detect-trends

name	detect-trends
version	2.0.0
description	Detect trending topics across multiple articles by analyzing entity co-occurrence and cross-source mentions. Uses memory search to find recent articles and LLM analysis to identify emerging trends.
license	MIT
allowed-tools	["use_llm","memory/search","memory/add","memory/link"]
metadata	{"domain":"news","category":"analysis","requires-approval":false,"confidence":0.85,"mcp-servers":[]}

Detect Trends

Detect trending topics across multiple analyzed articles.

When to Use

Use this skill when you need to:

• Identify topics mentioned across multiple sources
• Detect emerging trends in AI news
• Track topic momentum over time
• Prioritize topics for digest composition

Instructions

Step 1: Search for Recent Analyses

Use memory/search to find recent article analyses:

• namespace: "news/analyses"
• query: "recent AI news analysis"
• limit: 100 (or appropriate window)

This returns articles with their extracted entities and categories.

Step 2: Extract Entity Mentions

From the search results, collect all entities and track:

• Which articles mention each entity
• How many times each entity appears
• Which sources mention each entity

Entity normalization:

• Lowercase and strip whitespace
• Handle variations (e.g., "GPT-4" = "GPT4" = "gpt-4")

Step 3: Calculate Trend Scores

For each entity, calculate a trend score:

Score formula:

score = mention_count * source_diversity_bonus
source_diversity_bonus = 1.0 + (unique_sources - 1) * 0.2

Thresholds for trend qualification:

• Minimum 2 mentions
• Minimum 2 different articles

Step 4: Analyze with LLM

Use use_llm to refine trend detection:

Trend analysis prompt:

Given these entity mention statistics from recent AI news:

{entity_stats}

Identify the top trending topics and classify each as:
- breaking: Rapidly emerging (momentum > 5)
- hot: Actively trending (momentum 2-5)
- rising: Emerging (momentum 0-2)
- established: Stable coverage
- fading: Declining interest

Filter out generic terms like "AI", "technology", "company".

Return as JSON list with: topic, status, mention_count, related_topics

Step 5: Store Trends in Memory

For each identified trend, use memory/add:

• type: "trend"
• namespace: "news/trends"
• data: {topic, status, article_count, mention_count, momentum, related_topics}
• metadata: {detected_at, source_articles}

Step 6: Link Trends to Articles

Use memory/link to connect trends to source articles:

• source_id: trend ID
• target_id: article ID (for each contributing article)
• relation_type: "DETECTED_FROM"

Step 7: Return Results

Return trending topics including:

• Topic name
• Status (breaking, hot, rising, established, fading)
• Mention count
• Source articles
• Related topics

Tool Usage Guidance

memory/search

• Search namespace "news/analyses" for recent analyses
• Use broad query to capture all recent content
• Limit appropriately for time window

use_llm

• Use for trend classification and noise filtering
• Provide entity statistics as context
• Request structured JSON output

memory/add

• Store trends as type "trend"
• Include momentum and status

memory/link

• Create DETECTED_FROM relationships
• Links each trend to contributing articles

Trend Status Definitions

Breaking (Momentum > 5)

• Rapidly emerging topic
• Mentioned in 10+ articles recently
• Requires immediate attention

Hot (Momentum 2-5)

• Actively trending topic
• High current interest

Rising (Momentum 0-2)

• Emerging topic gaining traction
• Growing interest

Established (Momentum ≈ 0)

• Stable topic with consistent coverage
• Ongoing interest

Fading (Momentum < -1)

• Topic losing relevance
• Declining interest

Trend Data Schema

{
  "id": "trend-abc123",
  "topic": "GPT-5",
  "status": "breaking",
  "article_count": 15,
  "mention_count": 23,
  "momentum": 8.5,
  "related_topics": ["OpenAI", "AGI", "language models"],
  "source_articles": ["article-1", "article-2", "..."],
  "detected_at": "2026-01-31T12:00:00Z"
}

Noise Filtering

Filter out generic terms that aren't meaningful trends:

• "AI", "artificial intelligence", "machine learning", "ML"
• "technology", "tech", "company", "research"
• "model", "system", "data", "algorithm"

Error Handling

• If search returns no results, return empty trends
• If LLM analysis fails, use raw entity counts
• Log but continue if memory operations fail

Success Criteria

• Trends accurately reflect current news landscape
• Breaking/hot topics are identified correctly
• Noise is filtered out (no generic terms)
• Trends are linked to source articles