| 摘要 |
A subscriber portfolio model is created by forming mutually exclusive and completely exhaustive groups of subscribers in a way that maximizes the intra-group similarity and inter-group dissimilarity of the value of a target attribute among subscribers in the groups. The subscriber groups thus generated are self-descriptive, as the groups are defined by the attributes used to form them. A segmenting algorithm groups subscribers along permutations of attributes. Relevant data about subscribers is collected from various sources, correlated, and subscriber identities removed to ensure privacy. After analyzing the aggregated data, a subscriber portfolio model is built. The subscriber portfolio model is created in an iterative process, where in each step, a selected subscriber group is subdivided into smaller subscriber groups along different value ranges of a statistically decisive attribute that is selected and evaluated based on specific methods. The subdivision is performed by computing filtering conditions based on an in-depth statistical analysis.
|
