With the explosive growth of geological data, considerable researches are focused on accurately retrieving specific information from massive data and fully exploiting the potential knowledge and information in unstructured data. Currently, the researches on unstructured content retrieval mostly ignore the association of semantics and knowledge or only consider the association of a singular granularity, which leads to the lost of concepts with the same semantics but expressed in different forms during the retrieval process. To address these problems, this paper has made some enhancements, and the main contributions include: (1) Define a decision rule, split unstructured geological survey data into content fragments, and construct a more fine-grained geologic textual semantic description model. (2) Present a multi-constraint fusion feature weighted model to extract the thematic feature items from the content fragments. (3) From the three granularity of document, content-item, feature-item, the associations of the same-granularity and cross-granularity are merged to construct a multi-granularity geological text hypernetwork model. (4) The experiments verify that the proposed approaches can improve the precision and recall rate of unstructured content retrieval.

随着地质数据的爆炸性增长，大量研究集中在准确地从海量数据中检索特定信息以及充分利用非结构化数据中的潜在知识和信息。当前，关于非结构化内容检索的研究大多忽略了语义和知识的关联，或者只考虑了奇异粒度的关联，这导致了在检索过程中具有相同语义但以不同形式表示的概念的丢失。为了解决这些问题，本文进行了一些改进，主要贡献包括：（1）定义决策规则，将非结构化地质调查数据拆分为内容片段，并构建更细粒度的地质文本语义描述模型。（2）提出了一种多约束融合特征加权模型，从内容片段中提取主题特征项。（3）从文档，内容项，特征项这三个粒度上，将同粒度和跨粒度的关联进行合并，构造出多粒度地质文本超网络模型。（4）实验证明，该方法可以提高非结构化内容检索的准确性和查全率。