Dynamic Word Cloud Generator

Abstract

A small text-corpus visualiser built as a teaching exercise on top of nltk, scikit-learn, and the wordcloud library. Tokens are weighted by TF-IDF rather than raw frequency so that domain-specific terms surface above generic high-frequency words. The renderer respects an arbitrary alpha mask (a Pokémon Arceus silhouette, a fall-leaf shape, a Mexico outline), and an animated variant updates the layout frame-by-frame so the cloud "grows" as the corpus is fed in. This is an exploration / demonstration project, not a research result.

Pipeline

1. Tokenise & clean. Lowercase, strip punctuation, drop English stop-words from nltk.corpus.stopwords, lemmatise with WordNetLemmatizer.
2. TF-IDF weighting. Run TfidfVectorizer on the cleaned token stream against a background corpus (English Wikipedia sample). The score for term t in document d is:

   tf-idf(t, d)  =  tf(t, d) · log(N / df(t))

   Where tf is term frequency in d, N is total documents in the background corpus, and df is document frequency of t. This dampens generic English words and amplifies domain-specific ones.
3. Mask-aware layout. Pass the binary alpha channel of an input PNG (Arceus silhouette, leaf shape, Mexico outline) into WordCloud(mask=...). The layout engine packs words inside the mask; pixels with alpha = 0 are forbidden.
4. Animation. The dynamic variant feeds the corpus in chunks and re-renders the cloud each chunk. Frames are stitched into a GIF with imageio.mimsave.

Built collaboratively with Connor Carpenter, Ryan Lay, Samyak Karnavat, and Yash Shah.