
These explanation designs are sorted from least to most and its recommender system, as mentioned in section 1.1, by the amount of information they convey regarding the and thus knew and have interacted with the content and inputs relevant to the recommendation. The textual ex- diferent modules. The group consisted of nine women planation only focuses on one input, with the inline reply and two men, of which four finished graduate school, six being able to also show which specific input triggered college, and one high school. Age-wise, 2 participants the recommendation, whereas the tags are able to dis- were between 21-30, 5 between 31-40, 3 between 41-50 play all relevant input categories that are related to the and 1 between 51-60. All 11 users noted to use the inrecommendation. The word-cloud further builds on this ternet on the regular basis, with 6 participants stating by also displaying the relative importance of each input to be average computer and IT users, and 5 participants related to the recommendation, and the FI shows the ex- stating to be advanced computer and IT users. act sorting of input according to importance. The added percentages give the most transparency regarding the inputs, by also displaying the exact values used by the underlying RS.
Insights through XAI (+) At the start of the study, users were briefed on the pur- Six users liked the fact that they were able to gain more pose and context of the think-aloud study, and gave their insight through this explanation modality. Four users consent to having the audio recorded, after which they also stated that the percentages were a “nice-to-know”, iflled in the ResQue demographics questionnaire [ 20]. making the explanation more useful and informative. Afterwards, they were guided through the pain logbook, which they had to fill in with recent pain-episode they Negative sentiment towards XAI (-) experienced in mind. Having done so, they received some information regarding the recommendations that On the flip-side, two users disliked the addition of displayare going to be given, along with the explanations. We ing percentages, stating that when it comes to emotions briefly went over the six explanation designs in a fixed and feelings, certain aspects are not quantifiable . U4 order, after which we asked the participant to “explain stated: “Personally I think feelings are not quantifiable. what they like or dislike about the explanation” sepa- The bars are good, but don’t put an exact number on it. It’s rately for each design once they have seen them all. To okay if you’re communicating frequencies, like how often conclude this preference elicitation, the users had to sort an emotion occurred for example.”. the explanations by preference, with 1 being their most preferred one, and 6 their least preferred. They also had Visual/information overload (-) to give (or repeat) a key reason as to why they are giving Two users also stated that the addition of percentages is each explanation a certain ranking. The audio recordings unnecessary, mentioning that only using bars to comof both the preference elicitation and ranking are used municate importances is suficient. afterwards for a thematic analysis.
Rank: 2 · The feature importance explanation was The thematic analysis was done in two phases, with the among the most preferred explanations, liked for the ifrst phase consisting of deriving granular themes from fact that is was able to give a summary of the user input the thematic analysis with two researchers, and the sec- ( = 11), as well as being able to give additional insights ond phase focusing on merging them to higher level ( = 2). themes with a third researcher. The resulting higher level themes are displayed in Figure 3, along with the frequencies in which they occur per explanation design. Provides summary (+) The agreement percentage of the first phase two-coder Six users found the feature importance bars to be a clear thematic analysis is 88.1%, with Cohen’s kappa being way of communicating input topics and their importance. = 0.66, resulting in a substantial inter-coder agree- Four users stated that it gives them a nice overview of ment [21]. their input.
Insights through XAI (+)
Taking the average raking scores of all explanation de- they were able to get from the feature importances. U4 signs, we are now able to rank the 6 explanation modal- mentioned: “There are of course no numbers given, but ities from best to worst ranked, along with the results I can assume that I am really frustrated, and a bit less from the thematic analysis to explain why each explana- angry. I find it interesting to reflect on results that come tion type scored poorly or adequately. Figure 2 shows out of a questionnaire.” the frequencies of the rankings given to each explanation design.
Negative sentiment towards XAI (-)
Rank: 1 (best) · This explanation type was favored by most users, mainly due to the fact that it provided the most insight and transparency ( = 10). Only three out of 11 people found the addition of the percentages to feature importance bars to be ineficacious.
stating that they agreed with the general content, but not as to why one topic was deemed more important over others. This caused these users to slightly dislike and distrust the system, and give it a lower ranking.
them information as to what contributed towards the recommendation, but not why, like the textual explanation did. U6 stated: “There is not a lot of background given. It shows that these inputs contributed to my recommendation, but not why.” Three users were fond of the additional insights they got from the tags and the general themes that were present in their input. U3 stated: “When inputting my feelings I did not necessarily perceive them as negative or angry. But based on these tags, I’m able to see: okay, this is how the app interprets my feelings.” 3.3. Tags Rank: 3 · Tags scored relatively better than the previous three explanations in terms of average ranking, and were liked for their summative ability ( = 8). Only people who disliked having a lot of information, were less in favor of the tag explanation ( = 2).
Visual/information overload (-)
provided too much information. U6 stated: “Yes it’s clear, but less practical. I tend to focus on one thing at a time.”
Provides summary (+) Rank: 4 · Purely textual explanations received mixed reactions during the think-aloud study. When users liked Four users found using tags to be a nice way of providing or agreed with the recommendation, the textual explanaa summary of their input. Four users also stated that tion was a welcome addition helping them understand the doing in such a way is a clear and concise method of recommendation process and the recommendation itself, explaining why the recommendation is given. and gave users a nice summary of why the recommendation matched their inputs ( = 8). However, when the recommendation wasn’t in line with the user’s expectations, the textual explanation highlighted the mismatch even more and caused a poor reception of the recom- Problem with representation (-) mender system in general ( = 5). Here is an overview of these topics: Only some minor and infrequent negative remarks were given surrounding inline replies. Three users disliked the fact that by highlighting or repeating their negative input, they are more confronted with it. One user additionally mentioned that this explanation feels like the recommendation is only tuned to one input instead of multiple user inputs, making it feel too specific .
Provides summary (+)
summarize their input quite well, albeit only focusing on one topic (the most relevant one) surrounding the recommendation.
Positive sentiment towards explanation (+) Rank: 6 (last) · The word cloud received the lowest avTwo users stated that the written explanation was con- erage score. In general, users like the addition of displayifrming and comforting . One user also stated that the ing keyword or topic importance, however using a word wording of the textual explanation felt less confronting cloud to do so proves to be an inferior solution. The theregarding their negative input. matic analysis points out two main negative themes as to why this explanation is disliked: problems with represenNegative sentiment towards explanation (-) tation and content ( = 9) and visual/information overload ( = 4) and one positive theme, insights through explanation ( = 4).
On the other hand, three users mentioned that they cannot relate to the recommendation, and that the textual explanation highlighted this fact. U4 also found the ex- Problems with representation (-) planation to also be provoking, stating the following: “I know that I’m frustrated and that it does not help. However, Three users pointed out having keyword size commuexplaining that acts like waving a red flag in front of a nicate importance was unclear, and would rather have bull.” something concrete like bars indicating exact relevance. Three users also pointed out that the inconsistent sizes 3.5. Inline-reply inherent to the design of word clouds were visually displeasing. Two users additionally stated highlighting Rank: 5 · During the think-aloud study, the inline reply important keywords might be too confronting with rereceived relatively positive feedback and comments re- spect to their own input, e.g. if a user inputs that they garding the succinct summary it gave of the users input are feeling sad, having it displayed as a large word might ( = 7), with only some minor remarks regarding the confront the user too much with their state of mind. presentation of the explanation ( = 3). However, it scored quite low during the preference ranking itself due Visual/information overload (-) to other explanation modalities simply being preferred over the inline-reply.
Provides summary (+)
more concrete, and one user additionally stated that showing which message triggered the recommendation requires less analysis from the user.
Insights through explanation (+) Three users liked the fact that the inline-reply raises awareness of the fact that the recommendation is related to one of their own inputs. U3 stated: “I find it better than the textual explanation. There, they state ’You seem to be frustrated’, and here you really are made aware of the fact that it’s your own input.“
in such a way unnecessary, one of which additionally stated that adding the information in such way is too distracting.
Insights through explanation (+)
of keyword relevance gives more insight due to not only showing the relevant topics, but their importance as well.
People experiencing (chronic) pain or illness can feel distress when receiving negative information surrounding their state. In our study, we noticed that highlighting keywords that are potentially negative (e.g. negative emotions, reactions, etc.), can cause distress with users and therefore make them dislike the explanation. This was apparent with the inline reply and word cloud explanations, where visually highlighting negative sentiments that relate to the recommendation caused users to dislike the explanation.
Due to the fact that tags and FI/FI+% are able to display multiple input categories, users positively expressed that this would provide them more control over the recommendation process, if the design or implementation allows for it. One user suggested that tapping certain topics could be useful to request recommendations in a more user-controlled way. Other users additionally suggested U9:“It’s nice if you can individually remove certain topics”, and U7: “... especially of you notice something that wasn’t interpreted the way you intended it”.
4.3. Design FI through a lay user’s Users generally liked the holistic approach of the feature
importances, and were more inclined to look into the
perspective recommendation itself. When asked why they liked the
The FI and FI+% designs were favored by most users, recommendations more when explained using FI
comgiving most users the insight and summary they needed. pared to the purely textual explanation, they stated that
However, as mentioned in section 3.2, U4 interpreted the the FI were able to show them a general overview of them
FI bars as “... I can assume that I am really frustrated, as a person.
and a bit less angry”, indicating that they saw it as an On the other hand, there were also some users who
disoverview of their input, and not how strongly their input agreed with the ordering of keyword importances that
relates to the recommendation. In total, 10 out of 11 lay the feature importance bars were displaying, causing
users interpreted FI diferently than intended. Only U4 a slight increase in distrust towards the recommender
was able to correctly interpret the bars (after reading the system, ranking the explanation lower. This is to be
extext above the FI bars - “This is how your inputs relate pected, as increasing transparency of explanations can
to the recommendation”), saying “The frustrated bar is cause a higher drop in trust towards the system if the
the biggest, okay, so that contributes most to my recom- content of the explanation or recommendation does not
mendation”. Having a wrong interpretation could lead to align with the user’s expectation. However, the efect
confusion towards the system when, for example, a next of a misaligned textual explanation is still stronger, as
recommendation is shown, and the input keywords and users who did not agree with either the recommendation
their relevance change with respect to this new recom- or the explanation expressed a more negative sentiment
mendation. However, overcoming biases and changing towards the recommendation, and gave the textual
recmental models of lay users often proves to be dificult. ommendation a lower ranking. This is in line with similar
A possible design adaptations to the FI and FI+% design, research by Balog et al. [
This paper introduced several explanation designs for mobile pain related health recommendations, and compared them among lay users. Most users preferred the added transparency that was provided by the tags and FI / FI+% designs, stating that it gave them a brief and clear overview of their input which helped them understand why they received certain recommendations. Another interesting aspect is the fact that designs should be careful with visually highlighting negative sentiments of users. Designs that did so, i.e. the inline-reply and word cloud, were received poorly by users. Lastly, we confirmed that lay users might interpret certain visual explanations differently than intended, yet still prefer them over others. Given their feedback, we presented an adapted design of the favoured FI / FI+% explanation to be in line with what lay users expect.
The qualitative aspect of this study was already able to point out several key aspects related to designing health explanations for patients experiencing chronic pain. However, a larger scale quantitative user study is needed to further investigate these results. One such aspect is the fact that some users preferred textual explanations over explanations that ofered more information. Investigating whether this correlates to the user’s need for cognition (NFC), and what its implications are, can prove to be an interesting research direction similar to the research of Millecamp et al. [22]. Another aspect is the fact that while most users disliked being confronted with their negative input, some did not mind. This could be related to the "warriors vs. worriers" research, in which some users experiencing chronic pain actually prefer being exposed to negative feedback so they could address it, and could prove useful for further research [23]. Future research should also consider other designs to explain health recommendations and elaborate design guidelines that can be used by researchers and practitioners in this exciting domain. In addition, an interesting further line of research is to personalise these explanations on-thelfy, based on interaction data of end-users. As in work of [24], clicks and hover interactions as well as eye gaze data can be considered for such personalisation.
1–10