The tremendous growth of information on the Internet has been above our ability to process. A recommender system, which filters out useful information and generate recommendations, has been introduced to help users overcome the information overload problem and has been widely applied in an ever-increasing number of e-commercial websites. Collaborative filtering and content-based recommendation methods are two major approaches used in recommender systems. The collaborative filtering predicts items which a particular user prefers by using a database about the past preferences of users with similar interests. The content-based method analyses the content of the objects to generate a representative list of the user’s interests, and then compares the similarity of item descriptions. These two methods have some drawbacks in dealing with situations such as sparse data and cold start problems. Recently, hybrid methods combining collaborative filtering and content-based methods have been proposed to overcome these limitations. However, personalized recommender system attempt to penetrate people’s various demand and generate the tailored recommendations. A highly effective and personalised recommender system may still face new challenges including interestdrifting and multicriteria optimisation. For example, a user’s interest may change over time. They may no longer like a item which was strongly preferred. Another example is that a person’s preference is varying and always has multiple criteria. Classic collaborative filtering uses a single overall rating for prediction. It does not properly reflect the opinion on a item and the reason why people rated this item high or low. Unfortunately, the current recommender systems do not consider these important factors. First, we proposed a novel hybrid recommender system to overcome interest-drifting by embedding the time-sensitive functions into the recommendation process. The experimental results show that the intergraded approach with interest-drifting can constantly perform better and provide users with higher quality recommendations. Meanwhile, the experimental results on different size of training dataset show that our algorithm can boost the prediction accuracy for all configurations. The contributions of this proposed algorithm are in two main aspects. First, using time function to reflect users’ intersts changing in order to achieve higher quality of recommendations. Second, using intergraded methods to solve some problems such as sparsity and cold start. Then we developed a new technique to aggregate the multicriteria ratings for predicting more accurate recommendations. The results show that our algorithms outperforms the traditional collaborative filtering recommender system on both accuracy of predicting ratings and accuracy of recommendations. The one of contributions in this proposed method is that we introduced the multicriteria concept into recommender systems to reflect the users’ opinion more accurate. Another contribution is that we develop a linear method to aggregate multicriteria to single rating for higher quality of recommendations. Our experiments demonstrate that the recommendation achieved better performances when interest-drifting and multicriteria ratings were considered. The significance of our research study is that we consider incorporating interest-drifting, and multicriteria ratings into a recommender system to generate personalised and effective recommendations.
| Identifer | oai:union.ndltd.org:ADTP/252401 |
| Creators | Ma, Shanle |
| Source Sets | Australiasian Digital Theses Program |
| Detected Language | English |
Page generated in 0.0018 seconds