Book Impact Assessment Based On Multi-source Heterogeneous Data

As a hot topic concerned by academia and industry,the impact assessment of publications plays an important role in individuals and institutions.Currently,most impact related researches focus on assessments on journals or papers,few of them pay attention to book impact assessment.Meanwhile,book is an important form of publication,which is indispensable to the inheritance and development of knowledge.Therefore,book impact assessment has theoretical academic significance and practical value.Similar with methods for paper impact assessment,traditional hook impact assessment are based on artificial assessment methods,such as peer review.However,with the surge of books,artificial methods are difficult to evaluate books efficiently.Books' citations and alternative metrics(e.g.library holding,mention times on social media etc.)are used for book impact assessment.However,these frequency based metrics have ignored the content information about books,which is difficult to identify the real citation intentions and purchase motivation.In order to fill the gap,some existing researches used resource like reviews to conduct content mining,so as to assess book impacts.However,most researches are based on a single data source,and without fine-grained mining on relevant content information,which is difficult to obtain a comprehensive result of the book impact assessment.Hence,this paper conducted book impact assessment based on multisource and heterogeneous data,and mined content information via Natural Language Processing and other technologies.Then,we constructed the book impact assessment system based on content information and the existing frequency based metrics(e.g.citation,library holding etc.),so as to evaluate the impact of books comprehensively and efficiently.The main contents of this paper are as follows:(1)Book impact assessment system via multisource and heterogeneous data.This paper integrated multisource and heterogeneous data to construct multi-dimension impact assessment system.Meanwhile,we combined expert empowerment method and analytic hierarchy process to calculate weights of metrics.(2)Book impact assessment based on online reviews.This paper obtained online review based assessment metrics via multi-granularity sentiment analysis.Specifically,we compared four different text representation methods to achieve the optimal performance of document-level of sentiment analysis.Meanwhile,in order to solve the problem of coarsely granularity and non-categorization of existing aspect extraction researches,we combined deep learning and clustering algorithms to extract and category aspects simultaneously.Finally,as poor domain adaptability of the general sentiment lexicons,this paper constructed a domain sentiment lexicon based on the conjunction relationships,and considered sentiment polarities of fuzzy sentiment words(such as large,small,high and low)in different corpus.(3)Book impact assessment based on catalogues and cited literatures.This paper used topic models to conducted depth and breadth analysis,so as to extract topics in books'catalogues and cited literatures,and then calculated corresponding topic distributions respectively.Finally,we got impact assessment metrics based on books' catalogues and cited literatures.(4)Book impact assessment based on citation contexts.This paper identified citation functions via a supervised learning method,and compared four different methods for text representation.Meanwhile,we determined citation strength of each book,and finally we got citation context based impact assessment metrics.(5)Book impact assessment via integration of multisource and heterogeneous data.This paper integrated the content based assessment metrics and the frequency based assessment metrics to obtain the comprehensive results of book impact assessment,and we verified our results by conducting correlation analysis with experts' ratings on books.Finally,on the basis of summarizing the existing results,we summarized this paper and pointed out directions for further improvement.