Analysis of evaluations of living donors for liver transplantation utilizing text mining approach

本研究的目的是利用近年来流行的文本挖掘技术，对2004年收集的活体肝移植（LDLT）供者的免费问卷进行分析，使用两种文本挖掘软件True Teller（TT：Nomura Research Institute Ltd.）IBM SPSS Text Analysis for Surveys（STAfS：IBM）。本文主要分析了自由答案中词汇的使用频率和共现情况，并对软件在分析过程中的可用性和局限性进行了评价。健康焦虑文本挖掘结果表明，“模糊”、“未来”、“焦虑”、“最近”和“没有”等词汇的使用频率和共现率较高。这表明，他们对未来的健康感到模糊的不安，但目前没有症状。“元气”与“疲劳”、“肠”与“梗阻”也各有不同，且这三组被认知程度不同。另一方面，当非卫生行业的调查数据包含了几乎相同现象的各种表达时，词典的功能（包括同义词登记）对分析产生了影响，在TT和STAfS的分析过程中，词典的制作花费了大量的时间，但它们的分析结果是相似的。但是TT具有从数据导入到输出图形的自动化用户界面，STAfS是一个使用IBM SPSS Statistics准备数据集进行统计分析的工具，对后续分析具有很高的灵活性，文本挖掘使用客观的过程，可以从大量的文本数据中直观地掌握难以区分的词的频率和关系，这对今后分析自由回答问题具有重要意义。

The purpose of this study was to analyze the free answers of the investigation about the living donor liver transplantation (LDLT) donor which collected in 2004, utilizing text mining which had become popular in late years.We used two software for text mining ; True Teller (TT : Nomura Research Institute Ltd.) and IBM SPSS Text Analysis for Surveys (STAfS : IBM). We mainly analyzed frequencies and co-occurrences of words in free answers, and evaluated usability and limitations of software in analyzing procedures.Text mining about anxiety about health showed that "vague", "future", "anxiety", "recent" and "nothing" had high frequencies and co-occurrences. It suggested that they felt a vague uneasiness for future health without present symptoms. "Vitality" and "fatigue", and "intestine" and "obstruction" also have them, and these three groups were recognized different from each other. Their anxieties included not only vague but specific.On the other hand, the dictionary function including the synonym registration had an influence on the analysis when survey data from the non-health profession which includes various expressions of nearly same phenomenon.In both analytical procedures of TT and STAfS, the making of dictionaries for analyses took a lot of time, and results of them were similar. But TT had automatic user interface from data import to output graph, STAfS was a tool for preparing dataset for statistic analysis using IBM SPSS Statistics and had high flexibility for following analysis.Using an objective procedure, text mining could visually grasp frequency and the relationship of the word that was hard to distinguish from a large quantity of text data, and there was significance to use it for future analyses of free answers.