The Design And Implementation Of Scientific Literature Recommendation Subsystem In PKUSpace

As the rapid development of information technology and the explosive increase of digital resources, the traditional information retrieval technology that bases on keyword search can't meet people's requirements increasingly. Personalized recommendation systems have appeared in this case. As an important part of personalized services, the recommendation system has great advantages: it automatically collects users'characteristic information and makes recommendations actively for users according to the user profile, and it also tracks the change of users'interest, and adjusts recommendations in real time accordingly.This paper designs and implements a scientific literature recommendation subsystem in PKUSpace, which is supported by the NSF project. PKUSpace is platform built for research and communication, which provides the storage, sharing, searching and navigation of scientific literature as the basic functions. By applying web2.0 technologies, the platform also provides functions including collaborative tagging, sharing reading notes and view for personal literatures. Based on functions above, the tag mining and scientific literature recommendation services are also available.The most popular technologies for recommendation are content-based recommendation and collaborative recommendation, both of which have their advantages and shortcomings. Many systems try to combine both of these two technologies for better results.This paper adopts a recommendation framework which combines the content-based recommendation technology and collaborative recommendation technology. This framework is based on collaborative tags, and also utilizes the content information and citation information of the literature for the construction of user profile and literature model. The novelty of this paper is as follows: Using collaborative tags for recommendations in this system. Collaborative tagging is a popular way for organizing resources in web2.0 system, so the method adopted by this paper can apply to all systems which provides the function of collaborative tagging.Representing the user's profile based on tag text. Most of recommendation systems which utilize tags just simply use tags as the judgment for whether a user is interested to a certain resource. However, the tag texts are not utilized.Extending user's profile by literatures'content and citation information.Representing the matching between user model and resource model by the degree that a user is interested to a certain resource. When computing this degree, this paper uses the vector dot product instead of the popular cosine similarity.Combining the collaborative recommendation technology and content-based technology by multiple ways, and the weight of these two technologies can be adjusted to adapt to the increase of the system scale.