Relevance Feedback Method Based on Vector Space Basis Change

Vector Space Basis Change in Information Retrieval

“…Finally, we compare our proposed model with all existing VSBC-based models. We show that our model has better performance over models of [8,20,15,16,18,17]…”

“…According to M elucci [20], a context is modeled by a vector space basis and its evolution is mod eled by a VSBC. M eluccideveloped a new contextbased model called IRiX: if B is a basis which describes a context, L (B ) is the event that a vector belongs to the subspace spanned by B and PB is a projector to this subspace, then the probability that a vector y is in the context described by B is [15,16] developed RF algorithms based on a VSBC. These RF algorithms improve the results of known models (BM25 model, Rocchio model).…”

“…This transition matrix should minimize the sum (S1 ) of squared distances between each relevant document and gR ( gR is the centroid of relevant documents) and should maximize the sum (S2 ) of squared distances between each irrelevant docu ment and gR. According to [15] (IBM1 model), this transition matrix should minimize the quotient…”

“…With respect to the initial vector space basis (index terms), relevant and irrelevant documents share some terms (at least the terms of the query which selected these documents). According to [15,16], the technique of VSBC is effective in separating relevant documents and irrelevant ones.…”

“…The best framework that could bring the VSBC technique into application is RF: the user shows relevant and irrelevant documents in an initial rank ing and instead of reformulating the query, we change the vector space basis in which it is written (as well as the documents). In [15,16] from the relevant ones. These approaches were evaluated on a RF framework and on a Pseudo Relevance Feedback (PRF) framework in [18].…”