PlaceWeave: Understanding Place Through Social Video Narratives and Graph-Enhanced Local Knowledge

Paper Title

PlaceWeave: Understanding Place Through Social Video Narratives and Graph-Enhanced Local Knowledge

Publication Info

• Topic area: Human-centered trip planning using AI and social video narratives.
• Keywords: Localness, social video, knowledge graph, trip planning, AI assistant, Graph-RAG, community media, local exploration, itinerary systems, spatial sensemaking.

Background and Problem

• Problem / challenge: Existing location tools flatten the rich, relational aspects of places into simplistic metrics like ratings and coordinates. They fail to capture the "localness" of a place, forcing users to navigate fragmented workflows across multiple tools.
• Significance: Understanding localness is crucial for creating authentic, community-sensitive itineraries, especially for travelers seeking non-touristy experiences. Current systems lack the ability to integrate experiential cues from social media into actionable plans.
• Motivation and related work: Prior research emphasizes the social and cultural dimensions of place but struggles to operationalize these concepts for computational systems. Short-form videos like those on TikTok offer rich, multimodal insights into local practices, but these signals remain unstructured and underutilized in trip planning.

Solution

• Proposed approach: PlaceWeave, a human-centered trip-planning system that uses a graph-enhanced AI pipeline to extract and represent localness from TikTok videos, integrating this information into a unified interface for planning.
• Novelty:
  1. A framework-informed method for constructing a place knowledge graph from short-form social video, encoding localness as relational, evidence-backed attributes.
  2. A localness-aware Graph-RAG pipeline that grounds conversational assistance, map overlays, and route planning in the same evidence.
  3. A unified interface combining maps, conversational AI, and itinerary tools to reduce fragmentation and enhance spatial sensemaking.
  4. A user study demonstrating PlaceWeave’s ability to support local-feeling plans and reduce fragmented workflows compared to a baseline toolchain.
• Procedure and key techniques:
  • Multimodal entity extraction from TikTok videos, including visual, audio, and textual features.
  • Construction of a place knowledge graph with nodes for locations, activities, ambience, and temporal rhythms, linked to supporting evidence.
  • A Graph-RAG pipeline for retrieving and synthesizing localness attributes, enabling AI-powered recommendations and itinerary planning.
  • A user interface integrating an interactive map, conversational assistant, insights panel, mental map canvas, and route planner.

Results

• Concrete findings:
  • SUS score: 74.2 (above usability threshold).
  • High ratings for perceived localness (4.00/5), recommendation quality (4.22/5), creativity support (4.19/5), and satisfaction (4.10/5).
  • Knowledge graph: 28,341 nodes and 96,507 edges, with low hallucination rates for concrete attributes like "hidden gems" (≤0.11).
• Advantage over baselines:
  • Unified interface reduced fragmentation compared to a baseline workflow using ChatGPT, TikTok, Google Maps, and Draw.io.
  • Participants discovered more authentic, local-feeling places and created itineraries with higher perceived localness and authenticity.
  • Tight coupling of conversational AI with maps and itineraries improved trust and usability.
• Experiments / evaluation:
  • Within-subject study with 18 participants planning trips in two conditions (PlaceWeave vs. baseline).
  • Measures included SUS, custom Likert scales for localness and satisfaction, and qualitative analysis of think-aloud protocols and interviews.
• Limitations and future work:
  • Overrepresentation of younger demographics and popular venues due to reliance on TikTok.
  • Information density in the interface led to occasional user overwhelm.
  • Future work includes expanding data sources, improving visual hierarchy, and conducting in-the-wild evaluations of real-world trip outcomes.

Summary

PlaceWeave introduces a novel approach to trip planning by leveraging TikTok videos and a graph-enhanced AI pipeline to extract and represent localness. Its unified interface integrates maps, conversational AI, and itinerary tools, enabling users to create authentic, community-sensitive plans while reducing workflow fragmentation. A user study demonstrated significant advantages over baseline tools in usability, perceived localness, and planning efficiency. Future work will address representational biases, interface complexity, and real-world validation to further enhance its applicability and community alignment.