Chapter 9 - Conclusions

In this thesis, we have described the architecture and implementation of a movie recommender system, called the MOVSOM. The system incorporates the SOM, developed by Teuvo Kohonen [kohonen01som] as well as known recommendation techniques. Early versions of MOVSOM was developed in two projects, in connection with the courses "artificial neural networks" and "information retrieval on the internet" taught at the university of Uppsala during the year 2004. This thesis has since the fall of 2004 been an ongoing project that finally resulted in this master thesis. In this thesis we have shown that MOVSOM is comparable to traditional techniques used by recommender systems when it comes to accuracy of the given recommendations. MOVSOM offers a highly interactive user interface, and implements transparency that gains trust among its users, as defined by recent studies. A new way of presenting recommendations is utilized by MOVSOM. We claim that this visualization of recommendation is more intuitive than the traditional way of presenting recommendations as one-dimensional lists, in the sense of visualized similarity among the items. Our study of different prediction algorithms has convinced us that the accuracy of present algorithms are as good as they can be, in the sense that a few decimals in either direction of the mean absolute error doesn't have any impact on the overall acceptance by the user of the recommendation. The use of the SOM algorithm as a part of a recommender system is promising, its topologically preserving capabilities has been shown to be a very suitable feature to use in a recommender system.

Copyright © 2005-2024 MOVSOM™ Five By Five