Word Frequency Cloud

Our word frequency cloud generator processes your text through a six-stage pipeline — from raw text to an interactive, exportable visualization. Everything runs client-side in your browser, so your data never touches a server.

Stage 1: Tokenization

The first step splits text into tokens using Unicode-aware regular expressions. It recognizes letter sequences from many scripts, but does not perform language-specific word segmentation. In languages that commonly omit spaces between words, a long sequence may be treated as one token. The tokenizer preserves:

• Contractions — "don't", "it's", "we'll" stay as single tokens
• Hyphenated words — "well-being", "state-of-the-art" remain intact
• Numbers — optionally included or excluded via a toggle
• Accented characters — letters such as "café", "naïve", and "résumé" are included in tokens

Leading and trailing apostrophes or hyphens are stripped, and by default all tokens are lowercased so "Hello" and "hello" are counted together. Enable case-sensitive mode to treat them as separate words.

Stage 2: Stop Word Filtering

Common English function words — articles, prepositions, pronouns, and auxiliary verbs like "the", "is", "at", "which", and "on" — can be filtered using the built-in English stop-word list. This list may not suit every text or language.

• Toggle filtering on or off with a single switch
• View the full list via the expandable "stop words" panel
• Add custom exclusions — enter your own words or phrases (comma or newline separated) to filter out domain-specific terms, character names, or any noise

For n-gram analysis (bigrams and trigrams), stop word filtering works slightly differently: phrases that start or end with a stop word are removed (e.g. "the machine" and "learning the" are filtered, but "machine learning" is kept). This preserves meaningful phrases while discarding noise.

Stage 3: N-Gram Generation

Beyond single-word analysis, the tool can detect bigrams (two-word phrases) and trigrams (three-word phrases). This reveals meaningful collocations that single-word counts miss:

• Bigrams uncover pairs like "machine learning", "customer service", "data science"
• Trigrams capture longer phrases like "artificial intelligence", "user experience design"

N-grams are generated using a sliding window over the token sequence, then filtered for stop words at the phrase boundaries. N-grams can help you inspect recurring adjacent phrases in a text.

Stage 4: Frequency Counting

Each surviving token (or n-gram) is counted. The tool calculates:

• Raw count — how many times each word appears
• Percentage share — the word's proportion of all analyzed tokens
• Rank — sorted by frequency (descending), with alphabetical tie-breaking

A minimum frequency filter lets you surface only words that appear at least N times — useful for large texts where you want to ignore one-off mentions. A minimum word length filter removes short tokens that may slip through.

Stage 5: Cloud Layout Algorithm

Rendering the cloud is the most technically interesting step. The tool uses an Archimedean spiral placement algorithm with axis-aligned bounding box (AABB) collision detection:

1. Sort words by frequency (highest first)
2. Scale font sizes logarithmically — the most frequent word gets the largest font, the least frequent gets the smallest, with a logarithmic distribution that prevents the smallest words from becoming unreadable
3. Measure each word's pixel dimensions using a hidden HTML5 canvas context for accurate width/height calculation (not a rough character-count estimate)
4. Place the first (largest) word at the center of the canvas
5. For each subsequent word, spiral outward from the center in an Archimedean pattern, testing each position for collisions with already-placed words
6. When a non-overlapping position is found (within canvas bounds), the word is placed; if the spiral completes without finding room, the word is skipped
7. Color is assigned from the selected scheme, and rotation is applied based on the orientation setting (horizontal, mixed with periodic vertical rotation, or fully vertical)

The layout checks word bounding boxes for overlap and may skip terms that do not fit. Its estimated text dimensions and SVG rendering can still make spacing differ slightly across browsers.

Stage 6: Interactive Display and Export

The cloud renders as a scalable SVG that responds to hover — moving your cursor over any word dims the others and reveals a tooltip with the exact occurrence count. A companion frequency table provides the raw data: rank, word, count, a visual frequency bar, and percentage share, all searchable and scrollable.

When you're ready to use the results elsewhere, export with one click:

• CSV — spreadsheet-ready file with word, count, and percentage columns
• JSON — structured data for programmatic use or API integration
• SVG — vector image of the cloud itself, scalable to any resolution
• Copy — copy the word:list pairs to your clipboard for quick pasting

Tokenization, counting, layout, and export are handled in your browser.