Method Similarity

libs/search/doc_vec.cpp:117–217

}

double QueryVec::Similarity(IdfMap & docIdfs, DocVec const & rhs)
{
  size_t kInvalidIndex = numeric_limits<size_t>::max();

  if (Empty() && rhs.Empty())
    return 1.0;

  if (Empty() || rhs.Empty())
    return 0.0;

  vector<size_t> rsMatchTo(rhs.GetNumTokens(), kInvalidIndex);

  double dot = 0;
  {
    size_t i = 0, j = 0;

    while (i < m_tfs.size() && j < rhs.GetNumTokens())
    {
      auto const & lt = m_tfs[i].m_token;
      auto const & rt = rhs.GetToken(j);

      if (lt < rt)
      {
        ++i;
      }
      else if (lt > rt)
      {
        ++j;
      }
      else
      {
        dot += GetFullTokenWeight(i) * rhs.GetWeight(docIdfs, j);
        rsMatchTo[j] = i;
        ++i;
        ++j;
      }
    }
  }

  auto const ln = Norm();
  auto const rn = rhs.Norm(docIdfs);

  // This similarity metric assumes that prefix is not matched in the document.
  double const similarityNoPrefix = ln > 0 && rn > 0 ? dot / sqrt(ln) / sqrt(rn) : 0;

  if (!m_prefix)
    return similarityNoPrefix;

  double similarityWithPrefix = 0;
  auto const & prefix = *m_prefix;

  // Let's try to match prefix token with all tokens in the
  // document, and compute the best cosine distance.
  for (size_t j = 0; j < rhs.GetNumTokens(); ++j)
  {
    auto const & t = rhs.GetToken(j);
    if (!strings::StartsWith(t.begin(), t.end(), prefix.begin(), prefix.end()))
      continue;