Use of Machine Learning Classification Models, both Image and Text, in Network Graphing

Introduction

Contemporary graffiti has transcended traditional static urban interventions to become a nomadic, hyper-connected system of communication. In the specific context of freight train graffiti, railcars operate as mobile canvases, carrying visual messages across vast intercontinental logistics corridors throughout North America (Mexico, the United States, and Canada). This phenomenon represents not merely a physical appropriation of industrial space, but what Cruz-Gómez (2014) defines as a geographically dispersed onlife community: an ecosystem where physical transit along railroad tracks is inextricably linked with digital circulation across social media networks, primarily Instagram.

As anthropologist Clifford Geertz observed, human experience is embedded within self-created webs of meaning. In signature graffiti, this web of significance aligns with Figueroa's (2014) conceptual framework: the Triad of the Self-Announcement. This triad articulates the simultaneous assertion of existence (“I exist” / Communalities), spatial footprint (“I was here” / Geographies), and identity (“I am” / Entities). Within this dispersed community, two key actor roles emerge:

Despite its rich sociocultural significance, empirical analysis of large-scale freight graffiti poses a severe methodological challenge due to the volume, ephemerality, and multimodal nature of the data. Traditional qualitative methods in the social sciences fall short when attempting to map thousands of social media posts, visual style variations, and domain-specific slang simultaneously.