[1J3-GS-10d] AI application

1:40 PM - 1:55 PM JST(4:40 AM - 4:55 AM UTC)

[1J3-GS-10d-01] A Food Name Recommendation System for Marketing Based on Social Media Analysis

〇Donglin Qian¹, Takuya Shintate¹ (1. NABLAS Inc.)

Keywords:

Recommend System,Marketing,Language Model

Food trends have a direct impact on manufacturing, inventory management, and logistics planning. We need to some methods that can early identify and organize food name candidates worth attention in marketing analysis from social media mentions. However, existing indicators for selecting such candidates are not sufficiently established in Japanese. In this study, we assume that food names are characterized by both low current frequency (rarity) and non-typical lexical composition (surprisingness). Based on this assumption, we propose a framework that integrates these two factors to rank food name candidates, thereby supporting exploratory analysis for food trend analysis. Specifically, food names are assigned to hierarchically defined tags, and rarity is estimated based on tag occurrence frequencies, while surprisingness is approximated from the atypicality of lexical associations derived from word embedding representations of the constituent terms. Furthermore, we construct a dataset of 35,449 Japanese food names accompanied by sentiments and impressions, collected from Twitter and Bluesky posts between 2024 and 2025. Experimental results demonstrate that the proposed method can estimate tags with practical accuracy and can qualitatively extract food name candidates that jointly exhibit rarity and surprisingness through ranking-based comparative analysis.