Background

As the wave of digital governance rises, government agencies across Taiwan face growing demands for public opinion analysis. From citizen discussions on social media to policy commentary in news outlets, traditional manual monitoring methods can no longer keep pace with the rapidly shifting landscape of public sentiment. Government agencies urgently need a systematic AI-powered sentiment analysis tool to scientifically understand how the public truly feels about government policies.

However, agency procurement must be conducted in accordance with the Government Procurement Act and its related sub-regulations, making the process relatively rigorous. A Joint Supply Contract is one channel that simplifies the procurement procedure: a centralized procuring agency handles the tender in advance, and after the contract is signed, individual agencies can order directly from the listed items within the contract's validity period without re-tendering case by case. As a result, whether a solution is already included as a contract item often directly affects the administrative cost and timeline for an agency to adopt a new tool.

This section is a general explanation of the procurement system and does not refer to any specific provision or case determination. The actual scope of applicability and operational requirements should still be determined by the competent authority's latest announcements and your agency's (or your company's legal counsel's) own assessment.Laws & Regulations Database of the Republic of China (Taiwan)

Challenges Faced

• The volume of discussion pouring in daily from news media, social platforms, and forums far exceeds what case officers can read post by post — it's more than staff capacity can handle
• The absence of automated sentiment analysis technology makes it impossible to quickly assess positive and negative sentiment trends
• Inconsistent media monitoring report formats across departments make cross-departmental comparisons and informed decision-making difficult
• To simplify the process through the Joint Supply Contract channel, you first need to confirm whether the required items and specifications are already listed in the current contract
• Data security and personal data protection requirements are strict — the system must comply with the deployment method, access control, and acceptance checklist items designated by the agency

Industry Solutions

This government agency procured through the Joint Supply Contract channel to adopt LargitData's InfoMiner sentiment analysis platform. Because the case-by-case tendering process was avoided, it took only two weeks from requirement confirmation to going live. The actual items, specifications, and contract validity period that each agency can order are still subject to the current announcement on the Government e-Procurement Platform, so we recommend checking before procurement.

Implementation Overview

• InfoMiner Social Listening: provides comprehensive real-time media monitoring, sentiment analysis, trend tracking, and automated reporting capabilities
• Customizable Dashboard: configure dedicated monitoring topics and keyword combinations aligned with the agency's policy priorities
• Automated Daily/Weekly Report System: automatically generates daily sentiment intelligence summary reports, saving the time staff would otherwise spend writing reports manually
• Deployed to agency requirements: Can be configured to match the deployment method, account permissions, and access control designated by the agency, helping meet the agency's security and personal data protection requirements; actual compliance status must be verified by the agency's information security unit against its own checklist

InfoMiner connects to public sources such as Taiwan's major news media, PTT, and Dcard, while platforms such as Facebook and LINE communities are connected according to platform authorization and available scope, and coverage may change as each platform's policy changes. The system uses natural language processing (NLP) to perform Chinese sentiment analysis, classifying discussion as positive, negative, or neutral; the classification results are meant to assist interpretation, and posts involving irony, sarcasm, or content that depends heavily on context can still be misclassified — for important issues, we recommend pairing this with manual review.

InfoMiner Social Listening →

Implementation Results

• Report generation time reduced from 4 hours to 30 minutes, an 80% efficiency improvement
• 24/7 real-time sentiment alerts are now in place, enabling key decision-makers to be notified at the first sign of critical issues
• A unified media monitoring analysis platform is shared across departments, breaking down information silos and improving cross-departmental communication efficiency
• The joint procurement mechanism significantly streamlined the procurement process, with deployment completed within two weeks from initial request to go-live
• The system operated stably and passed the agency's internal security review and acceptance procedures, earning recognition from the using unit

How to read these numbers: measurement basis and preconditions

The figures above — 50+ sources, 95% satisfaction, and an improvement from 4 hours to 30 minutes — are measurement results from this project under a specific time period and specific issue setting. The source count is the number of connected media and social sites, not the number of data records, and doesn't represent coverage of all platforms; the satisfaction figure comes from an internal agency user survey, with the sample being case officers who actually use the system. Results will vary for other agencies depending on the complexity of their issues, reporting format requirements, and existing processes.

When evaluating similar solutions, it's worth clarifying a few questions first: does the report production time include manual review and supervisor sign-off; does the 24-hour alert clock start from when the original post was published or when the data was ingested; which social sources are connected via platform authorization and may change with policy adjustments; and which security review items the agency itself is responsible for. Clarifying these definitions first is what makes comparing numbers across different vendors meaningful.

Implementation and acceptance checklist

A public-sector monitoring project should begin with a written issue map rather than an unlimited keyword list. Each policy topic needs inclusion and exclusion terms, source scope, alert severity, responsible unit and review cadence. During a pilot, analysts should compare system results with a manually reviewed sample and record false positives, missed items, duplicates and source delays. This creates a defensible baseline before automated reports become part of an operational workflow.

Acceptance should test the full loop: collection, classification, dashboard, alert delivery, analyst review, report approval and audit trail. Role-based accounts must be tested with real permission boundaries, while alerts need fallback recipients and delivery logs. Agencies should also identify which configuration, source and report changes are included in maintenance, because a stable model alone does not keep a monitoring program current.

• Define topic scope, exclusions and escalation levels with the owning unit.
• Validate coverage and classification with a time-bounded human-reviewed sample.
• Test permissions, alert delivery, report approval and export logs end to end.
• Document source limitations, maintenance responsibilities and change procedures.
• Review precision, missed events and response time at a fixed monthly meeting.