ResearchPod Summary
As public transport systems become more complex, providing accurate, timely, and relevant information is critical for maintaining passenger satisfaction. While mobile passenger information systems (PIS) have evolved to offer real-time updates, many passengers still face information overload during service disruptions. This thesis investigates whether personalizing travel information—tailoring it to a user's specific context and preferences—can mitigate this burden. The research aims to define what personalization means in a public transport context, how it can be measured, and whether passengers actually want more personalized, automated systems.
A central contribution of this work is the development of a 5-level taxonomy of personalization in PIS. By conducting a systematic literature review, the author identified that personalization is driven by core functionalities—target identification, situational awareness, and timing of provision. These functions depend on configuration choices regarding data sources, triggers, and initiators. To make this theoretical framework actionable, the study introduced an assessment tool consisting of 83 system actions. This tool allows transport authorities to benchmark their existing digital infrastructure against the defined levels of personalization, effectively diagnosing where a system stands and identifying technological bottlenecks that prevent progress to higher levels of automation.
To understand user sentiment, the study utilized a between-subjects experiment where participants interacted with three different prototypes representing varying levels of personalization. Using an adapted version of the Unified Theory of Acceptance and Use of Technology (UTAUT), the research found that while passengers value relevant information, they do not universally favor higher levels of automated, system-initiated personalization. The results show that privacy concerns significantly negatively influence acceptance at higher levels of automation. Crucially, the perceived benefit (performance expectancy) of these advanced, automated systems was not high enough to outweigh the increased sensitivity regarding personal data usage.
Sam: Give riders a transit app that knows their habits, and they get warier, not happier. That's the uncomfortable result at the end of a three-part study of personalized travel information. Passengers like some personalization. But as the app takes over more of the work, acceptance drops, largely over privacy.
Alex: Before the result, place it for me. Whose work is this, and what kind of document?
Sam: It's a licentiate thesis, a Swedish degree roughly halfway to a doctorate, by Michelle van Ardenne at KTH in Stockholm. It bundles three papers: a systematic literature review, a scoring tool for apps, and a passenger survey experiment.
Alex: I don't work in transport. Why should someone like me, or a transit planner, care?
Sam: Because authorities are spending money on this right now, mostly blind. The author observes that agencies and tech providers build similar features independently, with no shared standards. So a new operator can't tell in advance what personalization would add for its riders. We'll cover how the thesis defines levels of personalization, how it scores real apps, and how far to trust the passenger results.
Alex: Start with the problem. Real-time arrivals already exist. What's still broken?
Sam: Disruptions. When service breaks down, passengers report the information as insufficient. Part of the reason, the thesis argues, is stress. Uncertainty and time pressure make it harder to process a flood of updates. The proposed fix is relevance: show each rider only what matters to their trip, so there's less to wade through.
Alex: "Personalization" covers everything from a saved home address to an app that predicts your day. Is there even a shared definition?
Sam: No, and that's the gap. The literature has no universal definition, and earlier work offered six dimensions to describe any personalized system. What's the object being tailored. Who's the target. Who starts it. What data, from where. And what triggers it. But nobody had formally identified which combinations of those keep recurring.
Alex: So how did the first paper get from six dimensions to something usable?
Sam: A systematic review that kept 40 studies, coded along all six dimensions, plus how passenger and system interact. The key finding is that four dimensions drive how automated a system is: the initiator, the type of input data, its source, and the trigger. Object and target don't track automation at all.
The findings caution public transport authorities against viewing extreme automation as the default goal for information systems. Instead, the research highlights a critical trade-off: higher levels of personalization risk alienating users if privacy concerns are not addressed through transparency and trust-building measures. For practitioners, this study provides a concrete, terminology-consistent roadmap for procurement and system development, helping to align long-term technical innovation with real-world passenger needs and ethical data governance.
AI-generated third-party summary by ResearchPod. Not official content or an endorsement by the paper authors or affiliated organizations.
Alex: Give me an example of that working.
Sam: The thesis offers one. Say the passenger makes a request, using preferences they typed in themselves. That's reactive. Nothing happens until you ask. Now change only one thing: let the system initiate. The strategy shifts, and the system suddenly needs different data to act on its own. Sweden's Skånetrafiken app works in that first spirit: you state a route and time window for disruption alerts.
Alex: That sounds like the levels of self-driving cars, where the question is who's holding the wheel.
Sam: That's exactly the analogy the author draws. The resulting taxonomy has five levels, ordered by increasing system autonomy and decreasing user control. The review also found three core functionalities underneath. Target identification: who is this for. Situational awareness: does it reflect real-time conditions. And timing: when and how to deliver it. One finding I liked: the first step up is simply using real-time network data, so the system knows the status without asking you.
Alex: Levels on paper are one thing. How do you tell which level a real app sits at?
Sam: That's the second paper. It breaks each core functionality into sub-functions, then into concrete system actions per level. The result is 83 system actions. Each is scored as fully implemented, partially, or missing, which produces a readiness score for every level.
Alex: And when they pointed it at actual apps?
Sam: They tested two, through documents and field use in spring 2025. Stockholm's public transport app, from Stockholms Lokaltrafik, and the app of the Dutch rail operator, Nederlandse Spoorwegen. The tool distinguished between them. But both remained at the lowest level, mainly because target identification was underdeveloped. That's the function most directly constrained by privacy concerns around passenger profiling.
Alex: Two apps is a demonstration, not a validation. I'd want that on the record.
Sam: That's fair, and the scoring checks whether required actions exist, not whether users are satisfied. The author says so: it's a first replicable comparison, not a full evaluation. Now, the passenger side.
Alex: How did the third paper measure acceptance?
Sam: They built three prototypes extending the Stockholm app, each matched to a level. Each participant saw one prototype and rated it. The model adapts a standard technology-acceptance framework by adding perceived personalization, privacy concerns, and trust in the provider. They estimated it separately for each level.
Alex: What actually predicted whether people accepted it?
Sam: Three things. By far the strongest was performance expectancy, meaning the belief that it will actually help you. Its effect sat around 0.7 to 0.8 across models. Social influence helped too, more modestly. Privacy concern pulled acceptance down, but only at the higher levels. And the models explained roughly three quarters to four fifths of the variance in acceptance.
Alex: So trust sits upstream, presumably, by easing the privacy worry?
Sam: Yes, trust consistently reduced privacy concern. And ease of use worked indirectly, by raising perceived usefulness rather than acceptance directly. Then the headline: acceptance declined significantly as the level rose. Perceived usefulness went down, privacy concern went up, and social influence stayed flat.
Alex: Is that a rejection of the higher levels, or just less enthusiasm?
Sam: Less enthusiasm. The thesis says higher levels are still viewed positively on average, just with more uncertainty. Riders don't see enough added value to justify the extra data use.
Alex: How much weight should I put on that? It's stated intention toward mock-ups.
Sam: I'd hold it loosely. It's intention, not behaviour, from prototypes, in Stockholm. Three prototypes also can't stand in for all five levels. And the excerpt I read doesn't give the sample size. The author's own biggest caveat is different, though. The taxonomy comes from academic literature, so practices in operating apps may be missing. And framing everything around shifting mental effort to the system is one lens; others could produce different taxonomies.
Alex: There's a real question buried there: is full automation even the goal?
Sam: The author raises exactly that. Maybe the target is good collaboration between rider and system, with transparency and control. The thesis calls privacy-preserving technology and policy the priority, suspecting it could raise both technical feasibility and acceptance.
Alex: So who should read the whole thing, and where should they start?
Sam: Transit authorities, app developers, and anyone studying technology acceptance or personalization. Start with the final chapter on policy implications and limitations. Then read the Paper III summary on why acceptance falls with automation. If you build or procure apps, the Paper II framework is your practical checklist.
Alex: And for everyone else?
Sam: The question isn't whether to personalize. It's how far, and on whose terms.
Alex: Something to think about next time your app pings you mid-delay.