Criteria for classifying free-form opinions in the job stress survey as anonymous

📌 First Answer: Deleting a name does not automatically make the free description anonymous.

In the free-form section of the job stress survey, even without a name, examples include “Last month, during the ○○ project, the team leader…”, “Out of the three people on the night shift, only I…”, and “After returning from parental leave…” Clues that can identify an individual ...can be included. The Personal Information Protection Act considers information to be personal information even if it cannot identify an individual on its own, if it can be easily combined with other information to identify an individual.

Therefore, the goal of anonymous classification is not simple deletion. Reorganizing the content to a level where it is difficult to identify the author, while retaining the meaning necessary for organizational improvement. no see.

🧭 It is safe to handle free writing in 4 steps.

1. Restriction on access to the original text — The original text is viewed by only a minimum number of people and is not kept in the general HR shared folder.
2. Removal of identification clues — Look for re-identification clues such as names, job titles, departments, projects, dates, rare events, and unique expressions.
3. Subject classification — Cod into organizational improvement categories such as workload, roles, relationships, manager feedback, change management, recovery, and system accessibility.
4. Rewriting the shared version — Replace quotations from the original text with summary sentences and aggregates focusing on 'what problems are recurring.'

🔎 What expressions can serve as identification clues?

🧩 'Anonymous information' and 'pseudonymized information' are not used interchangeably.

Under the Personal Information Protection Act, pseudonymized information is personal information processed in such a way that a specific individual cannot be identified without additional information, and special provisions may apply for purposes such as statistical compilation, scientific research, and public interest record preservation as prescribed by law. On the other hand, Article 58-2 stipulates that the Personal Information Protection Act does not apply to information where an individual can no longer be identified even if other information is used, taking into reasonable consideration of time, cost, technology, etc.

Even if referred to as 'anonymous opinion' in a working report, it is difficult to conclude that it is completely anonymous information in a legal sense if there remains a possibility of identifying the author again. Internal files Original / De-identified/Pseudonymized Version / Aggregated/Summary It is better to distinguish and display the processing status like this.

🗂️ We record 'work factors' rather than people in the classification table

✂️ There are many cases where 'generalization' is needed rather than deletion.

If too much free-form text is deleted, the organization cannot determine what needs to be changed, and if the original text is presented exactly as is, the author may be revealed. Therefore Preserve the meaning of the event and generalize unique identification clues The way it is done is useful.

👥 Minority groups handle free narrative more conservatively.

Groups with few members, such as teams of five or fewer, one person in a specific rank, one foreign employee, or one employee returning from leave, can be easily identified even if their names are omitted. In such cases, you may consider grouping by a higher-level organization or reporting by aggregating multiple rounds, rather than citing original documents at the departmental level.

The minimum personnel requirement is not set as a single number by law, but must be determined based on the size of the organization, response items, combined information, and reporting targets to report re-identification risk.

🧠 Even if AI or automatic classification is used, a final anonymity review is performed separately.

While free-form text can be grouped by topic using text classification models or generative AI, you must first verify which system the original text is being input into, as it may contain personally identifiable clues and health information. It is advisable to avoid unauthorized uploads to external AI services or operations that leave personally identifiable text in logs or training data.

As the Personal Information Protection Commission fully revises the guidelines for processing pseudonymized information in March 2026 Preparation → Risk Review → Pseudonymization → Appropriateness Review → Safe Management We presented a risk-based processing system. For unstructured data, such as free-form narratives, the step of human re-verification of re-identification clues is particularly important.

✅ Anonymous Classification Checklist

❓ FAQ

Recommended Posts

📚 Source and Verification Criteria

• Article 2 of the Personal Information Protection Act — Definition of Personal Information and Pseudonymized Information
• Article 58-2 of the Personal Information Protection Act — Exclusion of Application for Information That Can No Longer Identify an Individual
• Personal Information Protection Commission · Complete Revision of Pseudonymized Information Processing Guidelines (March 2026)

This document outlines general practical guidelines for classifying and sharing free-form opinions from job stress surveys for organizational improvement purposes. Whether information is "anonymous" is not determined solely by the deletion of names; factors such as the potential for combination with other information, organizational size, reporting scope, technology, and costs must also be considered. For projects where the actual level of pseudonymization or anonymization is critical, a review by a personal information protection officer or expert may be required. Date of verification of laws and official documents: September 30, 2026.