Page 191 - Proceedings of the 2017 ITU Kaleidoscope

Challenges for a data-driven society

Fig. 5. Trend Analysis and Document Find Function

The trend analysis function is implemented to display the 5.2. Results
graph by summing the occurrence rate of each document for We implemented three different experiments to show the
each year. This function is implemented using the effects of the number of topics and iterations in Table 4. For
FusionCharts API [20]. The occurrence rate is the the topic modeling experiments, we set the elements of
percentage of a representative keyword in the topic that TASIS to Y series and the year range from 1998 to 2016. In
represents each document, an important criterion for trend the experiment, each cluster is a topic. At first, after the
analysis. TASIS implemented 1 topic and 1000 iterations, we can see
that service, network, and function are keywords with high
5. EXPERIMENTS AND RESULTS weights in Y series documents (a). It is only shown to word
count. Second, the TASIS performed 10 topics and 1
5.1. Experimental data iteration and then sorted in descending order the topics by
We collected 252 documents of Y series (Global the dirichlet parameter (b). The result are based on the word
information infrastructure, internet protocol aspects, and count because of the one iteration. As the LDA algorithm
next-generation networks) from the ITU-T for topic modeling of the TASIS is based on the expectation-
Recommendations in Table 3. To experiment with the Topic maximization (EM) algorithm [16], if the number of
Modeling Function and the Trend Analysis and Document iterations is set to 1, the optimized result cannot be extracted.
Find Function, we used all Y series documents. Finally, the TASIS implemented 10 topics and 1000
Iterations (c). As per (c), we have also verified topic 0,
Table 3. Documents for Experiments which includes service, ngn, and network, and extracted it as
high proportional topic, and topic 1, which includes service,
Documents Title resource, and cloud, being confirmed the second highest
proportional topic. We can thus understand various topics
Y.100 - Global information infrastructure representing Y series documents. As a result, the topic
Y.999 modeling using LDA to extract qualified topics is necessary
Y.1000 - to specify appropriate the topic number and iterations.
Y.1999 Internet protocol aspects
Y.2000 - Next Generation Networks (NGN) As per Figure 5, TASIS extracted the documents related to
Y.2999
Y.3000 - Future networks the IoT as the result of the experiment for the Document
Y.3499 Find Function. Among the 252 documents, the most
Y.3500 - relevant international standard documents are
Y.3999 Cloud Computing Recommendation ITU-T Y.2074: The Requirements for
Y.4000 - Internet of things (IoT) and smart cities and Internet of things devices and operation of Internet of things
Y.4999 communities applications during disasters and Y.2066: Common
requirements of the Internet of things. On the other hand, the

– 175 –