ITEM RECOMMENDATION DEVICE, ITEM RECOMMENDATION METHOD AND ITEM RECOMMENDATION PROGRAM

• Main Authors: TODA HIROYUKI, TSUTSUMIDA KYOTA, KANEKIYO TOMOYUKI, NAKAYAMA JOJI, MAEBASHI KARIN
• Format: Patent
• Language: English; Japanese
• Published: 11.06.2015
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

PROBLEM TO BE SOLVED: To acquire a recommendation result at higher speed by reducing computational complexity in item recommendation using RWR.SOLUTION: An item recommendation device using RWR includes: transition matrix generation means 130 for generating a transition matrix representing a utilization history relation between a user and an item and a relation between the item and metadata while using item history storage means 120; relation degree vector calculation means 140 in which it is discriminated for each utilization item of a recommended user acquired from the item history storage means 120 whether a relation degree vector of the utilization item is stored in relation degree vector storage means 150, the relation degree vector is acquired if stored, a relation degree vector based on RWR of the utilization item is calculated while using the transition matrix if not, and the relation degree vector is stored in the relation degree vector storage means 150; and relation degree vector coupling means 160 for calculating a weighted linear sum of the relation degree vector and outputting an item with a high relation degree as a recommendation item. 【課題】RWRを用いたアイテム推薦における計算量を削減し、より高速に推薦結果を取得する。【解決手段】RWRを用いたアイテム推薦装置であって、アイテム履歴記憶手段１２０を用いて、ユーザとアイテムの利用履歴関係およびアイテムとメタデータの関係を表す遷移行列を生成する遷移行列生成手段１３０と、アイテム履歴記憶手段１２０から取得した被推薦ユーザの利用アイテム毎に当該利用アイテムの関連度ベクトルが関連度ベクトル格納手段１５０に格納されているか否かを判別し、格納されている場合は当該関連度ベクトルを取得し、格納されていない場合は当該利用アイテムのRWRによる関連度ベクトルを遷移行列を用いて算出し、関連度ベクトル格納手段１５０に格納する関連度ベクトル算出手段１４０と、関連度ベクトルの重み付き線形和を算出し、関連度が高いアイテムを推薦アイテムとして出力する関連度ベクトル結合手段１６０とを有する。【選択図】図１