Data aggregator
platform development
in Dubai

An aggregator in the UAE receives queries in two languages and in several spellings of each. The search box is where that mix either works or falls apart. A visitor who types a hotel name in Arabic and gets nothing, while the same hotel is listed in English, will assume the site is empty and leave.

Layout comes first, because it is visible before any query runs. Arabic reads from right to left, so an Arabic interface mirrors the page. The filter panel moves to the right side. Result cards place the image, the name and the price in the reverse order. Arrows swap direction. Price sliders start at the right. Icons that suggest direction, such as a back arrow, have to flip, while icons that do not, such as a search glass, usually stay as they are.

Mixed content inside a card needs care. An Arabic listing may contain an English brand name, a phone number and a price in digits. The browser handles much of this through the bidirectional text rules, but punctuation and parentheses can land on the wrong side when a line mixes directions. Test cards with real mixed data, not with placeholder text in one language.

Now the search itself. Arabic has several letters that people write in more than one way. The alef appears with or without a hamza above or below. The final ya is often typed as alef maqsura, and the ta marbuta at the end of a word is often typed as ha. A visitor may type either form. The index should normalise these variants to a single form, both when data goes in and when a query arrives, so both spellings find one listing.

Diacritics are another layer. Most Arabic text online has no short vowel marks. Some sources include them. A search that treats a word with marks and the same word without them as different words will miss matches. Strip the marks before indexing and before matching. The same goes for the tatweel, the stretching character sometimes used for decoration in names.

Digits come in two sets. Arabic text may use Eastern Arabic numerals, while English text uses Western ones. A visitor looking for a flat on a certain floor or a car of a certain model year may type either. Convert both sets to one before comparing. Display can still follow the language of the interface.

Then there is the query written in Latin letters. Many people in the region type Arabic words and names using English characters, often with numbers standing in for sounds that have no Latin letter, such as 3 for ain or 7 for the strong ha. There is no single standard spelling. A search for a district name can arrive in four or five versions.

Transliterated queries are best handled with a curated list rather than an automatic converter. For each important entity in the data, such as districts, brands, cuisines or car makes, store the Arabic name, the English name and the common Latin spellings people actually type. The search logs show which spellings appear. Add them regularly.

Fuzzy matching helps with the rest, but it needs limits. Short Arabic words can be one letter apart and mean completely different things. Tune the tolerance for each language separately.

Finally, decide what language the results appear in. A visitor who searched in Arabic on the English interface probably wants Arabic names shown. Store both names for every listing. Pick the display name from the query as well as the interface setting.