Applying Text Mining To Technology Opportunities Analysis In Biomedical Field

S&T innovation is one of the most important driving forces of social development and national competitiveness.The abundant online science and technology intelligence has provided new ideas for supporting S&T innovation on one hand,but on the other hand,it's challenged to extract useful information from such a mount of data by using efficient methods.Biomedical field is a strategic and multidisciplinary sector for accelerating regional economic growth and innovative competitiveness.While it has great potential in market with long-term prospect,its innovation process is characterized as high payoff with high investment,high risk,and time consuming.This dissertation was started from the NSF project “Revealing Innovation Pathway:Hybrid Science Maps for Technology Assessment and Foresight”.Focusing on biomedical field with a gold nanoparticles(GNPs)case study,this research is trying to apply text mining to technology opportunities analysis,so as to provide decision support to researchers and to improve innovation efficiency.1.This dissertation proposed a text feature extraction method based on conditional random fields(CRFs)and natural language processing(NLP).Text content in biomedical literature is different from it in other fields,while terms in biomedical literature usually have specific meanings from some perspective.In this study,we firstly defined four categories of text features: 1)chemicals and compounds;2)organisms and organs;3)genes and proteins;4)experiments related and tools.Then,by generating a keyword list,this feature extraction method combined NLP with CRFs.In GNPs case,feature terms were extracted from titles and abstracts.And they are classified into four categories automatically.Comparing with the traditional NLP result,the distribution of high frequency terms is more concentrated in our new method with less noise.It reduces the cost of data cleaning and can be more effective for forthcoming analysis.2.A model for generating technology translational pathways in biomedical research from multidimentions was described.It's necessary to consider different stages from basic research to clinical study in biomedical field.Biomedical literature should be indexed with different stages or applications before topical and pathway analysis.In this dissertation,this task was completed by using feature selection and text classification.To achieve translational pathways,research topics were generated using principal component analysis and these topics were connected by common terms.In GNPs case,all data was divided into nine subsets from different dimensions.Translational pathways of ten major topics were portrayed on detection,therapy,and imaging.Such translational pathways reveal the difference between different applications and they help researchers to identify technology development trends and research hotspots.3.A technology opportunities identification method based on link prediction was proposed.It is a structural manifestation of research activities to construct cooccurrence network based on feature terms.Research activities contribute to this network by linking different terms.The traditional co-occurrence network usually has this redundancy problem.In this study,it's improved by introducing categories of feature terms.Next,we used link prediction method to capture future links which may represent potential research opportunities.In GNPs case,a technology network on cancer treatment was built using the above method.Similarities between different fearure terms were calculated using five algorithms,and twenty candidate links for furture research potentials were listed.Comping with qualitative analysis,such quantitive method provide new clues for chasing future research opportunities.