Personalized scientific literature recommendation based on user's research interest

摘要

As the rapid growth of digital scientific and technical literatures, the scientific researchers need personalized retrieval to satisfy their research interests urgently. Unlike previous methods that just use simple keyword matching, we strengthen the semantic information of scientific literature by merging metadata such as title, keywords, abstract and citation in this dissertation. Then we use vector space model with tf-idf value of each term to model each scientific literature. We structure user's interest model by subject term vector with different weight. Different weight means different concern degree to each subject term by user. To emphasize the quality of recommended literature, we not only calculate the similarity between user's research interest and literatures, but also consider the total times cited of recommended literatures. As experimental dataset, we collect literatures from Web of Science under the topic of “pressure sensor”. The experimental results show that this recommendation method could well indentify user's research interest. So, it can improve efficiency of user literature retrieval and increase the accuracy of the literature recommendation.