Study On Online Keyword Auction Agent Bidding Strategies

Internet, especially search engine, has become an important tool for people to searchinformation. As one of the important economic basis to the search engine,the keywordadvertising effectively satisfies the marketing needs of advertisers, and brings huge profits tothe search engine providers. The keyword auction develops from search engine. It becomesthe most successful micro-economic theory applications in the Internet,and promotes theresearch of the mechanisms behind the keywords auction by many scholars. Recently, thekeyword auction has becoming toa hottest research field of Electronic commerce. Tradingagent competition, referred as TAC, is co-sponsored by Carnegie Mellon University and otherschools. TAC is designed to simulate the real market behavior and applied the currenttheoretical research in artificial intelligence to implementationthe transactionprocess. TAC/AA is a virtual keyword advertising auction platform. Researchers can applythe studies to this platform and verify the effectiveness of the algorithm. This paper focuseson the TAC/AA platform, designs an Agent model to export a continuous optimal biddingstrategy. It proposes a dynamic and diverse elite PSO algorithm. This algorithm effectivesolves the local convergence problem in TAC optimizer.At first, this paper analyzes international auction keyword researches in recent years anddescribes the TAC platform design idea and competition rules, and then introduces thekeyword auction theory, Agent theory and the theory of evolutionary algorithms. Wedesigned TAC-HEU-AA (THA) Agent model for the TAC competition keyword auctionplatform. Following by the model, we analyze the efficiency of the predictor algorithm andverify the necessity of each model by experiments. The THA model is compared with Agentswhich participate in the TAC/AA final round. By comparing the benefits of Agents, THAAgent model is verifiedon the validity. Then we analyze the advantages and disadvantages ofTHA model. We present a novel elite selection strategy based on a variety of particle swarmoptimization (DME-PSO) for the THA optimizer. We define three trends in the movement ofparticles and four particles in movement behavior. The particle is selected by the motionbehavior of particles in order to maintain population diversity. In the mutation process, thispaper defines an elite population saturation phenomenon. According to this phenomenon, the population selects variant behavior.We compare the DME-PSO algorithm with greedyalgorithm, multi-population genetic algorithm and add perturbation particle swarm algorithmin the knapsack problem. We found that DME-PSOalgorithm perform better than currentalgorithm with the increment of items.