|International Journal of Applied Information Systems|
|Foundation of Computer Science (FCS), NY, USA|
|Volume 10 - Number 1|
|Year of Publication: 2015|
|Authors: A.T. Olaniran, I.O. Awoyelu, A.O. Amoo, B.O. Akinyemi, R.O. Abimbola|
A.T. Olaniran, I.O. Awoyelu, A.O. Amoo, B.O. Akinyemi, R.O. Abimbola . An Enhanced Hybrid Item Recommender Model for Nigerian Online Stores. International Journal of Applied Information Systems. 10, 1 ( November 2015), 31-42. DOI=10.5120/ijais2015451459
Item recommendation is the process of recommending goods sold on online stores to visitors and existing customers of the store to aid their shopping transactions processes. Majority of the online stores in Nigeria have their shopping systems implemented similar to foreign online stores’ templates. Adapting these foreign shopping system templates to meeting the needs of Nigerian consumers has been quite challenging. This is due to the unavailability and sparsity of ratings needed by the systems for the generation of these recommendations, thus Nigerian online stores focus on the provision of non-personalized recommendations. The peculiarities of Nigerian consumers call for the provision of personalized item recommendations using alternative information other than ratings information. A hybrid item recommender system that has been demographically enhanced is being proposed in this paper. The model was formulated using the search method, user profiling and association rule mining for the content-based item recommendations. The vector similarity and the adjusted cosine similarity methods were used for formulating the collaborative item recommendations. The demographic item recommendations were then formulated using the clustering and association rule methods. The performance of the system was evaluated using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The results of the performance evaluation carried out on the system showed that the system was able to reduce the Mean Absolute Error of the existing system by 61.24% and the Root Mean Square Error by 37.23% in content-based recommendations. In collaborative recommendations, evaluation results further showed that the new system was able to reduce the Mean Absolute Error of the existing system by 63.16% and the Root Mean Square Error by 39.30%.