Towards using Segmentation-based Techniques to Personalize Mobility Behavior Interventions

This paper describes our initial work towards a segmentation-based approach to personalized digital behavior change interventions in the domain of sustainable, multi-modal urban transport. Segmentation is a key concept in market research, and within the transport domain, Anable has argued that there are segments of travelers that are relatively homogenous in terms of their mobility attitudes and behaviors. We describe an approach aimed at tailoring behavior change notifications by using segmentation-based techniques for user profiling. We report results from a Mechanical Turk study in which we obtained a crowd-sourced categorization of motivational messages. This is a first step towards understanding how to better deliver persuasive messages to relevant users profiles and situational contexts in the urban mobility domain. We conclude by discussing future steps of our work that should inform the deployment of persuasion profiling techniques to achieve sustainable mobility goals.


Introduction
We are investigating how to maximize the persuasion potential for nudging citizens towards more sustainable transport choices via an urban mobility platform based on mobile and web interfaces.In our earlier work, we discussed how behavior change theories can be integrated into a sustainable urban mobility platform [1,2].Building on the theories and techniques of Michie et al. [3,4,5] we were also inspired by the segmentation work of Anable [6,7].In this paper, we aim to combine the profiling tools of Anable's segmentation analysis with digital intervention techniques to deploy targeted digital interventions that prompt people to use more sustainable transport modes.The SUPERHUB project (http://superhub-project.eu) aims to do this by a combination of: i) prompting intention formation, ii) setting and reviewing specific goals; iii) providing monitoring, feedback, and rewards; iv) supporting (social) comparison; v) aiding decision-making.To improve behavior change interventions for urban mobility, our recent research has identified a considerable potential in exploring and experimenting with the following techniques:  Just-in-time prompts and notifications.Participants are prompted at appropriate times to change their behavior, for example to provide a lift or use public transport. Notifications or prompts are personalized according to user characteristics and context.Notifications to encourage behavior change are sent by taking into account domain-relevant user profiles and activities in order to raise their effectiveness in terms of user acceptance and persuasion.Previous work suggests that personalized or tailored messages are more useful than generic ones in promoting behavior change [8].Noar et al. [9] provides a Metaanalytic review of tailored Health Behavior Change interventions and provides evidence for the effectiveness of tailoring.Tailored messages may be especially useful when emotional arousal facilitates behavior change interventions [10,11].Research has shown that users show significant individual differences in their response to influence attempts [12,13,14,15].This suggests that using persuasion profiles (estimates of an individual user's susceptibility to different influencing strategies) to adapt persuasive systems should be considered [16].In this paper, we discuss how a segmentation-based approach can be used, where motivational messages are tailored to Anable's traveler segments.We also present the results of an initial study to select and categorize motivational messages.We finish by describing how this work will be extended to obtain an effective algorithm for selecting motivational messages based on the traveler segment a user belongs to.

A segmentation-based approach
The preliminary user research undertaken by large scale questionnaires and numerous focus groups early in our project found that different people have very different concerns regarding their travel choices.Some people are committed to the environment and will do what they can to reduce their carbon footprint.Others are less concerned and it will take a lot more persuading than simply showing CO 2 comparisons for them to make more sustainable travel choices.Anable [5] stated that travel research methodology and policy interventions often overlook how the combination of instrumental, situational and psychological factors affects travel choice and differs for distinct groups of people.Understanding what will motivate people to change their behavior is a key element of any successful intervention.For example, visualizing the amount of CO 2 produced over a year may work for some, whereas for others finding out the amount of money they could save by taking the bus rather than driving may be more motivating.Different people will respond more or less to different cues and this represents a major research challenge in understanding how to develop effective persuasive interventions for everyone, not just those already concerned about the environment.To enable a more tailored approach, we are considering the different types of 'traveler profile' proposed by the 'Segment' methodology developed by Anable [6] which deploys a version of psychological theory of attitude-behavior relations (Theory Planned Behavior -TPB) to score travelers on specific attitude statements.Anable [6] proposes eight distinct attitudinal segments, as shown in Table 1.The population segments are distinguished by their attachment to the car, self-identification with alternative travel modes and motivations for fitness and environmental protection.A series of so-called 'golden questions' were developed (answered via a Likert scale of 1-5, 1 being strongly disagree or very unlikely, 5 being strongly agree or very likely) to assign travelers to these segments.Example questions are: "I am not the kind of person who rides a bicycle", "I feel I should walk more to Towards using Segmentation-based Techniques to Personalize Mobility Behavior Interventions 3

Validation of motivational messages
This section describes our approach to categorize and validate motivational messages for use in personalized behavior change interventions.We are currently carrying out experimentation by utilizing Amazon's Mechanical Turk (MT) to gather crowd sourced intelligence about which kinds of message would be best suited to motivate Anable's defined traveler profiles.Initially we produced and selected a wide range of messages including quotations that we thought could motivate sustainable travel behavior in our users.This first MT study shows how we validated the messages to be used in each of the motivational categories shown in Table 2.We adopted the approach used by Dennis et al [18,19], who investigated the categorization of emotional support statements.

Participants
Participants were recruited from Amazon's Mechanical Turk service (MT, 2013), a crowd-sourcing tool.Participants (called workers) complete small tasks (called HITs) made available by requesters and are paid a small sum for completing the task successfully.For this validation experiment (HIT), participants had to be based in the US and have an acceptance rate of 90% (meaning that 90% of the work they do is accepted by other requesters as good quality) and were paid $0.70.We used a Cloze Test [20] for English fluency due to the language based nature of the study.Workers who failed the test were excluded.30 participants completed the experiment and were 27% male.24% of participants were 18-25, 43% were 26-40 and 33% were 41-65.The average time taken to complete the experiment was around 11 minutes.

Procedure
Participants were introduced to the categories and their definitions (as described in Table 2).Next, they were shown a message and asked to place it into one of the categories (still seeing the definitions), as shown in Figure 1.This was repeated for each of the 74 messages.Participants were advised that there were no right or wrong answers and that it was their opinion that counted.

Validation Measure
We use Free-Marginal Kappa [21] as a metric for establishing how well categorized our messages were.The kappa value describes agreement amongst raters, with 1 indicating unanimous agreement, 0.7 excellent and 0.4 moderate agreement.To be reliably categorized, the kappa score for the message had to be ≥ 0. keep fit", and "I feel a moral obligation to reduce carbon emissions".The SEGMENT project [17] has shown that these segments are common and workable across Europe.The proportions of the segments vary from country to country in relation to the value people put on status, cost, time, environment, social norms etc.

Results
Table 3 shows the messages with kappa ≥ 0.4.To decide on which messages to put forward to the next phase and potentially use as motivational messages during our next project trial, we selected those that had the highest kappa values and that had at least a kappa ≥ 0.4 (so, had an adequate level of agreement between the participants).The results showed that some categories obtained a much higher level of agreement between participants than others, for example, the 'positive aspects of cycling' agreement levels and therefore kappa values were very high, whilst other categories such as 'Advice on Sustainable Travel' and 'Sustainability Self-reflection' led to lower levels of agreement, and hardly any messages with kappa ≥ 0.4.This can be explained by the fact that some of the original messages had components of more than one category, for example, the following message "Did you know that many people use public transport at least once per week?" contains a positive statement about public transport and is also a social comparison statement, which effectively divided the participants choice between the two categories.Some refinement of the messages and categories has been made to enable clearer categorization for the next round of MT experiments.We have removed the 'Advice' category and any 'mixed message' messages will be replaced with less complex ones that provide a clearer persuasive message.A further iteration of the above procedure will be carried out with another 30 participants to allow validation of additional messages.

Conclusions and Future Work
In this paper, we have presented our early and ongoing work in designing persuasive notifications tailored to relevant travelers' profiles for behavior change interventions in the sustainable transport domain.We have reported work conducted to gather a corpus of 74 motivational messages and categorize them into 9 categories, with 45 messages being reliably categorized.The next step is to run a study which presents participants with each traveler segment profile and asks them to provide the most appropriate notifications using the validated messages.This will result in an algorithm that selects message categories (and messages) depending on the traveler segment.It would have been possible simply to select message categories and messages we thought would be appropriate for each of the segment types.However, by going through the process of crowd sourced intelligence for initially grouping messages by category and then choosing which messages (and hence, categories) would be most relevant to each segment type, we expect to obtain a more effective persuasive system.

Figure 1 .
Figure 1.Example of how messages were presented to participants during the MT study, message statement presented at the top of the page, participants select a category from the choices below.

Table 1 . Segmentation of people based on mobility attitudes, from Anable (2010).
Active Aspirers Would like to cut down on car use and agree that the bus can be quicker, but still see problems with using public transport.See themselves as cyclists and also regard walking as healthy.Have a high moral obligation to the environment and are highly motivated to use active modes of transport.Practical Travelers Only use the car when necessary and believe cars reduce quality of life.Enjoy cycling and will walk when it is more practical than cycling.See local pollution and congestion as is-sues but are not motivated by climate change.Have no intention of reducing car use or increasing Public Transport use.Car Contemplators See cars as status symbols and believe car use should be unrestricted.Would rather use the bus than cycle but see lots of problems with using public transport.Have a neutral attitude to the environment and are not motivated by fitness.Tend to be younger with the highest proportion of students.

Benefits of Cycling (BC)
"Think of bicycles as rideable art that can just about save the world."Grant Petersen 1 "Nothing compares to the simple pleasure of riding a bike."John F. Kennedy, 35th President of the United States 1 "Cycling is a joy and faster than many other modes of transport, depending on the time of day.It clears the head."David Byrne Did you know that traffic jams can be harmful for your health?The pollutants can get inside the car.
"Not having to own a car has made me realize what a waste of time the automobile is."Diane Johnson 0.48 EAI Endorsed Transactions on Ambient Systems 03 -10 2014 | Volume 1 | Issue 4 | e5 Towards using Segmentation-based Techniques to Personalize Mobility Behavior Interventions