#Japan #Tokyo #LLM #Digidigi #Association Star

Unveiling the 'Association Star' in Digidigi

Company Overview
ToDAO NADA Inc., based in Taito, Tokyo, under the leadership of CEO Yasunori Matsumoto, has recently added an innovative feature called 'Association Star' to their LLM brand recognition measurement tool, Digidigi. This new functionality allows for the visualization and measurement of how brands are associated within consumer minds. Unlike traditional recall measurement, which assesses whether a brand comes to mind when prompted with a category, the Association Star measures what thoughts arise when a specific brand is mentioned. This comes in the form of a two-layer associative word graph, illustrating how LLMs remember that brand.

Understanding Brand Associations

As users increasingly turn to LLMs for product selection and information gathering, the presence of brands in LLM responses is transforming into a key metric of brand recognition. Inside the LLM, brand names are not stored as isolated words but rather as interconnected clusters of meanings derived from contextually relevant terms encountered repetitively during the model's training. This means that how a brand is recalled depends on the surrounding linguistic context employed during LLM interactions.

In this LLM era, brand strategy must go beyond simply verifying if a brand is remembered (i.e., presence or absence), and instead focus on understanding what concepts are linked to it. This is where the Association Star shines, as it empirically reveals the structure of these associative concepts.

Positioning Within the Query-Key-Value Model

LLM response generation can be explained through an attention mechanism known as the Query-Key-Value (QKV) model. The Association Star simulates this by measuring the connections as follows:

• - Query: Recall question (e.g., "What comes to mind when I say PR effectiveness measurement tool?")
• - Key: The brand name being queried (central node)
• - Value: The set of associative words that emerge (1st layer and 2nd layer)

With the same brand name, the associations extracted may vary based on the context of the query. The Association Star conditions its measurements on specified recall questions, effectively capturing how different contexts influence the output. This design effectively eliminates ambiguity, allowing for precise measurement of associations without interference from similar terms, such as 'bank' referring to both a financial institution and a river bank.

Key Features of Association Star

1. Two-Layer Associative Structure Visualization:
The system visualizes the connections from the brand name to primary associative words, extending into further layers, creating a two-layer associative graph. Through this, users can observe shared terms (bundles) and also note instances where a brand reemerges in the context of associations (recirculation), providing key insights into the strength of brand associations.

2. Dependable Connections without Contextual Bias:
The tool employs three distinct measurement approaches, capturing only those words that appear consistently, irrespective of phrasing. This ensures that only stable associations are retained in the visualization.

3. Comparative Measurement Consistency:
By standardizing parameters, question phrasing, and aggregation rules, the measurements facilitate comparisons. While LLM outputs may vary under the same conditions, the fixed methodology allows discrepancies to be viewed as differences in memory processing rather than measurement conditions.

4. Storage and Update of Measurement Results:
Measurements are stored based on units of question phrases, subjects, and models, allowing for immediate retrieval of results in subsequent assessments. This means that changes in model types will initiate new measurements based on initial recall assessments.

5. Reporting Capabilities:
Results can be exported as PNG star charts and detailed data tables in Markdown, making them highly accessible for internal reporting and knowledge sharing.

Application Scenarios

• - Assessing how your brand is being remembered by LLMs
• - Identifying discrepancies between desired recall contexts and actual associative content
• - Comparing associative structures of competitors (measured under identical conditions)
• - Conducting longitudinal studies of messaging strategies across LLM model iterations

Conclusion

Feature Name: Association Star
Availability: Integrated within Digidigi (https://digidigi.qlipper.jp)
Launch Date: September 8, 2026
Measured Models: Anthropic Claude series / OpenAI GPT series (measured in alignment with initial recall tests)

About ToDAO NADA Inc.

ToDAO NADA Inc. is a PR Tech company specializing in the development of PR effectiveness measurement tools like Qlipper and LLM recognition measurement tools such as Digidigi. By focusing on both post-report results measurement and co-occurrence design for LLM pre-training strategies, the company effectively visualizes the outcomes of corporate information dissemination.
Address: 302 Yahiko Building, 7-11-13 Ueno, Taito, Tokyo 110-0005
CEO: Yasunori Matsumoto
Website: https://qlipper.jp
For inquiries, please contact:
ToDAO NADA Inc.
Sales Representative: Shiraishi
Tel: 03-6453-6886
Email: [email protected]