Web information integration system (WIIS) provides a novel mehtod for management of web data. It provides a unified, transparent interface for users who want to acess multiple sources on the web. The primary aim of WIIS is to help users effectively and efficiently querying heterogeneous and distributed data sources. Currently, WIIS has been an important research direction. This dissertation mainly focuses on some key problems in WIIS: common data model, query optimization, cooperation and learning to enhance performance of WIIS. The main contributions of this dissertation are as follows:(1) Firstly, the current approaches of information integration are summarized and classified. Two integration paradigms (materialized and virtual) are compared. Then we make a overview on the development of information integration systems, e.g mult-database systems, mediator/wrapper integration systems, federated database systems and information agent systems. Finally, a new query optimizer is designed based on the analysis the query processes and optimization in the WIIS, this optimizer can support dynamic query optimization and enhance the performation of systems effieciently.(2) A new data model called XCDM is presented based on two semi-structured data model OEM and OIM. In web information integration system, it's a general method to provide an efficient common data model (CDM) in order to integrate multi-data source with different data model. XCDM combinates the characteristics of XML with directed connected graphic structure of OIM, it can express XML documents flexibly and can provide view in different level for users and applications. Also two other object algebra operation is complemented besides OIM object algebra. XCDM is accorded with the rules of CDM.(3) An imporved method to solve multi-join query optimization in WIIS based on GA is proposed this dissertation. According to multi-join query expressions, encode method, crossover operator and mutation operator are introduced. In WIIS, many data sources have limited query capacity which can be expressed by bind pattern. So our method is partitioned into two phases, in the first phase the search space of GA can be reduced by the bind pattern of data sources and in the second phase, G A exploits the result of phase I as heuristics and seeks the optimal query execution plan. This method can be used not only for search space of left-deep but also for hybrid search spaces. The experiment results show that our algorithm is more efficiently while there are many dataAbstractsources integrated in WIIS.(4) A new cooperation model called MACM is presentd and based on this model, an improved distributed reinforcement learning algorithm is also proposed. Now many efforts have been given to developing WIIS into intelligent cooperative information systems in order to fulfill complex task requesteb by users. Hence, much research is focused on efficient cooperation among the components (Agents) of WIIS. According to theory anaylsis, each agent can harmonize its own actions with others through learning when learning mechanism is introduced into intelligent cooperative information systems. Multi-agent cooperation based on reinforcement learning is studied in this dissertation. MACM provides a flexible mechanism to support learning in multi-agent systems and our new algorithm not only can reduce the store space of Q-table but also can coverge to optimal equilibrium rapidly.(5) We combine the distributed artificial intelligence with distributed database system and present an agent-based query optimization system model. In this model, we adopt deliberative agents based on BDI. Each source agent possesses the capacity of learning and reasoning based on its belief and goal, and multi agents interact with each other to support the dynamic query optimization and plan execution. Thus, the performance of the whole system can be improved efficiently. Also, we study the process of multi-agent reinforcement learning in selecting sources with lighter load to answ