面向高考阅读理解的句子语义相关度
Semantic relevancy between sentences for Chinese reading comprehension on college entrance examinations

高考阅读理解选择题是基于背景材料，通过对材料的“理解”从多个选项中选出最佳选项。由于提供的背景材料相对较短且关键信息极具隐藏性，答案可能无法在背景材料中直接找到，因此从背景材料中挖掘信息并与选项进行相关性分析是解答该类问题的关键，而句子级的语义相关性分析是背景材料与选项相关性分析的基础。该文通过对大量高考科技文文意理解类选择题进行分析，提出基于多维度投票算法的句子语义相关度计算方法。该方法将不同维度的语义相关性作为度量标准，运用投票算法的思想，选取问题的最佳选项。在近十年北京市高考真题上进行测试，解答准确率为53.84%，验证了该方法的有效性。
Abstract：Multiple-choice reading comprehension questions in the Chinese College Entrance Examination are based on the given background material with the reader selecting the best option from a number of options. The answer may not be directly found in the background material since the passage is relatively short and the key information is hidden. Thus, information mining from the background material and semantic relevancy analyses with options are keys to solving the problem, with sentence level semantic relevancy analysis as the foundation. This paper presents an algorithm to calculate the semantic relevancy between sentences based on Multi-Dimension Voting by analyzing large numbers of multiple-choice questions from Chinese scientific article text understanding questions from college entrance examinations. The method utilizes the voting algorithm to take advantage of different size metrics to select the best option. The algorithm accuracy for the national college entrance examination of Beijing text understanding questions is 53.84%, which verifies the validity of the method.