UniStack

An experimental duplicate question search algorithm for Stack Exchange.

Current Algorithm

As this is an experimental project, everything is subject to change.

Extract question sentence + context words, then model remianing words.

• Question sentence: words starting from an MD or Wxx tagged word till the end of the sentence.
• Context words: any NN tagged word.
• Model: PoS tagged bag-of-words (unigram).
• Weighting: TF-IDF.
• Similarity function: cosine.