InterPlace: Framework for Person-Place Relations

• The paper introduces a novel framework that leverages visit frequency and dwell time to uniquely characterize individual-place relationships.
• InterPlace is a mobility framework that clusters visited locations into seven interpretable categories, addressing the limitations of traditional trajectory models.
• The framework’s use of Gaussian Mixture Models and visitation motifs offers actionable insights for personalized recommendations, mobility prediction, and urban planning.

InterPlace is a user-level visit-characterization framework introduced in “Understanding Individual-Space Relationships to Inform and Enhance Location-Based Applications” (Amichi et al., 19 Feb 2025). It models the relationship between an individual and a place through two observable mobility signals—visit frequency and dwell time—and clusters a person’s visited locations into seven interpretable spatial relationship types. The framework addresses a limitation of conventional mobility and location-based models: they typically treat places as prediction targets, trajectory waypoints, or cluster members, but do not distinguish what a place means to a particular individual. InterPlace instead formalizes person-place relationships along a spectrum from exploratory and casual visits to routine and anchored-like visits, and analyzes how these categories differ across semantic, spatial, and temporal dimensions in Singapore and Beijing (Amichi et al., 19 Feb 2025).

1. Conceptual scope and motivation

InterPlace is grounded in the premise that human mobility is not reducible to destination prediction. The same location can function as a central habitual anchor, an occasional but familiar stop, a leisure or exploratory destination, or a rare site associated with an extended stay. In the formulation used by the paper, the relevant analytical question is not only where a person goes, but how that person relates to the places visited (Amichi et al., 19 Feb 2025).

The motivating contrast is with existing mobility models that primarily emphasize predicting next location, counting visits, modeling trajectories, or clustering places without interpreting the human-place relationship. InterPlace moves the analysis toward a more human-centered characterization of visits. The paper states that, within its scope, it is the first method to exploit visitation patterns and dwell times to characterize an individual’s relationship with specific locations (Amichi et al., 19 Feb 2025).

This shift is consequential because the categories sought by the framework are not generic place classes. They are individualized relationship types. A workplace, residence, park, restaurant, or travel stop is not assigned a fixed global meaning; rather, its meaning is inferred relative to a given user’s recurrent behavior. This suggests a personalized semantics of mobility in which attachment, routine, exploration, and transition are inferred from behavioral regularities rather than imposed by place labels alone.

2. Formalization and computational pipeline

The paper defines a mobility record as

$m_i = (lat_i, lon_i, t_i),$

where $(lat_i, lon_i)$ are coordinates and $t_i$ is the timestamp. A mobility trace for user $u$ is

$\mathcal{D}_u = \{m_i\}_{i=1}^{n},$

ordered in time as

$t_1 < t_2 < \dots < t_n.$

The global notation also includes $\mathcal{U}$ for the set of users, $\mathcal{L}$ for the set of locations, $N$ for the number of users, and $M$ for the number of locations (Amichi et al., 19 Feb 2025).

Before visit characterization, the framework evaluates data quality through temporal completeness and spatial completeness. Temporal completeness assesses whether records are present across time windows of size $(lat_i, lon_i)$0. Spatial completeness filters unrealistic jumps using elapsed time $(lat_i, lon_i)$1, distance between consecutive locations $(lat_i, lon_i)$2, and a maximum-speed threshold. The paper sets $(lat_i, lon_i)$3 km/h (Amichi et al., 19 Feb 2025).

The core feature space is deliberately minimal. Each user-place pair is described by:

• Visit frequency: how often a person visits a location.
• Dwell time: how long the person stays there when visiting.

InterPlace then clusters places for each user using a Gaussian Mixture Model (GMM) on these two features. Model selection is performed with BIC and AIC, testing cluster counts from 1 to 21 and four covariance types—spherical, diagonal, tied, and full. The elbow in the BIC/AIC curves occurs at 7 clusters, especially for the tied covariance model, and this yields the seven relationship groups $(lat_i, lon_i)$4–$(lat_i, lon_i)$5 (Amichi et al., 19 Feb 2025).

A further analytical layer is the study of visitation motifs, i.e., transitions between the inferred visit types. The paper uses these motifs to show how one kind of visit tends to follow another. This extends the framework beyond static categorization toward a transition-aware account of mobility behavior.

3. The seven spatial relationship types

The seven groups form a spectrum from rare, brief, and exploratory encounters with place to stable, central, and anchored-like relations. The paper interprets these groups as capturing exploratory, casual, routine, transitional, and anchored forms of person-place attachment (Amichi et al., 19 Feb 2025).

Group	Distinguishing pattern	Interpretation
G1	Frequency and dwell time below the lower 20%	Short-duration exploratory visits
G2	Frequency in the bottom 20%, dwell time long; typically 10 hours to less than a full day	Rare but long exploratory visits
G3	High dwell times exceeding a full day, with varying frequency	Routine-change / disruptive visits
G4	Frequency 20%–44% in Singapore, 16%–33% in Beijing; dwell time within the lower 20%	Casual visits
G5	Frequency roughly 20%–40%, with moderate to high dwell time	Moderate-frequency, moderate/high-dwell visits
G6	Visits in the upper 60% of the frequency distribution, with moderate dwell time	Regular routine locations
G7	Visits in the upper 20% of the frequency distribution, with significant daily dwell time	Anchored-like locations

The interpretive logic of the taxonomy is explicitly tied to the joint distribution of frequency and duration. Low frequency with short dwell corresponds to brief exploratory or incidental visits. Low frequency with long dwell indicates rare but meaningful or special-purpose visits. High frequency with long dwell identifies anchored, routine, and central places. Moderate frequency with moderate dwell captures regular but not dominant locations (Amichi et al., 19 Feb 2025).

The most important conceptual distinction is that InterPlace does not collapse all high-frequency locations into a single “important place” class. It differentiates between regular routine locations $(lat_i, lon_i)$6 and anchored-like locations $(lat_i, lon_i)$7, while also separating moderate-frequency semi-regular places $(lat_i, lon_i)$8 from short casual stops $(lat_i, lon_i)$9. Likewise, infrequent visits are not treated as homogeneous noise: $t_i$0, $t_i$1, and $t_i$2 distinguish brief exploration, extended occasional stays, and routine-disruptive episodes.

4. Empirical basis: datasets, cities, and observed differences

The study combines mobility and point-of-interest data from three sources (Amichi et al., 19 Feb 2025):

• Singapore dataset: 144,795 users, spanning Dec 1, 2022 to Jan 31, 2023, described as very dense and high-granularity.
• Geolife dataset: 182 users, spanning April 2007 to August 2012, with trajectories across 30+ cities but concentrated in Beijing.
• PlanetSense PoI dataset: 238,690 PoIs in Singapore and 1,677,835 PoIs in Beijing, organized into 44 semantic categories.

A notable methodological difference between the two cities concerns temporal completeness. Singapore exhibits more stable temporal completeness across longer observation windows, so the study keeps 30 days for Singapore. Beijing shows temporal-completeness degradation with longer windows, so the study uses 15 days for Beijing (Amichi et al., 19 Feb 2025).

Across both cities, the analysis finds a prevalence of anchored-like visits. However, the composition and context of those visits differ. The abstract reports that anchored-like visits in Singapore include recreational spaces, whereas in Beijing they are limited to residential, business, and educational sites. The detailed semantic analysis also reports a stronger Beijing presence of office_building, public_service, education, power_plant, and religious categories (Amichi et al., 19 Feb 2025).

The behavioral profiles differ as well. Singapore shows a more fluid mix of exploratory and routine behaviors, with exploratory visits able to evolve into stable anchored-like visits and with longer exploratory stays more visible. Beijing exhibits stronger attachment to key locations and a more pronounced $t_i$3, indicating stronger anchored behavior (Amichi et al., 19 Feb 2025).

Spatially, G1 is broadly dispersed in both cities. In Singapore, G7 is concentrated around residential areas. In Beijing, G6–G7 are concentrated in downtown areas, especially around Peking University and the Haidian business district (Amichi et al., 19 Feb 2025). These patterns support the paper’s broader claim that geographic and cultural context materially shape the structure of person-place relationships.

5. Interpretation, scope, and common confusions

A frequent misunderstanding would be to read InterPlace as a standard place-classification system. It is not. The framework does not primarily classify locations by global semantics, nor is it designed as a next-location predictor. Its central object is the individual-space relationship, inferred from user-specific visitation patterns and dwell times (Amichi et al., 19 Feb 2025).

A second misconception would be to equate anchored-like strictly with home or work. The paper explicitly uses home and work as central examples, but the empirical analysis shows that Singapore’s anchored-like category includes recreational spaces. Conversely, rare visits are not necessarily trivial: G2 corresponds to rare but long stays, and G3 captures routine-change or disruptive visits with dwell times exceeding a full day. These categories indicate that low frequency does not imply low significance (Amichi et al., 19 Feb 2025).

The results also argue against assuming that the same visit type has identical semantic content across cities. Singapore and Beijing differ in temporal completeness, anchored-like composition, exploratory visibility, and spatial concentration. A plausible implication is that models built on one city’s inferred relationship types should not be transferred to another city without attention to local geographic and cultural context.

The term InterPlace is also potentially confusing because similarly named or place-related systems exist in unrelated domains. These include PLACE, a 3D human-scene interaction generation method (Zhang et al., 2020); TextInPlace, an indoor visual place recognition framework for repetitive structures (Tao et al., 9 Mar 2025); a context-based meetup-location geoprocessing framework described as InterPlace in a road-network setting (Wang et al., 2018); and a later quantum hardware co-design framework published under the same name (Du et al., 12 Sep 2025). None of these address the individualized mobility-relational problem posed in (Amichi et al., 19 Feb 2025).

6. Applications and significance for mobility research

The paper emphasizes several application areas for InterPlace (Amichi et al., 19 Feb 2025). In personalized recommendations, distinguishing between casual, routine, exploratory, and anchored places can align suggestions more closely with user intent. In mobility prediction, recognizing whether a user is in an exploratory or anchored mode can improve predictive accuracy. In public health, separating stable from transient visits can support exposure analysis, intervention targeting, and contact modeling. In transportation planning, the framework can reveal routine flows, important urban hubs, and shifts in commuting or activity patterns.

The broader significance lies in analytical interpretability. InterPlace replaces a purely trajectory-centric view with a relational one: the operative question becomes not simply “where did the user go?” but “what role does this place play in the user’s mobility life?” This reframing is especially relevant for location-based services, where personalization depends not only on location identity but on the stability, centrality, and affective or routine status of locations relative to the individual (Amichi et al., 19 Feb 2025).

Because the framework uses only visit frequency and dwell time, its representational basis is compact. Yet the Singapore–Beijing comparison shows that these two signals are sufficient to recover meaningful heterogeneity across semantic, temporal, and spatial dimensions. The resulting picture is a mobility analysis framework in which routine, exploration, casuality, disruption, and anchoring are treated as empirically inferable properties of person-place relationships rather than as residual byproducts of trajectory prediction.