An investigation into recommendation algorithms with application to dynamic environments

Geier, Florian (2010). An investigation into recommendation algorithms with application to dynamic environments. Student dissertation for The Open University module M801 MSc in Software Development Research Dissertation.

Please note that this student dissertation is made available in the format that it was submitted for examination, thus the author has not been able to correct errors and/or departures from academic standards in areas such as referencing.



Today, recommender systems are widely used in various domains. There are a lot of methods to generate recommendations and numerous parameters to adjust these methods. Each one of them has its individual strengths and weaknesses in certain situations. Based on the real-world use case of an existing location-based service application, the research at hand proves that by employing a custom-made hybrid recommender system, it is possible to exploit these strengths and, at the same time, to limit the weaknesses. The project analyses some of the most popular recommendation algorithms with respect to their predictive accuracy on datasets with different characteristics. For this purpose, a suitable evaluation method was designed and implemented in the form of an experimental setup and protocol. Six runtime factors characterizing each dataset are identified and investigated. The results show that these runtime factors have a direct influence on the quality of the recommendations and that they affect different recommendation algorithms in different, sometimes oppositional ways. These results are used to explore whether the predictive accuracy of a superordinate hybrid recommender system in a dynamic environment can be improved, when compared with each single subordinate recommendation algorithm. This is achieved by dynamically selecting or weighting the results of the respective sub-algorithms in consideration of the current situation of use. For this purpose, two different hybrid recommendation algorithms were developed and analysed in direct comparison with the conventional algorithms. It is demonstrated that one of them achieves the most accurate results over the whole range of runtime factor values under investigation, thus effectively eluding the limitations of the specific sub-algorithms it utilizes. The second hybrid algorithm achieves relatively good results, but falls short of the author’s expectations to outperform all other methods including the first hybrid.

Viewing alternatives

Download history


Public Attention

Altmetrics from Altmetric

Number of Citations

Citations from Dimensions

Item Actions