How a Restaurant App Picks What to RecommendHow a Restaurant App Picks What to Recommend

Analytic Hierarchy Process is a familiar way to break a messy choice into a hierarchy of criteria and then compare them in pairs. It works well when the criteria are few. The authors point at the step that follows, where every candidate restaurant has to be scored against every criterion, and argue that this step degrades as the list of alternatives grows. A diner's judgment about one warung against another is rarely crisp, and the comparison matrix starts to misrepresent what the person actually meant.

The mobile application keeps the input short. A user states a budget and a travel radius, and the system filters the candidate list before any weighting happens.

• Distance, computed from coordinates with the Haversine formula rather than from a street network.
• Price, checked against a stated budget such as Rp 30,000 for a meal.
• Rating, the quality signal carried over from the culinary data the system already had.

The modification is to treat the alternative weights as fuzzy rather than exact, which lets a diner express a preference as a range instead of a hard number. The team then ran 30 test cases and compared the application's pick with what human experts chose for the same situations.

• The fuzzy alternative weighting matched the expert decision in 66.67 percent of the 30 cases.
• The original, non-fuzzy AHP process managed 30 percent on the same cases.

An equal weighting baseline, where every criterion counts the same, matched the experts in about 36.7 percent of cases. That number matters because it sets the floor. Beating a naive average is not much of an achievement. More than doubling it is worth reporting, and it suggests the vague, human way of comparing two restaurants carries real information that a crisp matrix throws away.