search - Combining TF-IDF (cosine similarity) with pagerank? -
search - Combining TF-IDF (cosine similarity) with pagerank? -
given query have cosine score document. have documents pagerank. there standard way of combining two?
i thinking of multiply them
total_score = cosine-score * pagerank because if low on either pagerank or cosine-score, document not interesting.
or preferable have weighted sum?
total_score = weight1 * cosine-score + weight2 * pagerank is better? might have 0 cosine score, high pagerank, , page show among results.
the weighted sum improve ranking rule.
it helps break problem retrieval/ filtering step , ranking step. problem outlined weighted sum approach no longer holds.
the process outlined in this paper sergey brin , lawrence page uses variant of vector/ cosine model retrieval , seems kind of weighted sum ranking weights determined user activity (see section 4.5.1). using approach document 0 cosine not pass retrieval/ filtering step , not considered ranking.
search search-engine tf-idf cosine-similarity
Comments
Post a Comment