Abstract:
This research proposes an approach of the Dynamic Parallel Web Crawler
which aims at improving efficiency of Thai Web Search Engines in both the quality of
the result set of web pages retrieved and the storage optimization. The research
framework consists of three main components. First, the concept of parallel crawling
was applied to facilitate the crawling task to be performed in parallel and dynamic.
Second, an Ant Colony Algorithm was adopted to increase the speed of the crawler
and to gain more crawling search area. Since the algorithm was designed to support an
optimized search conducted in the distributed environment, it is thus very applicable to
the Dynamic Web Crawler solution. Finally, an agent based concept was used by
introducing an Ant Crawler Agent to support several crawling tasks to be responsible
for each crawler agent. This enables the management, and representation of the
crawling result is be more flexible and effective.
To verify the research idea, a prototype system of the proposed model of the
Dynamic Parallel Ant Crawler (DPAC) was designed and developed using JSP and
Servlet. In the experiment, the distributed system environment consisting of three
servers was established to implement the core function of the parallel agent crawler.
We evaluated the system in three parts: Retrieval Performance Evaluation, Resultbased
Evaluation and User Acceptance Test (UAT). The Retrieval Performance
Evaluation is conducted to compare the throughputs between a single Ant Crawler
System (ACS) and a parallel Ant Crawler System (ACS) using three servers. We use
100 keywords to test with the DPAC run on a single ACS and parallel ACS. The result
shows that the parallel Ant Crawler yields a greater number of URLs relevant to the
user’s requirements compared with the single Ant Crawler. In addition, with the same
number of throughputs, the parallel ACS reduced the crawling time significantly. For
the Result-based Evaluation, the accuracy rate of the set of URLs generated by the
DPAC system was focused on and the results were compared with recognized
commercial search engines. The results revealed that the DPAC generates the set of
URLs covering the given keywords and the ranking order of those URLs in an
accurate and consistent way as the commercial tools do. Using 1,000 keywords, the
results showed that DPAC outperformed the other two search engines. Finally, the
prototype was also measured by means of the UAT test in order to assess the user
satisfaction in using our search engine. With 800 keywords of the test among 40 users,
the user acceptance rate was 97.25%. As a consequence, that this approach offers
another sound and proven solution for the research area of the Thai Search Engine