{
  "abstract": "Agentic AI systems that combine perception with memory must repeatedly solve a context alignment problem:\n\ndeciding whether a current observation matches something already held in long-term memory, or whether it\n\nis new. The mechanism used to make that decision, symbolic (lexical) or sub-symbolic (embedding-based),\n\ndetermines whether the system’s memory generalises across vocabulary variation or fragments into redundant,\n\nnear-duplicate entries. This study investigates that choice inside an agentic memory architecture, using\n\nHuman Activity Recognition as the applied testbed rather than the object of study. Building on the\n\nmemory-driven architecture proposed by Shahabian Alashti et al. [1], a Jaccard-and-Bayesian symbolic\n\ncontext alignment engine is replaced with a semantic engine that computes cosine similarity over Sentence\n\nTransformer embeddings, while every other component of the pipeline, including the Bayesian posterior\n\nand the 60% confidence-gating threshold, is held fixed. Both configurations are evaluated in a closed loop\n\non seen (Robot House Multi-View, N = 420) and unseen (Toyota Smarthome, N = 930) data, at roughly\n\nthree times the scale of the pilot study this project began as. Long-term memory improves accuracy over a\n\nmemory-free baseline on both datasets, confirming its functional role in the architecture. On seen data, the\n\nsemantic mechanism outperforms the symbolic baseline narrowly (61.60% versus 60.90%). On unseen data\n\nthe gap widens substantially (49.80% versus 37.40%), while the supervised classifier alone collapses from\n\n94.00% to 38.30% once evaluated outside its training distribution. These results indicate that the advantage\n\nof semantic over symbolic context alignment is concentrated in generalisation under distribution shift rather\n\nthan in-distribution accuracy, a finding with implications for the design of context alignment mechanisms in\n\nmemory-driven AI systems beyond activity recognition specifically.",
  "abstract_shown": true,
  "affiliations": [
    "University of Hertfordshire"
  ],
  "article_type": null,
  "arxiv": null,
  "author_total": 2,
  "authors": [
    {
      "affs": [
        0
      ],
      "name": "Ganah Gideon"
    },
    {
      "affs": [
        0
      ],
      "name": "Abolfazl Zaraki",
      "orcid": "0000-0001-6204-7865"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/av8w5qqnhtpc4.bib",
  "cited_by": 0,
  "doi": "10.5281/zenodo.23598103",
  "fulltext_url": null,
  "id": "av8w5qqnhtpc4",
  "issns": [],
  "landing_url": "https://zenodo.org/records/23598103",
  "language": null,
  "licence": "CC-BY-4.0",
  "licence_label": "CC BY 4.0",
  "licence_source": "article",
  "licence_url": "https://creativecommons.org/licenses/by/4.0/",
  "metadata_licence": null,
  "native_subjects": [],
  "openalex": null,
  "pdf_url": "https://zenodo.org/api/records/23598103/files/Semantic_Context_Alignment_.pdf/content",
  "pmcid": null,
  "pmid": null,
  "published": "2026-10-23",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/av8w5qqnhtpc4.ris",
  "sources": [
    {
      "first_seen": "2026-11-06T12:21:08Z",
      "native_id": "23598103",
      "source": "zenodo",
      "url": "https://zenodo.org/records/23598103"
    }
  ],
  "subjects": [],
  "title": "Semantic Context Alignment with Open-World Learning for Memory-Driven Human Activity Recognition",
  "title_text": "Semantic Context Alignment with Open-World Learning for Memory-Driven Human Activity Recognition",
  "type": "preprint",
  "updated_at": "2026-11-06T12:21:08Z",
  "url": "https://research.draconitic.com/paper/av8w5qqnhtpc4",
  "venue": "Zenodo",
  "versions": []
}