A Data Mining Based Publish/Subscribe System over Structured Peer-to-Peer Networks - Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing - page 5

S NF
max load
where S NF is the percentage of attribute sets which contain frequent itemset, and
maxload is the maximal matching load threshold that is specified by users. When
the value of maxload is smaller, the number of virtual frequent itemsets is larger.

1 −

M ≥

3.2 Peer Initialization

In order to make full use of nodes and improve scalability of the system, the DHT-
based overlay network is applied in DMPSS. As previously described, DMPSS can
be deployed on any DHT-based overlay network, such as Chord, CAN, Pastry, etc.
In DMPSS, after a peer joins in, it is initialized according to the DHT network that
the user selects, and then it downloads frequent itemsets and their supports, as well
as virtual frequent itemsets from existing peers.

3.3 Subscription Installation

A subscription is installed in terms of an attribute set which is a frequent itemset
or a virtual frequent itemset, as shown in Algorithm 1. The attribute set is obtained
as follows. The subscriber extracts the attribute set Att r sub from sub , and compares
it with frequent itemsets. If Att r sub doesn't contain any frequent itemset, then one
virtual frequent itemset is selected randomly. If Att r sub contains one frequent item-
set, then this frequent itemset is the selected attribute set. If Att r sub contains more
than one frequent itemsets, then the frequent itemset with the smallest support is se-
lected. After the attribute set is specified, a string is obtained by bunching the names
of attributes in the attribute set lexicographically. Then a key K is produced by hash-
ing the string. Finally, the subscription is delivered to the related node in the overlay
network. When more than one frequent itemsets are contained in Att r sub , selecting
the one with the smallest support can keep matching load balancing among RPs.
In DMPSS, each subscriber records all the subscriptions it sends and their corre-
sponding keys. When a subscriber wants to cancel a subscription, it sends a cancel
message to the node according to the key in recorder. Moreover, in order to avoid
losing subscription information due to node departure, we suggest backing up sub-
scriptions on neighbor nodes in the overlay network.

3.4 Event Publication

An event is published in terms of an attribute set list, as shown in Algorithm 2.
The attribute set list is obtained as follows. The publisher extracts the attribute set
Att r e from e , and checks if there are frequent itemsets are contained in Att r e .All
the frequent itemsets that are contained in Att r e are added to the attribute set list.
Besides, all the virtual frequent itemsets are added to the attribute set list no matter
whether they are contained in Att r e . Assuming there are N attribute sets in the list,