The general agents used by BAsCET are already enumerated in the Building the logical part of a Concept Network representing bibliographic references node. They will now be described more precisely. Let's remember that these agents can work with any type of reference, and even any type of structure, would it be macro- or micro-. This algorithm of search for similar sub-strings is employed by separator seeker and by instance seeker. Next, you can read about the mechanism for building hierarchical structures in the Blackboard, that takes into account the solution's hierarchical structure, and manages conflicts between seeker agents (that's what is presented in later paragraphs). This mechanism is used by all the agents that may build an object in the Blackboard. For the agents' descriptions themselves, see: separator seeker, instance seeker, field seeker, zone seeker, and at last, stop agent.

In the blackboard, and in the frame of BAsCET's application on bibliographic references recognition, several types of objects may be built in the Blackboard:

instances of a generic node in the Concept Network
instances of a separator node, they contain textual and typographic information, in XML.
instances of a specific node of the Concept Network, and contain only textual information.

All these objects are linked inside a hierarchical structure, mostly represented by a link towards the containing field, and their location in this field, as shown by Figure 1. This location is represented by the location of the object's first character inside the containing field. An object is called a description of the containing field. Every field may have several (complementary) descriptions.

Figure 1: Hierarchical structure of the Blackboard

                        *+---------------+-----------------------+     * field:doc *
                        *| <Times-Roman> | C. B. Bose and S. Kuo | ... *************
                        *+-------+-------+---------------+-------+     *
                                 |                       |
                                 v                       v
            *******************************         ****************************************
            *               * sep:-author *         *+-----------------------+*field:author*
            * <Times-Roman> ***************         *|C. B. Bose| and |S. Kuo|**************
            *               *    0-12     *         *+--+-------++----+----+-+*   13-33    *
            *******************************         ***/*********|*********|****************
                                                      /          |         |
 *************************                          L            v         v
 *         * father node *      ********************** *************** ******************
 * content ***************      *            *field:a* *     *sep:a-a* *        *field:a*
 *         * location    *      * C. B. Bose ********* * and ********* * S. Kuo *********
 *************************      *            * 13-22 * *     * 23-27 * *        * 28-33 *
                                ********************** *************** ******************

But the right structures are rarely built at the first stage. That's the reason for the existence of a mechanism allowing to keep only one description at one location inside a field, the best for the agents that built it (or that want to build one).

There is a conflict between two description whenever one contains one another, or whenever they have a common part. When there is a conflict, one has to decide what description to keep alive. This decision is based on a score computed for each of them. This score depends on the object's length (i.e. the length of its content), and on its happiness. The happiness is more taken into account than the length.

Log in or registerto write something here or to contact authors.