摘要
目的设计并开发文本配图系统,实现面向文本数据的在线自动配图。方法基于图片和文本之间的描述关系构建“图片-标签”二分网络,然后基于“图片-标签”的二分网络,利用重启随机游走模型进行图片与标签之间的相关性计算。采用TextRank模型提取关键字,并将关键字构成的集合作为查询,将关键字视为标签。基于离线计算结果,在线整合标签与图片之间的相关性,得到文本与图片的相关性。依据相关性由大到小进行排序,并返回前k个最相关的图片。结果实验结果表明,前5个返回结果的MAP值能够达到0.839,能够准确地返回用户期望的图片。结论系统能够依据输入文本进行准确的图片匹配。
The paper aims to design and develop a text matching picture system to realize on-line automatic picture matching facing text data.Based on the description relationship between pictures and text,a“picture-tag”bipartite network was built.And then based on the“picture-tag”bipartite network,the relevance scores between pictures and tags was computed by random walk with restart.The keywords of the input text were extracted by TextRank model and the set composed of key words was queried with the key words as tags.Based on the off-line calculation result,the relevance scores between text and pictures were obtained by integrating the relevance scores between pictures and tags.And the top k most relevant pictures were returned according to the relevance scores.The experimental results show that the MAP score of the top 5 five returned pictures can reach 0.839,which can accurately return the expected pictures.The system can accurately match pictures for input text.
作者
张明西
乐水波
李学民
董一鹏
ZHANG Ming-xi;YUE Shui-bo;LI Xue-min;DONG Yi-peng(University of Shanghai for Science and Technology,Shanghai 200093,China)
出处
《包装工程》
CAS
北大核心
2020年第19期252-258,共7页
Packaging Engineering
基金
国家自然科学基金(62002225)
上海市自然科学基金(16ZR1422800)
上海理工大学国家级项目培育基金(16HJPY-QN04)
国家新闻出版广电总局准重点实验室招标课题(ZBKT201809)。
