Integration of structured and unstructured data goes much deeper than supporting large objects in a database. Through an architecture overview of the IBM Content Manager, this paper examines some of the requirements, challenges, and solutions in managing a large volume of content and in support of a wide range of content applications. The discussion touches upon system architecture, data model, and access control.

We describe a data model for structured office information objects, which we generically call “documents,” and a practically useful algebraic language for the retrieval and manipulation of such objects. Documents are viewed as hierarchical structures; their layout (presentation) aspect is to be treated separately. The syntax and semantics of the language are defined precisely in terms of the formal model, an extended relational algebra. The proposed approach has several new features, some of which are particularly useful for the management of office information. The data model is based on nested sequences of tuples rather than nested relations. Therefore, sorting and sequence operations and the explicit handling of duplicates can be described by the model. Furthermore, this is the first model based on a many-sorted instead of a one-sorted algebra, which means that atomic data values as well as nested structures are objects of the algebra. As a consequence, arithmetic operations, aggregate functions, and so forth can be treated inside the model and need not be introduced as query language extensions to the model. Many-sorted algebra also allows arbitrary algebra expressions (with Boolean result) to be admitted as selection or join conditions and the results of arbitrary expressions to be embedded into tuples. In contrast to other formal models, this algebra can be used directly as a rich query language for office documents with precisely defined semantics.

In this paper, we study a special class of trees called $\alpha $-$\beta $ leaf trees with degree r. These trees arise in information retrieval problems as well as prefix coding problems. An efficient method for constructing optimal trees is presented. The method is combinatorial in nature instead of the integer programming approach followed by other authors to study similar problems. The construction time is linear to the number of leaves of the tree.

Based on the Mattsom-Solomon polynomial, a class of algebraic linear error-correcting codes is proposed, which includes the Bose-Chaudhuri-Hocquenghen (BCH) codes, Goppa codes, and Srivastava codes as subclasses. Several constructive bounds on the minimum distance of these codes are derived and are shown to be achievable using either Berlekamp's iterative decoding algorithm or Goppa's method based on divided difference. Moreover, it is shown that this class of codes asymptotically approaches the Varshamov-Gilbert bound asn \rightarrow \infty. Although some binary Goppa codes were previously known to haven \leq 2^m, r \leq m \cdot t, andd \geq 2t+ 1, it is shown that a much larger class of codes also possesses such parameters. Finally, shortened codes are considered. With a limited computer search, a number of good codes were found. It is also observed that the proposed codes have no fundamental difference from those recently given by Helgert.