Encoding Semantic Knowledge in Vector Space for Biomedical Information

描述(由申请人提供)：从生物医学文献中有效地检索信息对于支持循证临床实践至关重要。然而，研究表明，临床医生很难使用现有的电子信息资源接口生成足够具体的查询。由于生物医学研究文献的迅速扩散，需要开发工具来使临床医生和研究人员能够找到和检索感兴趣的文献。将文档表示为高维空间中的向量的向量空间模型在信息检索中得到了很好的应用。然而，由于该模型基于术语(或模型变体中的概念)对文档进行索引，这限制了可以查询这些文档的特殊性。在我们最近的研究中，我们开发了基于谓词的语义索引(PSI)，这是一种基于向量的模型，它将SemRep系统从MEDLINE中提取的对象-关系-对象三元组(或谓词)形式的知识编码到向量空间。在拟议的研究中，我们将开发和评估一个新的基于PSI的信息检索模型。这一模式将使人们能够使用概念和关系搜索文献，以回答诸如“什么是用来治疗结核病”之类的具体问题。该模型代表了信息检索研究的一个新方向，我们的假设是，基于预测的文档表示将能够指定比现有模型更精确的查询。为了验证这一假设，该模型将使用OHSUMED测试集进行评估，并使用标准性能度量与传统的向量空间模型进行比较。

DESCRIPTION (provided by applicant): Effective information retrieval from the biomedical literature is essential to supporting evidence-based clinical practice. However, research suggests that clinicians have difficulty generating sufficiently specific queries using existing interfaces to electronic information resources. On account of the rapid proliferation of the biomedical research literature, there is a need for the development of tools to enable clinicians and researchers to find and retrieve documents of interest. The vector space model, in which documents are represented as vectors in a high-dimensional space, is well established in information retrieval. However, as this model indexes documents on the basis of terms (or concepts in variants of the model), this limits the specificity with which these documents can be queried. In our recent research, we have developed Predication-based Semantic Indexing (PSI), a vector-based model which encodes knowledge in the form of object-relation-object triplets (or predications) extracted from MEDLINE by the SemRep system, into vector space. In the proposed research we will develop and evaluate a new model of information retrieval based on PSI. This model will enable searching for documents using concepts and relations, in order to answer specific questions such as "what is used to treat Tuberculosis". This model represents a new direction in information retrieval research, and our hypothesis is that document representations based on predications will enable the specification of queries that are more precise than are possible with existing models. To test this hypothesis, the model will be evaluated using the OHSUMED test set, and compared to the traditional vector space model using standard performance metrics.