Information Technology Reference
In-Depth Information
4
Experimental Results
This section describes a set of experiments performed to assess the benefit of our
approach. We compare the method proposed in [7] to assign MIS values to items with
our approach. The method [7] is as follows:
MIS
(4)
The is the actual frequency of item in the TDB. MIN denotes the user-
specified least minimum item support value of all items. σ01 is a parame-
ter that controls how the MIS values for items should be related to their frequencies.
If σ0 , we have only one min_sup, MIN, which is the same as the traditional fre-
quent pattern mining. If σ1 and , is the MIS value for item
.The approaches are implemented in Java and all the experiments are performed in
an Intel Core i7 2.90GHz with 7.7 GB of memory, and running on Windows 7.
For our experiments, we generated a number of data sets from the tourism informa-
tion database to test our approach. For each point of interest (poi, in short), we used
other four categories from ontology, who, when, why and what, which are utilized to
describe poi to construct a dataset. Here, we use the results from one data set to illu-
strate. The others are similar and thus omitted. The number of transaction is 97. We
set up MIN values =1. Fig. 4 shows the number of frequent patterns found. We let
σ1 and vary ʱ from 1 to 5. We see from Fig. 4 that the number of frequent
patterns is significantly reduced by the method proposed in [7] when ʱ is not too
large. The number of frequent patterns found by our approach is close to the average
of number of frequent patterns found by the method proposed in [7]. This result indi-
cates that it can prevent to generate meaningless frequent patterns by using our main-
tenance method for automatic min_sup specifying.
Fig. 4. POI dataset
Search WWH ::




Custom Search