基于语义处理技术的信息检索模型

信息爆炸是当今信息社会的一大特点，如何在海量的信息中有效地找到所需信息因而成为了一个关键问题，语义检索技术是解决这一问题非常有潜力的方法。本文对信息检索中的若干关键问题进行了研究，提出了基于语义处理技术的信息检索模型——SPTIR，该模型主要包括以下关键技术：基于词义消歧的语义查询扩展、基于词汇语义相关性度量的查询优化和基于文档语义相关性的检索结果重排序。最后使用大型测试数据集和多项性能指标对SPTIR模型的检索性能进行了试验评估，实验结果充分验证了SPTIR模型的竞争优势以及该模型采用的各项语义处理技术对提高检索性能所起的积极作用。

应用搜索引擎计算语义相关度的实现

具备模仿人类判断能力的语义相关度在很多方面尤其是自然语言处理领域中处于非常重要的地位。已有的算法或依赖于WordNet层级结构或由于自身局限性无法满足精确计算的要求,由此提出了一种基于搜索引擎的语义相关度算法,根据对两关键词网络搜索时系统返回的搜索页数来计算二者的语义相关度值。通过与其他算法进行对比实验可看出该算法与专家值重合度要远高于其他算法,而且对于计算对象无词性、语法以及语言等方面的限制,优越性较为明显。

Assessing Ebola-related web search behaviour: insights and implications from an analytical study of Google Trends-based query volumes

Background:The 2014 Ebola epidemic in West Africa has attracted public interest worldwide,leading to millions of Ebola-related Internet searches being performed during the period of the epidemic.This study aimed to evaluate and interpret Google search queries for terms related to the Ebola outbreak both at the global level and in all countries where primary cases of Ebola occurred.The study also endeavoured to look at the correlation between the number of overall and weekly web searches and the number of overall and weekly new cases of Ebola.Methods:Google Trends(GT)was used to explore Internet activity related to Ebola.The study period was from 29 December 2013 to 14 June 2015.Pearson’s correlation was performed to correlate Ebola-related relative search volumes(RSVs)with the number of weekly and overall Ebola cases.Multivariate regression was performed using Ebola-related RSV as a dependent variable,and the overall number of Ebola cases and the Human Development Index were used as predictor variables.Results:The greatest RSV was registered in the three West African countries mainly affected by the Ebola epidemic.The queries varied in the different countries.Both quantitative and qualitative differences between the affected African countries and other Western countries with primary cases were noted,in relation to the different flux volumes and different time courses.In the affected African countries,web query search volumes were mostly concentrated in the capital areas.However,in Western countries,web queries were uniformly distributed over the national territory.In terms of the three countries mainly affected by the Ebola epidemic,the correlation between the number of new weekly cases of Ebola and the weekly GT index varied from weak to moderate.The correlation between the number of Ebola cases registered in all countries during the study period and the GT index was very high.Conclusion:Google Trends showed a coarse-grained nature,strongly correlating with global epidemiological data,but was weaker at country level,as it was prone to distortions induced by unbalanced media coverage and the digital divide.Global and local health agencies could usefully exploit GT data to identify disease-related information needs and plan proper communication strategies,particularly in the case of health-threatening events.

一种基于事实知识的实体相关度计算方法

在近来出现的面向实体的搜索服务中,准确地预测实体间的相关程度是至关重要的。该文提出了一种基于实体的事实知识,即利用"实体—属性—属性值"(SPO)记录进行实体相关度计算的方法。该文通过基于属性和属性值的两步概率估计,将实体表示为一个属性值词的概率分布列,并通过比对两个实体共享的属性值词汇得出二者的相关度。实验表明,在用于面向实体搜索的相关实体排序问题上,该文方法达到了80.9%的平均top-5准确率,优于词袋方法和基于查询日志共现的方法。此外,该文通过定量分析,考察了不同领域的用户需求特性对实体相关度计算结果的影响。

基于核函数的搜索引擎研究

通过语义相关度进行搜索引擎的设计是一种有效途径,现通过分析搜索时获得的页面数和各页面的关键词密度,提出一种基于核函数的语义相关度算法。同时,在标准测试集上进行数据实验,并与其它几种已有方法对比,结果显示该方法与专家打分值的Spearman相关系数最高,进一步表明了该算法的有效性,由于新算法中使用的关键词没有词性、语法等限制,且算法简便,所以有利于实际应用和推广。