Several keyword(s) were submitted to test the i-agent. The keywords covered a broad range of topics. Keywords submitted were single keywords, such as "Conference" and "Tennis", and multiple keywords, such as "Conference Australia", "Intelligent Agents " and "Information Retrieval". Volunteers tested the results of searching, first, by using the keyword(s) and their chosen search engine (AltaVista, Excite, Google, HotBot, Infoseek, Northernlight, etc.). The volunteers then used the i-agent program for search and retrieval from the WWW by giving the same keyword(s). The results from the two searches were then compared. For example, in one experiment, a volunteer used the keywords "Conference Australia" and performed a search using the AltaVista, Excite, Lycos and Yahoo search engines. These search engines, shown below, returned a large number of results.
It is very unlikely that a user will search the 26,194,461 results shown in the AltaVista query in Table 1. This could be due to users' past experience in not finding what they want, or it could be due to the time constraint that users have when looking for information. A business would consider the cost of obtaining the information and just what value exists after the first 600 pages found. It is very unlikely that a user will search more than 600 pages in a single query, and most users likely will not search more than the first 50.
Search query: Conference Australia | |
---|---|
Search Engines | Number of pages returned |
AltaVista | Conference: 26,194,461 and Australia: 34,334,654 |
Excite | 2,811,220 |
Lycos | 673,912 |
Yahoo | 40 categories and 257 sites |
In a more recent experiment, a volunteer used the search query of "Conference Australia". This volunteer extended the search to include several other search engines that are considered more popular in their use. The results illustrate that search engines and the results are changing dynamically. However, it is still very unlikely that a user will, for example, search the 1,300,000 results as shown in Google or the 3,688,456 shown in Lycos. The following table illustrates their results.
The dynamics of web data and information means that the simulation could be done any day and different results will be obtained. The essence of this research takes that into account by providing a way to continuously provide the user with their desired information while taking into account the dynamics of information on the Web. To evaluate the performance of the information filtering and retrieval of the above system, several simulations were performed. Simulation results for some of the keywords that are used to check the performance of the systems were: Conference, Conference Australia, Intelligent Agents, Information Retrieval, and Tennis. The results obtained were then passed to the evolutionary computing system, and final results were passed to volunteers for testing. The evolutionary computing, as described above, was then used to find the most relevant pages to this search query. The evolutionary computing was run for 200 generations. The top 200 URLs returned from the evolutionary computing were then presented to the volunteers to assess the results. The volunteers then compared the results obtained from i-agents to their own opinion from the web sites they visited and evaluated. It is interesting to see that the results obtained by combined i-agent and evolutionary computing are very good.
Search Engines | Number of pages returned |
---|---|
AltaVista | 766,674 |
| 1,300,000 |
HotBot | 807,500 |
Lycos | 3,688,456 |
Northernlight | 103,748 |
Yahoo | 251 pages with 20 hits per page |
Search query: Conference Australia | ||||
---|---|---|---|---|
Search Engines | Number of pages returned | Number of relevant pages | I-agent relevant pages | Number of relevant pages from i-agent and evolutionary algorithms |
AltaVista | Conference: 26,194,461 and Australia: 34,334,654 | 2,050 | 180 | 179 from 200 pages returned |
Excite | 2,811,220 | 1,889 | ||
Lycos | 673,912 | 2,210 | ||
Yahoo | 40 categories and 257 sites | 84 sites were relevant |
The number of URLs returned is feasible , and the users are provided with relevant web pages. Out of another 62 experiments performed, the combined i-agent and evolutionary computing system performance was good. The volunteers have reported that in more than 70 percent of the experiments (out of 62 experiments), the results of the combined i-agent and evolutionary computing system were very satisfactory. Table 4 shows that relevancy of the URLs, based on some of the queries given to i-agent and evolutionary computing.
Search query | Number of pages returned i-agent | Number of relevant pages returned by i-agent and evolutionary computing and fuzzy logic |
---|---|---|
Conference | 180 from 215 pages returned | 169 from 200 pages returned |
Conference and Australia | 127 from 223 pages returned | 127 from 200 pages returned |
Intelligent Agents, and Tennis. | 189 from 348 pages returned | 132 from 200 pages returned |
Information Retrieval | 117 from 216 pages returned | 111 from 200 pages returned |
Information Retrieval | 127 from 223 pages returned | 108 from 200 pages returned |