D2C Announces Research Results at One of Japan's Largest Academic Conferences on Natural Language Processing and Artificial Intelligence
D2C Inc. will present a novel research method at JSAI2026 that analyzes consumer interest in restaurant chains over time using telecommunications carrier search logs, quantifying potential pre-visit interest via NPMI without relying on purchase data.
📋 Article Processing Timeline
- 📰 Published: April 15, 2026 at 20:00
- 🔍 Collected: April 15, 2026 at 11:31
- 🤖 AI Analyzed: April 19, 2026 at 09:10 (93h 38m after Collected)
D2C Inc. (Headquarters: Minato-ku, Tokyo; President and CEO: Yuki Oka; hereinafter "D2C") announced that it presented research results by Daichi Inoue and other members of the company at the "32nd Annual Meeting of the Association for Natural Language Processing (NLP2026)" held in March 2026, and will present the latest research results further developing that content at the "40th Annual Conference of the Japanese Society for Artificial Intelligence (JSAI2026)" to be held in June 2026.
This research proposes a method to chronologically analyze consumer interest in specific restaurant chains using massive search logs and member attribute information (gender, age group) held by a telecommunications carrier. Specifically, keywords searched together with store names are extracted from search queries. The strength of the connection between those words is quantified using an index measuring the "depth of relevance" (NPMI), rather than mere frequency of appearance. This made it possible to grasp as objective data "from which customer demographic" and "in what context" each chain is garnering attention.
In NLP2026 conducted previously, the effectiveness of the customer demographic analysis method using search logs was demonstrated, but for this upcoming JSAI2026, the research was further deepened to realize time-series analysis on a monthly basis. It clarified how limited-time products, collaborative measures with other companies, and even social events changed the structure of consumer interest. This research result visualizes consumer trends that cannot be fully captured by purchasing data alone, and is expected to contribute to the realization of more precise marketing measures.
The details of the presentation at JSAI2026 are as follows:
[Title] Time-Series Analysis of Restaurant Chain Customer Demographics Based on Telecommunication Carrier Search Logs
[Presenter] Daichi Inoue (Data Insight Department, Single ID Marketing Division, D2C Inc.)
[Date and Time] Friday, June 12, 2026, 12:30 - 14:00
[Session Number] 5Yin-A
[Session Venue] Venue Y (Exhibition Hall AB-1)
[Presentation Summary]
This research proposes a method to analyze consumer interest in restaurant chains utilizing massive search logs and member base attribute information held by a telecommunications carrier. In contrast to conventional customer analysis that relies on store visit history and purchase data, this research focuses on "search behavior" as potential pre-visit interest data.
Specifically, keywords co-occurring with store names are extracted from search queries, and the level of interest by gender and age is quantified using an index measuring their "strength of connection" = Normalized Pointwise Mutual Information (NPMI). Furthermore, by introducing a monthly time-series analysis, it clarified how limited-time products, collaboration measures with other companies, or social events brought about changes in the interest structure of each customer demographic.
As a result of the analysis, in addition to the existence of main customer demographics for each restaurant business format, differences in interest groups and chronological fluctuations were confirmed for each chain even within the same business format. It also demonstrated the ability to distinguish different patterns, such as a sustained rise in interest towards a specific measure, or a temporary concentration of interest caused by scandals.
Since this method allows understanding changes in customer interest without using store visit data or purchase history, it is expected to be applied to advanced solutions that support corporate decision-making, such as measuring the effectiveness of marketing measures, competitive comparisons, and the early identification of reputation risks.
Principal Component Analysis (PCA) for NPMI vectors by gender/age and NPMI by chain store (*)
* The first principal component mainly reflects gender differences, and the second principal component reflects age differences. A certain tendency in customer demographics can be confirmed for each business type, while differences in customer demographic characteristics can be seen for each chain even within the same business type. Furthermore, a post-survey revealed that some chains are consistent with their actual primary target demographics.
■ Conference Overview
[40th Annual Conference of the Japanese Society for Artificial Intelligence (JSAI2026)]
The JSAI Annual Conference is one of the largest academic conferences in Japan where the forefront of academia and technology regarding artificial intelligence (AI) gathers. At the annual conference held every year, researchers and engineers from industry, academia, and government assemble under one roof to present the latest research results and technological trends, introduce practical examples of companies, and hold panel discussions. It attracts attention from many participants as a place where knowledge and networks responsible for the future of AI converge.
- Session Period: Monday, June 8, 2026 - Friday, June 12, 2026
This research proposes a method to chronologically analyze consumer interest in specific restaurant chains using massive search logs and member attribute information (gender, age group) held by a telecommunications carrier. Specifically, keywords searched together with store names are extracted from search queries. The strength of the connection between those words is quantified using an index measuring the "depth of relevance" (NPMI), rather than mere frequency of appearance. This made it possible to grasp as objective data "from which customer demographic" and "in what context" each chain is garnering attention.
In NLP2026 conducted previously, the effectiveness of the customer demographic analysis method using search logs was demonstrated, but for this upcoming JSAI2026, the research was further deepened to realize time-series analysis on a monthly basis. It clarified how limited-time products, collaborative measures with other companies, and even social events changed the structure of consumer interest. This research result visualizes consumer trends that cannot be fully captured by purchasing data alone, and is expected to contribute to the realization of more precise marketing measures.
The details of the presentation at JSAI2026 are as follows:
[Title] Time-Series Analysis of Restaurant Chain Customer Demographics Based on Telecommunication Carrier Search Logs
[Presenter] Daichi Inoue (Data Insight Department, Single ID Marketing Division, D2C Inc.)
[Date and Time] Friday, June 12, 2026, 12:30 - 14:00
[Session Number] 5Yin-A
[Session Venue] Venue Y (Exhibition Hall AB-1)
[Presentation Summary]
This research proposes a method to analyze consumer interest in restaurant chains utilizing massive search logs and member base attribute information held by a telecommunications carrier. In contrast to conventional customer analysis that relies on store visit history and purchase data, this research focuses on "search behavior" as potential pre-visit interest data.
Specifically, keywords co-occurring with store names are extracted from search queries, and the level of interest by gender and age is quantified using an index measuring their "strength of connection" = Normalized Pointwise Mutual Information (NPMI). Furthermore, by introducing a monthly time-series analysis, it clarified how limited-time products, collaboration measures with other companies, or social events brought about changes in the interest structure of each customer demographic.
As a result of the analysis, in addition to the existence of main customer demographics for each restaurant business format, differences in interest groups and chronological fluctuations were confirmed for each chain even within the same business format. It also demonstrated the ability to distinguish different patterns, such as a sustained rise in interest towards a specific measure, or a temporary concentration of interest caused by scandals.
Since this method allows understanding changes in customer interest without using store visit data or purchase history, it is expected to be applied to advanced solutions that support corporate decision-making, such as measuring the effectiveness of marketing measures, competitive comparisons, and the early identification of reputation risks.
Principal Component Analysis (PCA) for NPMI vectors by gender/age and NPMI by chain store (*)
* The first principal component mainly reflects gender differences, and the second principal component reflects age differences. A certain tendency in customer demographics can be confirmed for each business type, while differences in customer demographic characteristics can be seen for each chain even within the same business type. Furthermore, a post-survey revealed that some chains are consistent with their actual primary target demographics.
■ Conference Overview
[40th Annual Conference of the Japanese Society for Artificial Intelligence (JSAI2026)]
The JSAI Annual Conference is one of the largest academic conferences in Japan where the forefront of academia and technology regarding artificial intelligence (AI) gathers. At the annual conference held every year, researchers and engineers from industry, academia, and government assemble under one roof to present the latest research results and technological trends, introduce practical examples of companies, and hold panel discussions. It attracts attention from many participants as a place where knowledge and networks responsible for the future of AI converge.
- Session Period: Monday, June 8, 2026 - Friday, June 12, 2026