THESIS
2015
xii, 55 pages : illustrations ; 30 cm
Abstract
In the era of information overload, it is hard for information consumers to find what they
are interested in among the massive information while it is also hard for information producers to distinguish their information to draw attention from users. In the environment of e-commerce, this contradiction is more intense actually. Recommendation system is such an important tool to address the above-mentioned problem, connecting users with the information they need. Collaborative Filtering and Supervised Learning are two kinds of frequently used recommendation algorithms in these systems. However, to the best of our knowledge, there is no published comparative research about the prediction performance between the above-mentioned recommendation algorithms on a large real-world e-commerce dat...[
Read more ]
In the era of information overload, it is hard for information consumers to find what they
are interested in among the massive information while it is also hard for information producers to distinguish their information to draw attention from users. In the environment of e-commerce, this contradiction is more intense actually. Recommendation system is such an important tool to address the above-mentioned problem, connecting users with the information they need. Collaborative Filtering and Supervised Learning are two kinds of frequently used recommendation algorithms in these systems. However, to the best of our knowledge, there is no published comparative research about the prediction performance between the above-mentioned recommendation algorithms on a large real-world e-commerce dataset.
Therefore, We implement two mainstream Collaborative Filtering algorithms and four
frequently-used Supervised Learning algorithms on a large real-world e-commerce customer behavior dataset and first compare their performance from three different angles experimentally. Our experimental results demonstrate that Supervised Learning algorithms outperform Collaborative Filtering algorithms on prediction power. Particularly, Random Forest was the best recommendation algorithm overall. We also provide some insights into conducting experiments on large-scale real-world dataset and some experience of solving related engineering problems, which can be learnt by other researchers or
anyone who attempts to build an e-commerce recommendation system.
Post a Comment