The Study Of Collaboration And Competition Model Among Agents

With the rapid growth of internet information, the search engine was facing tremendous pressure. General search engine returned a large amount of information, but the information quality was low for user's specific query. In order to solve this problem, topic crawlers were born. For traditional topic crawlers, they crawled independently, and without communication. The studies found that it was difficult to improve the efficiency of system by only improving the ability of independent crawling on topic crawlers. Therefore, multi-crawler system was studied. In multi-crawler system, it would cause repeated crawl without immediate communication among them. So it was necessary to study the collaboration and coordination. The main research contents in our paper were as follows:1. The cooperation and competition ability of the crawlers. Considering the cooperation ability of the crawlers from two aspects: firstly, whether the crawler had the ability to continue crawling; secondly, whether the crawler have relatively strong collaboration capabilities. In this paper, we used the remaining time and remaining storage space of the crawlers to measure their crawling ability, and combined the history of crawlers ' collaboration and the operational situation of the cooperation ability to measure their collaboration ability.2. Seduction Factor function and Interesting Function. In order to make the Agent take the initiative participate in the competition, this paper proposed Seduction Factor function and Interesting Function based on the whole system and the idea of a win-win situation in competition Agents. For seduction Factor function, we used the maximum value resources take out by the inviter and the competition ability to measure. For Interesting Function, we used the similarity between resources and theme, as well as the similar degree between resources and tasks performed by collaborators to measure.3. Multi-Agent Competition(MAC) model. MAC model was based on the based on market matching structure in a set of clearance prices, to form perfect match participate between the competition C- Agents and the resources of needed in collaboration.4. More-to-One(MTO) model and Many-to-Many(MTM) model. In the competition and cooperation activities of crawlers, the multiple Invite Agent(Invite Agent, IA) and an Invited Agent(Invited Agent, IDA) formed a many-to-one relationship(More- to- One, MTO) model, and the multiple Invite Agent and multiple Invited Agent(Invited Agent, IDA) formed a many-to-many relationship(Many-to-Many, MTM) model.Finally, to verify the correctness and superiority of the Multi-Agent competition and collaboration, the system achieved the crawlers proposed in this paper based on the JADE platform. We set up comparative experiments by comparing precision rate and recall rate to measure the necessity of the crawlers. Experimental results showed that our research could improve the intelligence of the crawler and the efficiency of the system.