Original title: Arabic text preprocessing for dynamic searchable symmetric encryption: An empirical evaluation and preprocessor selection framework
Kholoud Al-Saleh · September 2026
Searching through encrypted files without revealing your text to the cloud is tough. In Arabic, this is even harder because one root word can take dozens of forms. Without prep work, you miss matches and build massive, wasteful search indexes. To fix this, researchers tested four Arabic text-cleaning tools against a simple baseline that only normalizes text. These tools shrink words down to their stems or segments before encrypting them into a searchable index. The team found that stemmingCutting words down to their root form to group related words together. shrinks the vocabulary list by up to 83%, which makes the encrypted index much smaller and cheaper to transmit. However, the simplest baseline searched fastest simply because it matched the fewest documents; returning more matching files takes more time. Overall, light stemming and segmentation struck the best balance of finding the right files without grabbing irrelevant ones. Root stemmers lost accuracy without helping recall.
Imagine organizing a library where every book title can be written in dozens of different ways depending on grammar. You want to find books by topic, but the catalog is encrypted so the librarian can't see the titles. If you simplify titles too much, you lose meaning; if you don't simplify enough, the catalog becomes huge and slow. In this paper, the researchers tested different ways to simplify Arabic words before encrypting them for search. They found that the simplest simplification made searches fastest but missed many books, while a balanced approach found the most relevant books without slowing down too much.
For anyone building encrypted search systems for Arabic data, this work provides clear guidelines on which text preprocessing method to use. The wrong choice can make searches slower or miss relevant results. System designers can now balance speed, memory, and accuracy based on their specific needs.
The abstract notes this is the first framework for Arabic DSSE, but it is based on a single corpus (Khaleej) and a single DSSE scheme (IM-DSSE). Only four preprocessing strategies were evaluated, and scalability was tested only up to 45,500 documents. The findings may not generalize to other languages, schemes, or larger corpora.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Arabic text preprocessing for dynamic searchable symmetric encryption: An empirical evaluation and preprocessor selection framework — Kholoud Al-Saleh — PLoS ONE — doi:10.1371/journal.pone.0351281
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.