计算机科学
情报检索
语义相似性
语义匹配
排名(信息检索)
概率潜在语义分析
嵌入
图形
语义搜索
领域知识
文字嵌入
匹配(统计)
人工智能
语义网
理论计算机科学
数学
统计
作者
Xuan Wang,Jianchao Lin,Chunhui Ren,Jin‐Ming Chen
标识
DOI:10.1109/iccsnt56096.2022.9972953
摘要
The latent domain knowledge plays an important role in helping to improve efficacy of information search. In order to utilize and mine the latent semantic knowledge behind the query, we propose a knowledge graph-based semantic ranking method for efficient semantic query, where the keywords-based syntactical query is semantically enriched by mining out the latent semantic knowledge of keywords based on knowledge graph. First, a neural network-based knowledge graph embedding model is proposed, where word embedding and entity embedding are learned and optimized using an end-to-end approach, and the semantic knowledge from knowledge graph is fused into the distributed representations of entities. Second, the word-level similarity and entity-level similarity between queries and documents are modeled by interaction matrices, and the strong matching features and soft matching features of the interaction matrices are extracted respectively. Third, we rank documents according to the relevance scores calculated by the above two features. At last, the experiments were made over the related datasets against the traditional keywords-based method, and the results show that the method proposed in this paper can effectively improve the ranking performance.
科研通智能强力驱动
Strongly Powered by AbleSci AI