Evaluation Of Paper Reading Value Based On Text Mining And Surface Information

The preparation of scientific literatures is a necessary step and an important foundation for scientific research. But the current literature retrieval system only simply sorts the literatures searched in terms of keywords by indexes, such as cited number, published year, and can’t provide valuable information and efficient service to searchers. So, how to evaluate the reading value of literatures and how to provide knowledge service which combine objective value and subjective demand are an urgent subject with important practical significance for saving the researcher’s time. In this dissertation, the literature reading value was comprehensively evaluated from two aspects of subjective and objective, i.e., literature self value and the value relative to searchers, and an evaluation method for literature reading value was proposed based on surface information and text mining.First of all, to evaluate the self value of literature, subject frontier was quantitatively measured based on surface information of literature. In the process of research, keywords were selected as topic words, bursty and attention degree of topic words was chosen as main basis to evaluate the frontier of literature’s subject, and a new frontier metric of keywords was proposed, on which collection of front keywords was constructed, meanwhile, collection of front literature was achieved in terms of the constructed collection of front keywords. Furthermore, a new literature frontier evaluation method based on latent semantic analysis (LSA) was proposed in terms of literature’s surface information such as title, abstract and keywords.Then, in order to quantitatively evaluate subjective value of literature relative to researchers based on surface information of literature, downloading behavior of users that reflects their research background and concerns was analysed by using the text mining methods. In the process of quantitative evaluation, dimension of feature space for clustering was compressed by latent semantic indexing and the researchers’interest was analyzed by hierarchy clustering method, and then, subjective value of new literature relative to researchers was quantitatively evaluated in latent semantic space representing the researchers’interest.Finally, the comprehensive evaluation model of literature reading value was constructed by the transformed E-measure index based on self value as well as subjective value relative to researchers of literature, and the unknown parameters of the model were determined according to the change of kurtosis of reading value for test literatures.The experimental results show that the evaluation method of literature reading value proposed in this dissertation is more reasonable and effective comparing with the traditional evaluation method simply using single factor. The set of front keywords built by the proposed frontier metric method outperforms the academic hot spots achieved by the existing research platform. The positive correlation between our results and the impact factors of journal and published years of literature demonstrate that the proposed method is reasonable for evaluating literature frontier.