WEKO3
アイテム
Exploring weighting schemes between report assignments on medical informatics written by nursing and radiologic technology students
https://repo.qst.go.jp/records/71781
https://repo.qst.go.jp/records/717815485c7b2-c8e9-4619-b92c-d38c0a0f7895
| Item type | 会議発表用資料 / Presentation(1) | |||||
|---|---|---|---|---|---|---|
| 公開日 | 2015-08-07 | |||||
| タイトル | ||||||
| タイトル | Exploring weighting schemes between report assignments on medical informatics written by nursing and radiologic technology students | |||||
| 言語 | ||||||
| 言語 | eng | |||||
| 資源タイプ | ||||||
| 資源タイプ識別子 | http://purl.org/coar/resource_type/c_c94f | |||||
| 資源タイプ | conference object | |||||
| アクセス権 | ||||||
| アクセス権 | metadata only access | |||||
| アクセス権URI | http://purl.org/coar/access_right/c_14cb | |||||
| 著者 |
Nishimoto, Naoki
× Nishimoto, Naoki× M., Ito Yoichi× 横岡, 由姫× Yagahara, Ayako× Uesugi, Masahito× Tsuji, Shintaro× Fukuda, Akihisa× Ogasawara, Katsuhiko× 横岡 由姫 |
|||||
| 抄録 | ||||||
| 内容記述タイプ | Abstract | |||||
| 内容記述 | 1. Introduction Paramedical staffs such as nurses and radiologic technologists have encountered ethical issues daily in healthcare environments. These issues seldom represent major ethical dilemmas, but consist of minor issues that they face in their everyday contact with patients. Hussey and Allmark [1,2] reported on some of these issues faced by nurses. Although such studies discuss the results of questionnaire surveys or proposed revisions to educational programs, they rarely analyze differences in the background knowledge of students. Faculties should be able intuitively to comprehend similarities and differences among students with multiple majors through grading of their assignments. To the best of our knowledge, however, there are no articles reporting the quantitative analysis of the expression differences of medical informatics education based on background knowledge. The motivation for this study was to illustrate the differences in literal expressions for students’ background knowledge using statistical techniques or other quantitative method based on written materials. The vector space model is used to evaluate document similarity in information retrieval. To distinguish the documents, several weighting scheme was proposed. However which weighting is the best is not so clear. The purpose of this study is to explore the better weighting schemes of the vector space model and to extract the key words for the literacy among the nursing and radiologic technology major. 2. Material and methods A natural language processing technique was used for dividing sentences to words automatically and word counts were obtained. We used Juman (Kyoto University, Japan) for dividing the sentences. Before compute the inner product of i dimension term vector, we calculated the weighted count. Table 1 shows the local weight for TF (term frequency) and global weight for the IDF (inverse document frequency). “lij” corresponds to the weight for the ith term in the jth document. “gi” corresponds to the global weight for the ith term. “fij” is the count of the ith term in the jth document and Fi is the count of for the ith term over the entire documents. ni is the number of documents that has ith term. Generally, local weight affects the sensitivity and global weight affects positive predictive value for the information retrieval system[3-5]. \n3. Examples and computation Example: We analyzed report assignment sets for 85 nursing and 44 radiologic technology students. Students were sophomores and juniors in each major, respectively. They had completed general education and basic healthcare education, such as anatomy, physiology, and some specialties, but had not yet completed clinical training in a hospital. Students answered the above question in written, free-text format within a limit of 1,000 characters in the Japanese language. This assignment was designed to probe the dilemma between the ethical problem of privacy and contribution to academic progress. The students from each department were required to write their opinion clearly on whether they agreed or disagreed with the question. 4. Results Fig. 1 shows the similarity score distribution and heat map visualization between two students’ report based on the logarithmic TF and the probabilistic IDF. Using the particular combination, we found that the similarity score distribution did not diverge. For example, the combination of binary weight and IDF produce lower or higher similarity score. There were less variances. On the other hand, the combination of logarithmic TF and the probabilistic IDF made the similarity score diverged on both department. We made a heat map for the above combination. Similarity score has a distribution around 0.5 and lower score, we could point out the relatively high score on the particular report. Arrows were set at the corresponding report. Focusing on the 34th report in Radiologic department, it also had high scores among other reports indepent on the department. The red point on the diagonal line, it is the maximum similarity score because it compares the own report. Some pixels looked lower score, it shows overlapped pixels when they showed the heat map. 5. Discussion We used binary weight at first, some global weight for the IDF helped the similarity score diverged, it is clear that the binary weight makes the score goes extreme value. For 0 or 1 on binary weight, it is difficult to obtain the difference between the two reports. Although some contrasts become obvious, subtle difference in word usage would be lost by the weighting scheme. Table 1. Weighting scheme for term frequency and inverse document frequency. In spite that we compute the similarity score for the same datasets, the shape of the distribution has been changed depending on the weighting scheme. IDF and global frequency IDF weighting suppress frequency of the term and they draw the score to lower and upper from fig.1. The histogram combination of binary weight & probabilistic IDF was closed to normal. However, it is difficult to say that this weighting scheme is the better than the other because the target document is students’ reports. In this study, we did not apply any expression normalization because we could not predict what type of terms would be used. The procedure would help to make accurate part-of-speech tagging. Thus, although expression variation seems to have an effect on dividing sentences into meaningless particles, because the accuracy of JUMAN was 99% in previous research, we decided that we could ignore the effect of parsing error since the effect would be small. In conclusion, the weighting scheme affects the similarity score distribution. Further investigation is needed for which weighting scheme is the best solution for the target documents. |
|||||
| 会議概要（会議名, 開催地, 会期, 主催者等） | ||||||
| 内容記述タイプ | Other | |||||
| 内容記述 | The AMIA Annual Symposium | |||||
| 発表年月日 | ||||||
| 日付 | 2012-11-03 | |||||
| 日付タイプ | Issued | |||||
