Understanding attractions’ connection patterns based on intra-destination tourist mobility: A network motif approach

Urban tourism is a major driver of economic growth, with contemporary tourist mobility characterized by flexible, non-fixed itineraries. Advances in ICT and location-enabled mobile devices have yielded rich spatial-temporal datasets, enabling analyses of tourist mobility via GIS, time geography, and Markov models. Aggregating individual movements into attraction networks supports understanding of topological structures and management implications. However, most network analyses rely on descriptive metrics with limited reliability for pattern validation. Network motifs—recurring, statistically significant subgraphs—offer insights into functional properties of networks and can identify influential destinations. This study applies network motif analysis to an intra-destination tourist network using social media data from Suzhou, China, to link motifs directly to specific attractions and their attributes. The study addresses three questions: (1) What types of motifs do attractions constitute? (2) How are motifs linked to specific attractions? (3) How do motifs relate to attraction attributes (type and title)?

Network motifs are interconnection patterns that recur significantly above random expectation (Milo et al., 2002) and reveal functional dynamics of complex systems. Motif discovery approaches include network-centric enumeration (e.g., Mfinder, FanMod, Kavosh, G-tries) and motif-centric querying (e.g., Grochow, MODA). In tourism, complex network analysis has examined inter- and intra-destination flows using metrics such as density, efficiency, centralities, and core–periphery structure. Applications of motifs to mobility include studies of human daily mobility motifs and travel motifs incorporating temporal/semantic dimensions; however, travel motifs reflect individual itineraries rather than aggregated network structure. Prior tourism motif work is limited (e.g., global tourism flows via UCINET; a South Korea study applying network motif analytics), often constrained by location inaccuracies from cell tower data. This study fills the gap by using social media geotag data to map tourists’ movements to specific attractions, enabling motif detection that connects subgraph structure with attraction roles, types (natural/cultural/commercial), and titles (5A/4A/others).

Study area: Suzhou, China—an urban destination with rich cultural heritage (notably classical gardens) and significant visitation (>100 million domestic visits annually pre-COVID). Data: Geo-tagged Sina Weibo posts collected via API for Suzhou from 2012-04-12 to 2016-10-31, including post ID, user ID, text, images, longitude/latitude, timestamp, and user profile metadata. Tourist identification and preprocessing: (1) Exclude locals based on registered profile location (double-filtration approach). (2) Define length of stay as time between first and last post; exclude users staying >1 month. (3) Define tourists as users who posted within officially listed attractions from the Suzhou Tourism Bureau; use geo-coordinates to match posts to attractions. After filtering: 234,049 microblogs from 54,712 tourists; trajectories constructed by sorting posts temporally and mapping movements between attractions to a directed network. Network: 104 attractions (nodes) and 2,171 directed edges. Motif extraction: Use Kavosh algorithm to enumerate all k-size subgraphs (directed), group isomorphic subgraphs with NAUTY, generate randomized networks, and assess significance. Statistical measures: Frequency (count in input network), Z-score = (Np − Nrand)/σ across randomized networks, and P-value = proportion of random networks where motif frequency exceeds input. Significance thresholds (following Milo et al., 2002) using 1,000 random networks: P-value < 0.01, frequency > 4, Z-score > 1. Search restricted to 3- and 4-node motifs (k>4 did not meet extraction prerequisites where (k−1) motif distributions match random networks).

• Data and network summary: 104 attractions; 2,171 directed edges; 234,049 geo-tagged posts from 54,712 tourists (2012–2016).
• Motif set: Nine significant motifs identified—three 3-node motifs and six 4-node motifs—classified into four classes: chain, mutual dyad, double-linked mutual dyad, and fully connected triad. Three-node motifs constitute 37.61% of all subgraphs; four-node motifs constitute 29.67%.
• Dominant patterns: Three-attraction connections are dominated by chaining (ordered, non-return sequences). Four-node motifs include centrally linked variants representing basecamp/gateway patterns and tightly connected triads enabling free movement among three core attractions.
• Motif frequencies (selected): ID 1 (5.21%), ID 2 (14.91%), ID 3 (17.49%), ID 4 (4.21%), ID 5 (4.90%), ID 6 (5.10%), ID 7 (4.36%), ID 8 (6.14%), ID 9 (4.95%).
• Key attractions/roles: Guanqian Street, Jinji Lake, and Pingjiang Road dominate highest-degree motif nodes, indicating central roles around which most local patterns are organized. Zhouzhuang frequently serves as a gateway/transit attraction (tourists often continue to other attractions without returning due to its distance from Suzhou’s urban core). Hanshan Temple often precedes dispersal to other attractions, acting as a convergence point. Top B-node transit destinations include Tongli National Wetland Park, China Flower Botanical Garden, and Dabaidang Ecological Park (noted for floral/vegetation features, especially in spring).
• Attraction types (natural/cultural/commercial): Across chain motifs, node types are relatively balanced. In mutual dyads, node B and C types often invert (e.g., one dominated by natural, the other by cultural). In double-linked mutual dyads, nodes C and D show consistent type distributions; nodes A in motifs 4 and 5 have high commercial shares (tourist aggregation role), while in motif 6 node A is more of a transit node without high commercial share. In fully connected triads, nodes C and D are largely cultural; node A (hub) shows a balanced mix.
• Attraction titles (5A/4A/others): Nodes with higher degrees typically correspond to renowned 5A/4A attractions. In chain and mutual dyad motifs, node A (transit/hub) tends to have a higher share of famous (5A/4A) attractions than nodes B/C. In double-linked mutual dyads, node A has a much higher share of titled attractions than other nodes (which are mostly non-famous). In fully connected triads, all three fully connected nodes are often titled 5A/4A, indicating strong interflows among top-tier attractions, while peripheral B nodes linked only to A are mostly non-famous.
• Interpretation: The roles of core, transit, and gateway attractions are structurally embedded in recurring local patterns; network structure is strongly influenced by a small set of highly popular attractions.

Network motif analysis, adapted from complex systems biology, effectively uncovers local structural patterns that compose the overall tourist movement network at an intra-destination scale. Unlike itinerary-focused travel motifs, network motifs reveal frequent subgraph structures among attractions and are well-suited to detecting localized, recurrent movement phenomena. The identified motifs clarify roles for attractions—core hubs around which planning can be centered; transit nodes meriting enhanced transport connectivity; and gateway nodes where hospitality and guiding services should be strengthened. The concentration of crucial nodes among a few high-profile attractions implies management challenges, including potential overtourism risks. For destination design and marketing, recognizing how attractions connect within recurring patterns supports targeted product bundling, route curation, and development of substitutes/alternatives to distribute flows. Overall, the local relational features identified by motifs provide actionable insights to shape the broader destination network structure and visitor experience.

This study applies network motif analytics to social media-derived tourist movement data in Suzhou, China, to elucidate recurring local connection patterns among attractions. Using the Kavosh algorithm and strict significance criteria, nine significant motifs (three 3-node; six 4-node) are identified across four classes: chain, mutual dyad, double-linked mutual dyad, and fully connected triad. The city’s local attraction network is organized primarily around Guanqian Street, Jinji Lake, and Pingjiang Road, with specific roles for Zhouzhuang (gateway) and Hanshan Temple (convergence). Higher-degree motif nodes tend to be titled (5A/4A) and are predominantly cultural/commercial. The findings introduce a methodological framework for intra-destination network analysis that ties subgraph structures to concrete attractions and their attributes, offering implications for planning, marketing, and management. Future research should extend motif-based analyses across multiple destinations for comparative generalization and probe the mechanisms of attraction choice within motifs (e.g., itinerary satisfaction maximization).

Findings rely on social media data (Sina Weibo), which are subject to platform-specific adoption, temporal/country-level variation, and engagement biases (e.g., overrepresentation of highly active users). Geotagged posts capture intra-city movement only; patterns may differ across destinations, limiting generalizability. The study focuses on one city; multi-destination comparative analyses are needed. While motifs reveal structural recurrence, they do not directly recover individual itineraries, and additional modeling is needed to infer underlying decision mechanisms (e.g., attraction choice, satisfaction trade-offs).