An experimental duplicate question search algorithm for Stack Exchange.
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.