As a kind of normative and structured knowledge form,thesauri plays a significant role in the fields of information retrieval,enterprise knowledge management and service,as well as automated intelligent decision-making. Traditional closed-world thesauri have more and more severe drawbacks,such as high cost of expert-knowledge construction,low accuracy and narrow coverage of domain knowledge,obsolete knowledge-updating mechanism and weak capacity of knowledge expansion. It has become an important research domain that how to construct a proper,holistic and scientific system based on thesauri,as a powerful aid towards various systems and applications of intelligent information and knowledge management,which aims at knowledge ordering and sharing under the condition of the open interconnection. Related research has discussed in depth the structure,data sources,methods and language of thesauri integration,but lacks comprehensive induction and combing of multi-source fusion technology,theory and methods on thesauri,especially about how to embark on knowledge fusion in the open and interconnected environment.

    Introducing knowledge fusion,this paper,first of all,defines the conception and connotation of the paradigm of “knowledge fusion” in the light of Popper's Three World Theory,and discerns its difference from “data integration” and “information integration”,as to clarify the category and scope of “knowledge fusion” fundamentally; Secondly,from the perspectives of OWA(Open World Assumption),this paper presents a more comprehensive and systematic theoretical framework of knowledge fusion called “MtFFO” in combination with KRAFT framework and other related existing models. The framework mainly contains the external environment,internal core,input unit,key factors and methods. Through feature comparison of different data source (database,metadata,and the table),external Environment Sub-frame of Information Input(EESII)is expounded as a multi-level architecture of knowledge adjustment subsystem. Next the knowledge flow dealt with EESII will receive the pattern matching and semantic recognition as the control and guidance of fusion strategies and quality to achieve optimization of knowledge transformation and composition; the respectively corresponding actual operation process is knowledge mapping and merging,and the former can be divided into two methods,knowledge aggregation based on homogeneity and logical conversion system on the meta-thesaurus,and the latter consists of simply merging,reductively merging,fully merging,or synonymy & hyponymy merging. The best way we recommend is taking target requirement,detailed context,cost and quality of knowledge unit and schema in thesauri as well as feasibility into consideration,combined with the different technology of matching,mapping and merging,and importing expert knowledge and reasonable control,so as to improve the efficiency,accuracy and robustness of the fusion system. Another innovation of the “MtFFO” framework is,to deal with the difficulty of thesauri expansion and enrichment,its introduction of metadata,general knowledge base on the basis of graph model and machine learning algorithm to mix all kinds of domain taxonomy,vocabulary and database together for data mining and knowledge discovery of high quality knowledge unit.

    Although “MtFFO” is merely a theoretical framework without any validation by systems in practice,it can not only improve and develop the current theoretical system of knowledge fusion to a certain extent but also scientifically sort out and integrate a variety of relevant methods and crucial technology,providing thus theoretical basis and technical reference for solving the problems of thesauri interoperability,semantic understanding and self-enriching mechanism. 5 figs. 1 tab. 76 refs.