Estimation of similarity of items

Abstract

Similarity of items can be estimated by using a method including generating a prediction model that predicts an indicator of a target based on one or more attributes for items, by estimating a weight set, among weight sets, for each of the items, and estimating a similarity among the items for the target based on the weight sets of the prediction model.

What is claimed is: 
     
         1 . A method comprising:
 generating a prediction model that predicts an indicator of a target based on one or more attributes for each of a plurality of items, by estimating a weight set, among a plurality of weight sets, for each of the plurality of items;   estimating a similarity among the plurality of items for the target based on the plurality of weight sets of the prediction model; and   displaying on a display device the indicator and the one or more attributes of the plurality of items in accordance with the similarity.   
     
     
         2 . The method according to  claim 1 , wherein the generating of the prediction model includes estimating one or more weights for the target, wherein the prediction model predicts the indicator of the target further based on the one or more weights for the target. 
     
     
         3 . The method according to  claim 1 , wherein the generating of the prediction model includes estimating a posterior probability distribution of each of the plurality of weight sets. 
     
     
         4 . The method according to  claim 3 , wherein the estimating of the posterior probability distribution of each of the plurality of weight sets is performed by using a Gaussian distribution as a prior probability distribution of the plurality of weight sets. 
     
     
         5 . The method according to  claim 4 , further comprising:
 obtaining an input data comprising one or more data sets, each data set including the indicator, the one or more attributes of the plurality of items and one or more attributes for the target,   wherein the estimating of the posterior probability distribution of the plurality of weight sets is performed by calculating the posterior probability distribution p(w/D) according to:
     p ( w|D )= N ( w|m   D   , S   D ), 
   where D≡{(x 1 , y 1 ), (x 2 , y 2 ) . . . (x N , y N )}, N is the number of the data sets, x n  (1≤n≤N) is a vector including the one or more attributes for the target and M attribute sets for M items in an n-th data set, y n  is the indicator of the target in the n-th data set, w is a vector including one the or more weights for the target and M weight sets for M items, m D  is a vector derived from S D , a sum of multiplication of x n , and y n  for n=1, . . . , N, and S D  is a matrix derived from a sum of multiplication of x n   T  and x n  for n=1, . . . , N.   
     
     
         6 . The method according to  claim 2 , wherein the estimating of the similarity among the plurality of items includes estimating a similarity between two items among the plurality of items. 
     
     
         7 . The method according to  claim 6 , wherein the estimating of the similarity between two items is performed based on a distance between the two weight sets of the two items. 
     
     
         8 . The method according to  claim 6 , wherein the estimating the similarity between two items is performed based on at least one of a square distance, a Gaussian similarity, Normalized inner product similarity, and a Cosine similarity between the two weight sets of the two items. 
     
     
         9 . The method according to  claim 5 , wherein the estimating of the similarity among the plurality of items includes calculating the similarity s(i, j) of an item i and an item j according to:
     s ( i, j )=∫ w   f ( w   (i)   , w   (j) ) p ( w|D ) dw.  
   where f(w (i) , w (j) ) is the distance between the two weight sets of the item i and the item j.   
     
     
         10 . The method of  claim 1 , further comprising:
 grouping the plurality of items into two or more groups based on the similarity among the plurality of items.   
     
     
         11 . The method of  claim 1 , wherein the indicator of the target is a performance indicator of one or more stores, and the one or more attributes of the plurality of items are the one or more attributes of the plurality of retailers that are competitors of the one or more stores.