Application Of Case Based Recommender System To Advice New Importers And Exporters In Product Selection

Abstract

There are various products which are imported from different countries to Ethiopia and that are exported from Ethiopia to different countries. But to select products to import or export to new importers and exporters based on their interest is much difficult. There is no automated system that can assist new importers and exporters in product selection process. Therefore the main objective of this study is to design a prototype case based recommender system that can support new importers and exporters in product selection and to enable them to make timely decision. The knowledge is acquired using semi-structured interview, document analysis and observation to develop the prototype case based recommender system of import and export product selection. The acquired knowledge is modeled using hierarchical tree structure model and represented using feature value case representation. The prototype of the case based recommender system is implemented using JCOLIBRI2.0 tool. For this study, 1561 case were collected from ministry of trade and 7 attribute were selected as a description and solution attribute to develop the prototype. K-nearest neighbor algorithm was applied to compute the similarity between new case and cases retained in the case base. Then the system retrieves top 10 cases and adapts the solution through direct copy method. The solution is revised by domain experts and the final solution is retained for future use. The system recommends the amount of capital to import or export products based on the input or query. The recommendation system of fitness of the final solution to new problem is evaluated by domain experts and other users of the system. The system is evaluated in terms of its performance and user acceptance testing by different domain experts and importers and exporters. According to the evaluation of user acceptance testing, the average performance of the system is 87% and 80% of domain experts and importers and exporters respectively. The system also was evaluated similarity retrieval and relevant cases from retrieved cases performance using recall and precision. The average result of recall and precision is 85% and 59% respectively