SECRET IDENTITYWEEK 5 · THE BAG OF WORDS KNOWS WHO YOU ARE

0score

0machine

0streak

–round

This case

U unmaskI impostorR redacted

How the machine plays

It sees the same clues you do and the same four suspects. Each suspect is a bag of words: the counts on their page with stopwords removed, scaled to length one. The machine scores a suspect by adding up that suspect's weights for the clue words revealed so far, which is the cosine between your clue list and the page. It reads at least two clues, then commits as soon as one suspect scores twice the runner-up, or is forced to guess on the tenth clue; a wrong commitment scores zero, like yours. It never sees the answer. You see when it commits, never to whom, until you commit yourself.

Where the words come from

The 303 full Wikipedia pages from the course's week 5 release, tokenized with spaCy, lowercased, letters only. A character's own names are never clues. Raw counts is every word; no stopwords drops NLTK's 198 English function words; distinctive ranks by count × ln(303 / pages containing the word), which is next week's TF-IDF a week early. Impostor sentences come from one trigram model per Louvain community; a generated sentence is rejected if it shares any nine-word run with the corpus. Blanks are concordance lines; two of the three decoys are words that share left-and-right contexts with the answer, NLTK's similar().

Pictures on the reveal are the lead images of the Wikipedia articles, loaded from Wikipedia and credited; only their URLs are stored. Keyboard: 1–4 pick, Space for another clue or next, R / M for real / Markov.

Week 5 · from language to numbers

SECRET IDENTITY

Every one of the 303 Marvel pages is a bag of words. Unmask the character from as few words as you can, tell a real Wikipedia sentence from one a trigram model dreamed up, and put the blanked word back into a concordance line. A bag-of-words machine plays the same case and its score sits next to yours.

Case length

Case types

Clue order for unmask rounds

The other three suspects

Suspects drawn from

CASE OF THE DAYthe same 12 rounds for everyone who opens it on

SECRET IDENTITY

CASE CLOSED