【摘要】：閱讀理解是目前NLP領(lǐng)域的一個(gè)研究熱點(diǎn)。閱讀理解中好的復(fù)雜問(wèn)題解答策略不僅要進(jìn)行答案句的抽取,還要對(duì)答案句進(jìn)行融合、生成相應(yīng)的答案,但是目前的研究大多集中在前者。該文針對(duì)復(fù)雜問(wèn)題解答中的句子融合進(jìn)行研究,提出了一種兼顧句子重要信息、問(wèn)題關(guān)聯(lián)度與句子流暢度的句子融合方法。該方法的主要思想為:首先,基于句子拆分和詞重要度選擇待融合部分;然后,基于詞對(duì)齊進(jìn)行句子相同信息的合并;最后,利用基于依存關(guān)系、二元語(yǔ)言模型及詞重要度的整數(shù)線(xiàn)性規(guī)劃優(yōu)化生成句子。在歷年高考閱讀理解數(shù)據(jù)集上的測(cè)試結(jié)果表明,該方法取得了82.62%的F值,同時(shí)更好地保證了結(jié)果的可讀性及信息量。
[Abstract]:Reading comprehension is a hot research topic in the field of NLP at present. The good complex question solving strategy in reading comprehension not only needs to extract the answer sentence, but also merges the answer sentence to generate the corresponding answer, but most of the current research focuses on the former. In this paper, sentence fusion in complex problem solving is studied, and a sentence fusion method which takes into account the important information of sentence, the correlation degree of question and the fluency of sentence is proposed. The main ideas of this method are as follows: firstly, the fusion part is selected based on sentence resolution and word importance; then, the sentence is merged based on word alignment; finally, sentences are optimized by integer linear programming based on dependency, binary language model and word importance. The test results on the reading comprehension data set of college entrance examination over the years show that the method achieves 82.62% F value, and better ensures the readability and information of the results.
【作者單位】： 山西大學(xué)計(jì)算機(jī)與信息技術(shù)學(xué)院;山西大學(xué)計(jì)算智能與中文信息處理教育部重點(diǎn)實(shí)驗(yàn)室;
【基金】：國(guó)家高技術(shù)研究發(fā)展計(jì)劃(863計(jì)劃)項(xiàng)目(2015AA015407) 國(guó)家自然科學(xué)青年基金(61100138,61403238) 山西省自然科學(xué)基金(2011011016-2,2012021012-1) 山西省回國(guó)留學(xué)人員科研項(xiàng)目(2013-022) 山西省高�？萍奸_(kāi)發(fā)項(xiàng)目(20121117) 山西省2012年度留學(xué)回國(guó)人員科技活動(dòng)擇優(yōu)項(xiàng)目
【分類(lèi)號(hào)】：TP391.1
，

本文編號(hào)：2517566