2026-10-13T14:27:38Z https://api.figshare.com/v2/oai

oai:figshare.com:article/33959144 2026-09-21T18:01:37Z category_146 category_21 category_24 category_734 category_132 portal_5 item_type_3 month_year_09_2026

<p>Qualitative failure cases across preprocessors.</p> Kholoud Saad Al-Saleh (25085069) Space Science Biotechnology Evolutionary Biology Biological Sciences not elsewhere classified Infectious Diseases morphological segmentation achieve match morphological variants framework provides actionable creates substantial challenges concrete preprocessor choices lucene arabic analyzer arabic text preprocessing proportionally reducing encrypted enables keyword search stemming reduces vocabulary fastest per query evaluate vocabulary size incidence matrix dsse baseline &# 8212 query search latency set size rather limit retrieval recall vocabulary size encrypted search search latency light stemming arabic dsse size effect retrieval precision retrieval breadth index size search quality latency persists transmission cost systematic evaluation paper presents operate successfully memory scalability khoja stemmer isri stemmer fewest documents empirical evaluation driven primarily best trade assess scalability annotated 1 500 documents <div><p>Dynamic Searchable Symmetric Encryption (DSSE) enables keyword search over encrypted data without revealing plaintext to the server. Arabic morphological richness — where a single root generates dozens of surface forms — creates substantial challenges for encrypted search: unprocessed vocabularies inflate encrypted-index size and transmission cost, and limit retrieval recall by failing to match morphological variants. This paper presents the first empirically grounded preprocessor selection framework for Arabic DSSE, derived from a systematic evaluation of four Arabic preprocessing strategies — the Khoja stemmer, ISRI stemmer, Lucene Arabic Analyzer, and Farasa segmenter — against a normalization-only baseline within the Incidence Matrix DSSE (IM-DSSE) scheme on the Khaleej Arabic news corpus. We evaluate vocabulary size, search latency, search quality, and retrieval breadth on an annotated 1,400-document corpus using 56 benchmark queries, and assess scalability on cloud infrastructure across corpus sizes up to 45,500 documents. Stemming reduces vocabulary by up to 83% relative to the normalization-only baseline, proportionally reducing encrypted-index size and transmission cost. We find that per-query search latency is driven primarily by result-set size rather than vocabulary size: the normalization-only baseline is fastest per query because it matches the fewest documents, while preprocessors that broaden retrieval incur higher latency. At scale, all five configurations — including the normalization-only baseline — operate successfully to 45,500 documents, and the result-set-size effect on latency persists. In terms of search quality, light stemming and morphological segmentation achieve the best trade-off between retrieval precision and recall, while root-based stemmers sacrifice precision without commensurate recall gains. Based on these findings, the framework provides actionable, evidence-based guidelines mapping deployment requirements — memory scalability, search latency, search quality, and retrieval breadth — to concrete preprocessor choices for Arabic DSSE system designers.</p></div> 2026-09-21T18:01:14Z Dataset Dataset 10.1371/journal.pone.0351281.t009 https://figshare.com/articles/dataset/_p_Qualitative_failure_cases_across_preprocessors_p_/33959144 CC BY 4.0