Learning-Based Algorithms and NLP for Adaptive E-Participation in Smart Cities: A Systematic Review and Knowledge Engineering Perspective

The E-participation platforms of smart cities have not been very progressive to date and mainly provide one-size-fits-all interfaces, which do not suit citizens’ varying motives, digital literacy, or situational contexts. Mobile platforms for e-participation should be designed adaptively, but existing examples show that despite this adaptability, there is still a gap between the design of such platforms and the learning paradigm under which many people interact with the city. The mismatch between the two is the adaptive e-participation gap, which is highlighted in this paper by a PRISMA 2020-inspired systematic literature review and the creation of a benchmark on six deployed systems. Of the 220 records retrieved from five sources—Scopus, Web of Science, Google Scholar, IEEEXplore, and ACM Digital Library—22 studies met the inclusion criteria, of which 17 explicitly used learning-based algorithms, NLP, or knowledge engineering to form the algorithmic core. Building on this empirical groundwork, we propose three conceptual contributions: (i) a taxonomy of five learning-based algorithmic families mapped to the identified gaps, analyzed along shared axes of sample efficiency, fairness tractability, and their explainability mechanism; (ii) a conceptual knowledge engineering framework, specified through competency questions, that connects citizens, issues, proposals, policies, and outcomes into a semantic knowledge graph; and (iii) an incremental deployment roadmap (L0–L5). We also discuss ethical and political considerations that any algorithmic intervention in democratic participation must address. The taxonomy, framework, and roadmap are offered as propositions for empirical testing; their implementation is identified as future work. The empirical layer is descriptive rather than validated. Of the seventeen algorithmic-core studies, thirteen (76.5%) remain at L1–L2, four (23.5%) reach L3 or above, and none reach L5. A mixed-methods appraisal tool (MMAT) appraisal found that nineteen of the twenty-two context-corpus studies are empirically appraisable; three are conceptual frameworks that propose systems without evaluating them, and all three are in the algorithmic core.