This paper proposes a systematic data mining approach to study users’ Internet resource access actions for finding out behavior models as state-transition graphs. A series of Internet resource access actions are stored in a database of [user, resource-access-action, time] records. Such access actions are treated as basic behavior elements and form an action hierarchy which possesses different levels of radix codes. For every user, the data sequence is divided into a series of transactions and all the actions in a transaction constitute a special behavior pattern, called (inter-transaction) behavior. The behavior codes can be aggregated as behavior hierarchy also. Accordingly, each user can possess his/her own behavior model, formulated as a state-transition graph with behavior states and transition probability between behaviors. The overall mining process is computerized and validated by experiment. The example uses simulated sequential data to show how to combine AprioriAll algorithm and the proposed algorithm to construct a set of nested state-transition graphs.
[1]
L. Rabiner,et al.
An introduction to hidden Markov models
,
1986,
IEEE ASSP Magazine.
[2]
Nicolas Halbwachs,et al.
Synchronous Programming of Reactive Systems
,
1992,
CAV.
[3]
Mohammed J. Zaki.
Efficient enumeration of frequent sequences
,
1998,
CIKM '98.
[4]
Anthony K. H. Tung,et al.
Breaking the barrier of transactions: mining inter-transaction association rules
,
1999,
KDD '99.
[5]
Ramakrishnan Srikant,et al.
Mining sequential patterns
,
1995,
Proceedings of the Eleventh International Conference on Data Engineering.
[6]
Heikki Mannila,et al.
Discovering Frequent Episodes in Sequences
,
1995,
KDD.
[7]
Ramakrishnan Srikant,et al.
Mining Sequential Patterns: Generalizations and Performance Improvements
,
1996,
EDBT.