Query Of Cross-Media Scientific Research Achievements Based On Learing To Rank

With the development of the Internet and big data,the data scale on the Internet is becoming larger and larger.These data contain text,images,videos and other information.Unlike social media data and news data,scientific research achievements have the characteristics of many proper nouns and strong polysemy.The traditional single-modal query method containing only keywords can not meet the needs of scientific researchers and managers of the Ministry of Science and Technology.Scientific research project information and scientific research scholar information contain a large amount of scientific research achievement information.Evaluating the output ability of scientific research achievements from scientific research projects and scientific research teams can effectively assist managers in making decisions.In view of the above background,study the data collection and feature extraction methods of cross-media scientific research achievements,the query and sorting methods of cross-media scientific research achievements,and the research and development of cross-media scientific research achievements query system has profound significance.The main work of this thesis is as follows:(1)Study the collection,processing and storage methods of cross-media scientific research achievements and the learning methods of cross-media semantic features.Design a distributed module for data collection,processing and storage of cross-media scientific research achievements,which effectively solves the problem of how to collect multi-source heterogeneous scientific research achievements efficiently.Propose a text and image feature extraction method which suitable for cross media scientific research achievements query.The experimental results show that the text feature and image feature extraction methods have good semantic expression ability,and the extraction method of scientific and technological named entity can effectively extract entities in the text information of scientific research achievements.(2)Study the query method of cross-media scientific research achievements.Propose a cross media scientific research achievement retrieval method based on deep language model.In this method,the image and text feature vectors of cross-media scientific research achievements are generated into a unified semantic vector through cross-media semantic learning,and the search of cross-media scientific research achievements is realized in the unified semantic space.The experimental results show that the query effect of the method is better than the comparison method in all kinds of the same mode and different modes.(3)Study the query method of scientific research achievements based on learning to rank.Propose a pointwise method based on learning to rank and integrating the side information of scientific research achievements.This method makes use of the strong correlation between scientific research achievements and scientific research projects,takes scientific research achievements as the sorting basis of scientific research projects,carries out single document sorting learning and integrates it into query,which is very convenient for supervisors and researchers to evaluate the overall achievement output ability of scientific research projects.Experiments show the effectiveness of the method.(4)Design and implement cross-media scientific research achievement query system based on learning to rank.This system contains the query of scientific research achievements,scientific research teams and scientific research projects by using cross-media scientific research achievements data collection,processing,storage and feature extraction methods,cross-media scientific research achievements query methods and scientific research achievements sorting method based on learning to rank.Through three query methods,the scientific research achievements are queried from various aspects to show the output ability of scientific researchers and scientific research teams,and the system verifies the effectiveness and feasibility of these methods.