This repo seeds the reference account dataset used to drive style extraction for automated X content creation.

• data/reference_accounts.json contains the reference account set (username, niche, follower count, avg engagement rate, weight).
• data/reference_tweets.json contains the most recent fetched tweets for style analysis.
• data/topic_stream.json stores raw tweets collected for topic discovery.
• data/topic_candidates.json stores extracted topic candidates prior to validation.
• data/topic_signals.json stores scored topic signals with detailed scoring metadata.
• data/approved_topics.json stores approved topics that pass the scoring threshold.
• data/recent_topics.json stores recently used topics for deduplication.
• data/topic_discovery_config.json holds the topic discovery configuration.
• data/candidate_viral_tweets.json stores raw candidates before engagement normalization filtering.
• data/viral_tweet_dataset.json stores normalized, labeled, high-performing tweets for training.
• data/viral_tweet_config.json holds the viral tweet dataset configuration.
• data/tweet_patterns.json stores extracted structural templates for tweet generation.

The fetcher uses the X API v2. Set X_BEARER_TOKEN and run:

python scripts/fetch_reference_tweets.py --max-results 10 --update-accounts

Engagement rate is computed as:

avg_engagement_rate = avg((likes + replies + retweets + quotes) / follower_count) across recent tweets

Accounts stay in pending_engagement_verification until fresh tweets are pulled and rates are computed.

The topic discovery pipeline scans reference accounts and keyword searches to surface early, controversial, or high-engagement discussions. It writes the topic_stream, topic_candidates, topic_signals, and approved_topics datasets.

Run with:

python scripts/topic_discovery.py

You can adjust keywords, lookback window, signal thresholds, or scoring weights in data/topic_discovery_config.json.

Build the viral tweet dataset from manual CSV input with:

python scripts/build_viral_tweet_dataset.py

The script reads data/manual_viral_tweets.csv, writes data/candidate_viral_tweets.json, and promotes tweets into data/viral_tweet_dataset.json after normalization.

Fill data/manual_viral_tweets.csv with tweets you collect manually from the X UI and run:

python scripts/build_viral_tweet_dataset.py

Required columns are text, author_followers, like_count, reply_count, retweet_count, quote_count. Optional columns are tweet_id, url, created_at, author_username, author_id, impression_count, source_type, source_query.

Once the viral dataset is populated, extract reusable structure templates with:

python scripts/build_tweet_patterns.py

Rank generated tweet candidates with:

python scripts/rank_tweets.py --top 12