一种改进的基于关系的信息检索技术

发布时间：2019-03-24 14:18

【摘要】：有研究工作表明现有的基于关系的信息检索技术(RIR)优于基于项(term)或基于语义(concept)的IR技术,但仍存在显而易见的缺陷,即不能明确关系本身,只能表达概念A,B是存在关系的概念对。本文提出一种改进的基于关系的IR技术—IRIR(Improved Relation-based Information Retrieval),就是要明确关系的取值和属性,整合概念对和关系的信息为三元组表达式(triple),通过以下匹配方法获取未知信息。对于文本中出现的知识表示为R(rela-tion)[First Concept,Second Concept],对于疑问代词(如what)开头的查询表达为R(relation)[First Concept,Un-known],对于疑问副词(如how)开头的查询表达为R(Unknown)[First Concept,Second Concept],当文本与查询的三元组表达式中已知部分匹配一致时,则得到查询未知部分的一个取值。由此,既可以实现类似QA(query answer)功能,又可以完成精确信息检索。基于Drexel大学DMBioinformatics Lab开发的生物医学文献搜索引擎(2004版,简称为RIRS),我们开发了一个能实现IRIR技术和功能的实验IR引擎—IRIRS(Improved Relation-Based IR Sys-tem),该系统使用UMLS和WordNet两大权威本体库分别确定概念和关系,在博士入学考试英语阅读理解测试集上的实验结果令人满意,IRIRS将文字段级别的检索精确度MA PP(Mean average passage precision)从RIRS的64.44%提高到74.28%。这表明,在IR中应用改进的基于关系的信息检索技术是非常具有探索价值的。
[Abstract]:Some studies have shown that the existing relational-based information retrieval technology (RIR) is superior to the item-based (term) or semantic-based (concept)-based IR technology, but there is still an obvious defect, that is, the relationship itself cannot be clearly defined, and only the concept A can be expressed. B is the concept pair of existential relations. In this paper, an improved relationship-based IR technique-IRIR (Improved Relation-based Information Retrieval), is proposed to clarify the values and attributes of the relationship, and to integrate the information of the concept pair and relation as a triple expression (triple),. The following matching methods are used to obtain unknown information. For the knowledge that appears in the text is represented as R (rela-tion) [First Concept,Second Concept], for a query beginning with a interrogative pronoun (such as what) expressed as R (relation) [First Concept,Un-known], For a query beginning with a query adverb (such as how) expressed as R (Unknown) [First Concept,Second Concept], a value of the unknown part of the query is obtained when the text matches the known part of the triplet expression of the query. As a result, it can not only realize QA (query answer)-like function, but also complete accurate information retrieval. Based on the biomedical literature search engine (version 2004, RIRS), for short) developed by DMBioinformatics Lab of Drexel University, we developed an experimental IR engine-IRIRS (Improved Relation-Based IR Sys-tem, which can realize the technology and function of IRIR. The system uses two authoritative libraries, UMLS and WordNet, to determine the concepts and relationships respectively. The experimental results are satisfactory on the English reading comprehension test set of the doctorate entrance examination. IRIRS increased the text-level retrieval accuracy MA PP (Mean average passage precision) from 64.44% of RIRS to 74.28%. This shows that the application of improved relationship-based information retrieval technology in IR is of great value for exploration.
【作者单位】：国防科学技术大学计算机学院;北京航空航天大学计算机科学与工程学院;
【基金】：国家自然科学基金项目(60573057,60473057,90604007)的资助
【分类号】：TP391.3

【共引文献】