bestblogs-content-reviewer

BestBlogs Content Scoring Review Workflow

Skill Overview

The BestBlogs Content Scoring Review Workflow is an intelligent content quality audit tool used to review articles and tweets that have already been scored but not yet reviewed in the system each day. It evaluates the accuracy of AI scores, corrects scoring biases, and recommends high-quality content worth reading in depth from the items awaiting review.

Use Cases

1. Daily Content Quality Review

Review the scores of newly added articles and tweets in the BestBlogs system every morning and evening. Through AI-based evaluation and classification, the workflow ensures the accuracy of content scores. It supports parallel retrieval of content awaiting review, processes up to 200 items, and sorts them from highest to lowest score, helping content administrators quickly complete their daily review tasks.

2. Scoring Accuracy Evaluation and Correction

Systematically check for bias in AI scores based on source priority (HIGH/MEDIUM/LOW), score ranges (95+ milestones, 90–94 top-tier, 85–89 high-quality recommendations, 80–84 acceptable for display, etc.), and deduction rules (compiled or paraphrased content capped at 89, marketing-oriented content below 75, etc.). Content with scores that are too high or too low is corrected to ensure consistency and fairness across the scoring system.

3. Curating Today’s Reading List

Intelligently select articles and tweets worth reading in depth from the content awaiting review. Recommendations are divided into three priority levels: Must Read (3–5 items), Recommended (8–12 items), and Optional (5–8 items). Each recommendation includes a detailed reason (15–30 Chinese characters) explaining the content’s unique perspective, practical techniques, or value in revealing industry trends, thereby providing high-quality reading input for the deep-reading workflow.

Core Features

1. Intelligent Content Review and Classification

Systematically inspect each piece of content based on unified scoring rules, including source priority verification (LOW sources capped at 89, HIGH sources generally scoring no lower than 80), score-range alignment checks (95+ milestone events, 90–94 top-tier content, etc.), deduction red-flag checks (compiled or paraphrased content, introductory tutorials, marketing-oriented content, etc.), and content quality signal evaluation. Review results are automatically classified into four categories: 🌟 Recommended Reading, ✅ Score Reasonable, ⬆️ Score Too Low, and ⬇️ Score Too High, providing clear decision-making guidance for subsequent processing.

2. Structured Review Results

Output separate review tables for articles and tweets, grouped by category and including the title, source, score, suggested adjustment, and detailed reasoning. Each table entry provides an admin review link (https://admin.bestblogs.dev/article/review/{id}) for quick access to the original content. The workflow also outputs an overview of score statistics, including the score range, average score, score distribution, domain distribution, and identified scoring biases, helping administrators gain a comprehensive understanding of content quality.

3. Batch Marking and Score Adjustment

For confirmed items that are not recommended reading, batch-call the markNotQualified endpoint to mark them and adjust their scores. The workflow supports parallel execution of up to five requests per batch and outputs progress for each batch (for example, ✅ [1/30] RAW_xxx — Title — adjustScore: -3 → New score: 82), providing clear execution feedback. A single failure does not interrupt the batch; processing is paused after three consecutive failures and a possible systemic issue is reported. Recommended reading items retain a qualified: null status, producing a clean reading list for the deep-reading workflow.

Frequently Asked Questions

What are BestBlogs’ scoring standards?

BestBlogs uses a unified scoring system that evaluates content based on source priority and content quality. Source priority is divided into three levels—HIGH, MEDIUM, and LOW—each corresponding to different score ceilings and floors (HIGH sources can reach 95+, while LOW sources are capped at 89). Score ranges include: 95+ (milestone events, <1%), 90–94 (top-tier, ~5%), 85–89 (high-quality recommendations, ~15%), 80–84 (acceptable for display, ~25%), 75–79 (borderline, ~20%), and <75 (low quality, ~35%). Deduction rules include: compiled or paraphrased content capped at 89, introductory tutorials at 75–79, marketing-oriented content below 75, and low-quality aggregations at 70–76.

How do I use BestBlogs for daily content review?

Daily BestBlogs review consists of four stages:

Stage 1: Retrieve articles and tweets awaiting review in parallel, up to 200 items.

Stage 2: Use the scoring rules for AI-based evaluation and classification, checking source priority, score-range alignment, deduction red flags, and other dimensions.

Stage 3: Output structured review tables and a recommended reading list, then wait for user confirmation.

Stage 4: Execute batch marking operations, calling the markNotQualified endpoint to mark and adjust the scores of non-recommended content.

The entire workflow supports parallel processing and provides detailed progress updates and execution summaries.

How does BestBlogs correct AI scoring bias?

BestBlogs identifies and corrects AI scoring bias through a multidimensional review mechanism. First, it checks whether the source priority matches the score (a LOW-source score of 90+ is almost certainly too high, while a HIGH-source score below 80 may be too low). Next, it verifies whether the score range is reasonable (95+ should represent an industry milestone, while scores below 75 should represent low-quality content). It then checks for deduction red flags, such as compiled or paraphrased content scoring above 89 or marketing-oriented content receiving a high score. Finally, it evaluates content quality signals, including word count, reading time, and title clarity.

Based on the review results, content is classified as 🌟 Recommended Reading, ✅ Score Reasonable, ⬆️ Score Too Low, or ⬇️ Score Too High, with specific adjustment recommendations: minor adjustments of ±1–2, moderate adjustments of ±3–5, and substantial adjustments of ±6–10.