自己組織化マップによる行動履歴の類型化 : クレジットカード利用履歴を利用したキャッシング移行予測  [in Japanese] Behavior Pattern Extraction by Self-Organizing Maps of Personal Usage Histories : Predicting when credit-card users will switch to credit-card cashing based on personal credit histories  [in Japanese]

Access this Article

Search this Article

Author(s)

Abstract

本研究では,多量の個人履歴データから,顧客セグメントを抽出する次のような方法論を提案する.まず,単位時間ごとの行動のありようを自己組織化マップ(SOM)で類型化する.その上で,個人がどの類型に何回当てはまったかという頻度分布を,分布間距離を利用して再度SOMで類型化することで,観察された長期間の行動全体を類型化する.さらに,変量間の関連の共通性に着目し,得られたSOMによる行動類型を共通の一般化線形モデルで説明可能になる場合に統合するモデル統合分析で,顧客セグメントを与える.以上の方法により,長期間多面的に,かつ,悉皆的に収集された履歴データの分析方法を与える.具体例として,クレジットカード利用者のうち,どのような顧客がキャッシング利用に移行しやすい顧客であるかを識別する問題を扱う.各種ショッピング/キャッシング利用の月額や残高から,クレジットカード年間利用類型を与え,この利用類型にロジスティック回帰モデルでモデル統合分析を行なうことにより,キャッシング利用をプロモートする上で着眼すべき顧客特性を,作成された顧客セグメントごとに明らかにした.

In this paper, we propose a method to extract a customer segmentation from the voluminous data produced by personal usage histories. First, in order to categorize customer behavior types at specified points in time, we apply SOM (self-organizing maps) to usage records. Second, we define new distances between distributions of behavior types on the SOM map using distribution functions. We then map the distributions of behavior types in order to obtain customer types over the long-term. Finally, in order to obtain a customer segmentation whose segment has a homogeneous functional relation between variates, we propose a model merge method. This method merges neighboring customer types in an SOM map, in cases where the MDL (minimum description length) criterion of the generalized linear model on the merged customer type is smaller than the sum of MDLs of the models on customer types to be merged. In this way, we are able to analyze historical personal usage data exhaustively gathered from multiple sources over the long-term. In order to validate the proposed method, we predict which credit-card users are likely to switch to credit-card cashing based on their credit histories. First, we classify card users into monthly types by applying SOM to monthly usage records of card shopping and cashing. Next, using the distances between distributions over the map, we apply SOM to the distributions of monthly types, thus obtaining yearly customer types. Finally, to predict the switch, we estimate logistic models on yearly customer types, and combine these models using the model merge method. The proposed method reveals useful features of card users who switch to credit-card cashing.

Journal

  • Journal of Japan Industrial Management Association

    Journal of Japan Industrial Management Association 57(5), 404-412, 2006

    Japan Industrial Management Association

References:  8

Cited by:  2

Codes

  • NII Article ID (NAID)
    110007521688
  • NII NACSIS-CAT ID (NCID)
    AN10561806
  • Text Lang
    JPN
  • Article Type
    Journal Article
  • ISSN
    1342-2618
  • NDL Article ID
    8587397
  • NDL Source Classification
    ZD23(経済--企業・経営)
  • NDL Call No.
    Z4-298
  • Data Source
    CJP  CJPref  NDL  NII-ELS  J-STAGE 
Page Top