Порівняння підходів на основі парсингу та ШІ агентів для вирішення задачі відстеження цін товарів

Abstract

This paper compares two approaches to tracking food prices: web parsing using headless browsers and AI agent-based systems. Special attention is paid to the problem of inconsistent product names across different stores. It describes a parsing approach based on headless Chromium instances with fuzzy matching. The AI agent-based approach, in turn, relies on LLMs combined with web search. Both approaches were tested on the same queries and compared on various criteria, including coverage, execution time, infrastructure, and cost. The parser-based approach achieved lower coverage and accuracy than the AI agent, but a significant disadvantage of the latter is its cost. Further research should focus on improving the accuracy and reducing the cost of the AI agent