Procedure for Creating a Virtual Multibank Agent: The Intelligent Agent
The first problem that we encounter when designing an intelligent agent is that the information that is handled is presented in the same format (standardization of data), so that it is possible to offer the most precise results in the most efficient manner possible when searching for or requesting information, without requiring human supervision [Semet, F.& Taillard, E. ]. It has to do with converting the information into knowledge, referencing data contained on the websites into metadata, with a common layout on a given domain. As for the search system, if, for example, a system presents a significant phrase for each document searched in a list of hits, the essential information on each document can be given to the user. The advantage is that the information about a banking product is usually expressed in a very standardized language. Thus, by mentioning interest rates, the information is usually expressed in APR (annual percentage rate), NR (nominal rate), ER (equivalent monthly rate, semester, etc.); the term of the operation is usually defined as taxed term (in savings operations) or financing term (in loan operations); the capital as taxed or nominal to be financed; the commissions almost always have the same name: maintenance, opening, maintenance fees, administration, commission for early cancelation, commission for partial reimbursement, etc. After making an exhaustive list of the main defining components of a bank product and the usual language in which it is expressed, thus converting it into categories of information, we are then ready to begin the algorithmic design of our intelligent agent, in charge of comparing data from the same category and showing results.
In the design of the intelligent multibank agent proposed in this paper, the main objective that is sought is that it contains an interface that enables questions from the users in natural language about their needs for information about a banking product and that it also responds to these questions in natural language. It also contemplates the possibility that the intelligent agent redirects the user to the address that it deems most probable as a response to the user’s inquiry. It is necessary to direct the user’s inquiry to the use of a few key words related to the content of the banking product about which the inquiry is made. All of the information is grouped into hierarchal levels. The belonging of these key words to each one of these levels is determined by means of a series of numeric coefficients that indicate how significant the word considered is in the level in question. With this objective, weight vectors are assigned to each word, with each vector containing the word’s certainty or probability of belonging to each fuzzy set. The inputs to the system are the coefficients of belonging at each level (Figure 1).
The next step will be to establish a search engine that is in charge of determining the probability that the key words contained in the user’s inquiry belong to a certain fuzzy set at a specific level. For this reason, the search engine takes the weight vectors of the key words as inputs. For the fuzzy logic motor, we initially chose the program Matlab, with its “toolbox” of fuzzy logic, because it is easy to use, it is possible to integrate other “toolboxes” and it is a strong tool. review
Figure-1 Basic design of an intelligent agent