Despite the widespread adoption of electronic medical records (EMR), the variations in format and terminology across institutions hinder inter-institutional comparisons and feature extraction. This paper proposes a demonstration of extracting frequent disease-specific instruction sequences and efficiently visualizing them with statistical insights, e.g. statistical trends, and abnormal inspection result rates from real multi-institutional EMR data. The utility of the developed visualization tool was presented for decision support and improving clinical processes.

With the increasing use of electronic medical records, medical support from analysis of the accumulated medical information is expected. Currently, new treatment methods and drugs are being developed for the treatment of new diseases, but the transition history of medical orders has yet to be visualized for diseases such as COVID-19. In this paper, we use sequential pattern mining to extract frequent medical orders and then apply the longest common subsequence variant (LCSV) and merged sequence variant (MSV) to analyze the differences in treatment patterns at different times. We also propose three types of sliding window (time interval window, sequence number window, and time-sequence number window) to analyze the transition history of medical orders. As an example, we applied these methods to Japanese electronic medical records covering the first to the fifth waves of COVID-19 and analyzed the differences in medical-order patterns for the five infection waves and the transition history of medical orders. We then visualized the difference with MSV. The results showed that the proposed method can successfully visualize the differences in medical orders between infection waves, and the transition history of medical orders can be revealed. The validity of the results was confirmed by the medical staff involved.

Electronic medical record systems have been adopted by many large hospitals worldwide, enabling the recorded data to be analyzed by various computer-based techniques to gain a better understanding of hospital-based disease treatments. Among such techniques, sequential pattern mining, already widely used for data mining and knowledge discovery in other application domains, has shown great potential for discovering frequent patterns in sequences of disease treatments. However, studies have yet to evaluate the use of medical-order sequence variants , where a “frequent pattern” can include some limited variations to the pattern, or have considered the factors that lead to these variants. Such a study would be meaningful for medical tasks such as improving the quality of a particular treatment method, comparing treatments with multiple hospitals, recommending the best-suited treatment for each patient, and optimizing the running costs in hospitals. This article proposes methods for evaluating medical-order sequence variants and understanding variant factors based on a statistical approach. We consider the safety and efficiency of sequences and related information about the variants, such as gender, age, and test results from hospitals. Our proposal has been demonstrated as effective by experimentally evaluating an electronic medical record system’s real dataset and obtaining feedback from medical workers. The experimental results indicate that the medical treatment history and specimen test results after hospitalization are significant in identifying the factors that lead to variants.

Abstract There are many types of approaches for Paraphrase Identification (PI), an NLP task of determining whether a sentence pair has equivalent semantics. Traditional approaches mainly consist of unsupervised learning and feature engineering, which are computationally inexpensive. However, their task performance is moderate nowadays. To seek a method that can preserve the low computational costs of traditional approaches but yield better task performance, we take an investigation into neural network-based transfer learning approaches. We discover that by improving the usage of parameters efficiently for feature-based transfer, our research goal can be accomplished. Regarding the improvement, we propose a pre-trained task-specific architecture. The fixed parameters of the pre-trained architecture can be shared by multiple classifiers with small additional parameters. As a result, the computational cost left involving parameter update is only generated from classifier-tuning: the features output from the architecture combined with lexical overlap features are fed into a single classifier for tuning. Furthermore, the pre-trained task-specific architecture can be applied to natural language inference and semantic textual similarity tasks as well. Such technical novelty leads to slight consumption of computational and memory resources for each task and is also conducive to power-efficient continual learning. The experimental results show that our proposed method is competitive with adapter-BERT (a parameter-efficient fine-tuning approach) over some tasks while consuming only 16% trainable parameters and saving 69-96% time for parameter update.

Next-item recommendation solutions based on sequential pattern mining have been widely used in empirical studies. However, current solutions do not consider recommendations involving large combinations of items with varied values. For example, inspecting many specimens is key to understanding a patient's current health status and to checking a medical prescription's effectiveness. Typically, a specimen inspection may involve dozens of inspection items selected from more than a thousand possible items, with each item being associated with a measured value. The values themselves will differ for different items. Based on the pattern of previous item values, recommending the next specimen inspection from a huge number of candidate ones, requires that a combination of many inspection items must be processed efficiently. This paper presents a method for vectorizing a combination of items and item values, and identifying clusters of item-set types. From the item-set types that best suit the target input, a set of items is recommended using both frequency and uniqueness. The method was tested experimentally, using real data from a university hospital's electronic medical record system. The results showed that the proposed method can successfully recommend specific item types with the highest precision and recall ratios. The validity of the results was confirmed by the hospital's medical staff.

Electronic medical record (EMR) systems have now been widely adopted to support medical workers. There also has been much interest in the machine-based generation of clinical pathways that can utilize sequential pattern mining (SPM) to extract them from historical EMR systems. However, the existing methods do not protect individual privacy, even though they involve sensitive medical data. To ensure the privacy of individual data, this paper describes two algorithms that deploy differential privacy by adding noise during calculations in the SPM considering time interval for guaranteeing privacy. The proposals can limit the amount of added noise by adding noise to the frequency calculations of only a part of candidate closed sequences. Experiments on real medical datasets show that our proposal can ensure the robust and high utility of mining process even with minimum privacy budget and amount of added noise.

Sequential pattern mining (SPM) is widely used for data mining and knowledge discovery in various application domains. Recently, we have proposed an analyzing method to evaluate the sequence pattern variant (SPV) that is the original sequence containing frequent patterns including variants. Such a study is meaningful for medical tasks such as improving the quality of a disease's treatment method. This paper aims to evaluate the effectiveness of the proposed analyzing method in more detail when it was applied to Electronic Medical Record Systems. Using a real dataset, it is observed that the analyzing method is successful in statistically discovering the meaningful indicators that are leading to the difference between comparative SPVs, such as complicated risk, severity risk of the disease, the length of stay in the hospital and the total medical cost. Moreover, it is observed that the length of stay and the medical cost can gain more benefit from increasing the significance level parameter used in comparing the SPVs.

It is useful to employ electronic medical records to improve medical studies. Based on their experience, medical workers conventionally prepare clinical pathways as guidelines for the typical flow for the medical treatment of each disease. In this study, we propose an approach for verifying existing clinical pathways and recommend variants or new pathways by analyzing historical records. We propose a method based on the application of sequential pattern mining to record logs with handling time intervals between treatments. We also focus on the efficacy of medicines instead of their names because various medicines have the same efficacy and they change dynamically. We evaluated the proposed method using actual logs and the results demonstrated that the proposed method is effective.

We discuss the influences of electronic clinical pathways (ECPs) as a knowledge management system in a university hospital. We conducted a preliminary study wherein we analyzed records from the use of ECPs and interviewed eight nurses in charge of creating ECPs in their respective departments. We then conducted follow-up interviews with a chief nurse, another nurse, and four technical staffs. We found that constructing ECPs involves creating knowledge to enhance the skills and knowledge of nurse apprentices. In addition, ECPs function as tools for knowledge creation in teams that frequently generate and use them. Nurses can use ECPs to facilitate knowledge transfer from senior colleagues to apprentices through the processes of constructing and operating the ECP.

This paper presents a method of acquiring IsA assertions (hyponymy relations), AtLocation assertions (informing of location of objects) and LocatedNear assertions (informing of neigh boring locations) automatically from Japanese Wikipedia XML dump files. To extract IsA assertions, we use the Hyponymy extraction tool v1.0, which analyses definition, category and hierarchy structures of Wikipedia articles. The tool also produces information-rich taxonomy from which, using our original method, we can extract additional information, in this case AtLocation and LocatedNear type of assertions. Experiments showed that both methods produce positive results: we were able to acquire 5,866,680 IsA assertions with 99.0% reliability, 131,760 AtLocation assertion pairs with 93.0% reliability and 6,217 LocatedNear assertion pairs with 99.0% reliability. Our method exceeded the baseline system considering both precision and the number of acquired assertions.

This paper presents our research in detection of emotive (emotionally loaded) sentences. The task is defined as a text classification problem with an assumption that emotive sentences stand out both lexically and grammatically. The assumption is verified exper- imentally. The experiment is based on n-grams as well as more sophisticated patterns with disjointed elements. To deal with the sophisticated patterns a novel language modelling algorithm based on the idea of language combinatorics is applied. The results of experiments are explained with the standard means of Precision, Recall and balanced F-score. The algorithm also provides a refined list of most frequent sophisticated patterns typical for both emotive and non-emotive context.

The existence of social media has made computer-mediated communication more widespread among users around the world. This paper describes the development of an emoticon recommendation system that allows users to express their feelings with their input. In order to develop this system, an innovative emoticon database consisting of a table of emoticons with points expressed from each of 10 distinctive emotions was constructed. An evaluation experiment showed that 71.3% of user-selected emoticons were among the top 10 emoticons recommended by the proposed system.

In this paper we propose a method of filtering excessive amount of textual data acquired from the Internet. In our research on pun generation in Japanese we experienced problems with extensively long data processing time, caused by the amount of phonetic candidates generated (i.e. phrases that can be used to generate actual puns) by our system. Simple, naive approach in which we take into considerations only phrases with the highest occurrence in the Internet, can effect in deletion of those candidates that are actually usable. Thus, we propose a data filtering method in which we compare two Internet-based rankings: a co-occurrence ranking and a hit rate ranking, and select only candidates which occupy the same or similar positions in these rankings. In this work we analyze the effects of such data reduction, considering 1 cases: when the candidates are on exactly the same positions in both rankings, and when their positions differ by 1, 2, 3 and 4. The analysis is conducted on data acquired by comparing pun candidates generated by the system (and filtered with our method) with phrases that were actually used in puns created by humans. The results show that the proposed method can be used to filter excessive amounts of textual data acquired from the Internet.

The current understanding is that human-likeness of a robot leads to better human perception. However, the factors have not been thoroughly studied. We conducted a laboratory experiment to examine two questions: how verb acquisition ability affects human perceptions on human-likeness and familiarity of a humanoid robot, intention to use the robot, and enjoyment and satisfaction of the interaction, and whether human-likeness mediates the links between the effects of interaction of verb acquisition between the human perceptions. The experiment involved 48 participants, and we found that the robot that was able to acquire two Japanese verbs, "oku (to put/to place)" and "hanasu (to move away from)," was perceived by participants as more familiar and satisfying than the one that knew the verbs from the beginning. We also found that human-likeness mediated the links between the effect of verb acquisition ability and other perceptions toward the robot.

OBJECTIVE: At present, most clinical data are exchanged between organizations within a regional system. However, people traveling abroad may need to visit a hospital, which would make international exchange of clinical data very useful. BACKGROUND: Since 2007, a collaborative effort to achieve clinical data sharing has been carried out at Zhejiang University in China and Kyoto University and Miyazaki University in Japan; each is running a regional clinical information center. Methods An international layer system named Global Dolphin was constructed with several key services, sharing patients' health information between countries using a medical markup language (MML). The system was piloted with 39 test patients. RESULTS: The three regions above have records for 966,000 unique patients, which are available through Global Dolphin. Data exchanged successfully from Japan to China for the 39 study patients include 1001 MML files and 152 images. The MML files contained 197 free text-type paragraphs that needed human translation. Discussion The pilot test in Global Dolphin demonstrates that patient information can be shared across countries through international health data exchange. To achieve cross-border sharing of clinical data, some key issues had to be addressed: establishment of a super directory service across countries; data transformation; and unique one-language translation. Privacy protection was also taken into account. The system is now ready for live use. CONCLUSION: The project demonstrates a means of achieving worldwide accessibility of medical data, by which the integrity and continuity of patients' health information can be maintained.

This paper presents CAO, a system for affect analysis of emoticons. Emoticons are strings of symbols widely used in text-based online communication to convey emotions. It extracts emoticons from input and determines specific emotions they express. Firstly, by matching the extracted emoticons to a raw emoticon database, containing over ten thousand emoticon samples extracted from the Web and annotated automatically. The emoticons for which emotion types could not be determined using only this database, are automatically divided into semantic areas representing "mouths" or "eyes," based on the theory of kinesics. The areas are automatically annotated according to their co-occurrence in the database. The annotation is firstly based on the eye-mouth-eye triplet, and if no such triplet is found, all semantic areas are estimated separately. This provides the system coverage exceeding 3 million possibilities. The evaluation, performed on both training and test sets, confirmed the system's capability to sufficiently detect and extract any emoticon, analyze its semantic structure and estimate the potential emotion types expressed. The system achieved nearly ideal scores, outperforming existing emoticon analysis systems.

This paper presents CAO, a system for affect analysis of emoticons in Japanese online communication. Emoticons are strings of symbols widely used in text-based online communication to convey user emotions. The presented system extracts emoticons from input and determines the specific emotion types they express with a three-step procedure. First, it matches the extracted emoticons to a predetermined raw emoticon database. The database contains over 10,000 emoticon samples extracted from the Web and annotated automatically. The emoticons for which emotion types could not be determined using only this database, are automatically divided into semantic areas representing “mouths” or “eyes,” based on the idea of kinemes from the theory of kinesics. The areas are automatically annotated according to their co-occurrence in the database. The annotation is first based on the eye-mouth-eye triplet, and if no such triplet is found, all semantic areas are estimated separately. This provides hints about potential groups of expressed emotions, giving the system coverage exceeding 3 million possibilities. The evaluation, performed on both training and test sets, confirmed the system's capability to sufficiently detect and extract any emoticon, analyze its semantic structure, and estimate the potential emotion types expressed. The system achieved nearly ideal scores, outperforming existing emoticon analysis systems.

By our demonstration we want to introduce our achievements in combining different purpose algorithms to build a chatbot which is able to keep a conversation on any topic. It uses snippets of Internet search results to stay within a context, Nakamura's Emotion Dictionary to detect an emotional load existence and categorization of a textual utterance and a causal consequences retrieval algorithm when emotive features are not found. It is also able to detect a possibility to make a pun by analyzing the input sentence and create one if timing is adequate.

This paper introduces an early stage of a smart toy development project which combines several techniques to achieve a level of conversational skills and knowledge higher than currently available robots for children. We describe our ideas and achievements for three modules which we treat as the most important - topic unlimited talking engine, emotions recognizer and the moral behavior analyzer. We will also mention our novel evaluation method for freely speaking agents and possibilities of adding another module - an automatic joke generator.

This paper describes a novel approach for the automatic generation and evaluation of a trivial dialogue phrases database. A trivial dialogue phrase is defined as an expression used by a chatbot program as the answer of a user input. A transfer-like genetic algorithm (GA) method is used to generating the trivial dialogue phrases for the creation of a natural language generation (NLG) knowledge base. The automatic evaluation of a generated phrase is performed by producing n-grams and retrieving their frequencies from the World Wide Web (WWW). Preliminary experiments show very positive results.

This paper presents the characteristic differences of linguistic and acoustic features observed in different spoken dialogue situations and with different dialogue partners: human-human vs. humanmachine interactions. And it also presents influences of awareness of users on those characteristics. We compare the linguistic and acoustic features of the user’s speech to a spoken dialogue system and to a human operator in several goal setting and destination database searching tasks for a car navigation system. Because it is not clear enough whether different dialogue situations and different dialogue partners cause any differences of linguistic or acoustic features on one’s utterances in a speech interface system , we have performed experiments in several dialogue situations[4]. However, in these experiments the conditions such as voice quality and awareness of users such as impressions on the partner and prejudices against a system have not been considered. And so we collected a set of spoken dialogues in new dialogue situations. To investigate influence of voice quality, we also prepare recorded voice for response of dialogue partners and compared the influences of voice (natural voice, synthetic voice and recorded voice). We also made users answer questionnaire before and after the experiments and investigated characteristic differences caused by awareness of users. Additionally, in order to confirm the usefulness of the results of all experiments, we actually applied acoustic features of users’ utterances and identified the utterances made to a system.

This paper presents a SVM-based prediction approach for constructing personal recommendation system for TV programs. We have applied support vector machine (SVM) to personal prediction of online Internet electronic program guide (IEPG). Our basic idea is to combine SVM and feedback processing into our system, using user-watched histories as retraining data, to realize personal predictions. We evaluate the precision by experiments with open data. The results show that the proposed polynomial kernel SVM system offers a statistically significant increase in performance compared to other method, and this system demonstrates good dynamically adaptive capability.

This paper proposes a learning method for rules of soccer agents' actions in RoboCup Soccer. In general, soccer agents need to do high skilled actions in real time and they should have a mechanism to decide the actions. The soccer agents based on our method refer to some types of rules in order that they go into highly skilled actions. One type of them is the rules that are Hand-coded rules. Another type is the rules that are acquired using inductive learning and they are acquired from examples of actions, automatically. We describe the basic idea for them and indicate examples of rules.

We are aiming to detect local deceleration of Japanese spontaneous conversational speech. We have proposed the variable threshold (VT), which detects local speech rate deceleration from the sequence of time series of mora duration. In this paper, we add a constant term to the VT to detect local deceleration appropriately. The VT is applied to 167 samples of Japanese spontaneous speech taken from a spoken dialogue corpus. The results of the detection are compared with local decelerations which are perceived by a listener. The VT detects 64 phrases among whole 87 decelerated phrases. We confirm the reduction of the incorrect detection of non-decelerated portions by adding a constant term to the VT.

Fundamental studies about differential limen (DL) for word-based speech rate variations in Japanese are described. In our previous study, the DLs are expressed by subtractive difference of mora duration. In this report, however, to fit the expression for various global speech rate, the DLs are expressed by variation ratio of mora duration. We carry out auditory tests with stimuli made by equally lengthening or shortening a duration of a word in a sentence. The subjects’ focus of attention is diffused to get DLs that are used in the normal natural conversations. The obtained DLs are approximately 0.85 for acceleration and 1.18 for deceleration in variation ratio of mora duration.

The variable threshold(VT), which detects the speech rate deceleration, is proposed. The VT varies dynamically depending upon the duration of previous mora in the utterance. The VT should not change rapidly because listener cannot perceive small variations of mora duration. Thus, a set of functions with time constants which decide response speed of the VT is introduced. We apply the VT to six sentences of spontaneous conversational speech. The auditory test of detecting local speech rate deceleration is carried out for the evaluation. The possibility of detecting the local speech rate deceleration by the VT is indicated.

In this paper, we propose a syntactic analysis method using inductive learning from examples. In our proposed method, the system acquires the parsing rules using the examples of parsing result. And the system parses Japanese sentences using the acquired parsing rules. We consider that our proposed method can resolve problems of the rule-based approach and the example based approach. Moreover, we consider that this method can be applied to the other languages. We performed the experiment using transcripts of simulated spoken dialogue corpus. We consider the result of experiment about the ability of our proposed method.

There are many machine translation systems recently. However, the results of these machine translation systems include various errors on a selection of translated word, a dependency relation and so on. The purpose of our research is to correct these errors automatically and improve the translation accuracy by preprocessing. This paper presents a method for preprocessing in machine translation system using inductive learning and results of evaluation experiment.

A lot of unnatural sentences exist in the Japanese which the computer generates in machine translation, dialogue processing and so on. This causes user dissatisfaction. From the point of view of this situation; we previously proposed the method which paraphrases an unnatural Japanese sentence generated by a computer to its natural Japanese sentence with inductive learning. In order to confirm the effectiveness of our proposed method, we constructed the experimental system and carried out the evaluation experiment. The rate of,the number of natural sentences in the number of all input sentences improved from 32.4% to 55.3%. From the result of this evaluation experiment, we have confirmed the effectiveness of our method. In this paper, first of all we explain the outline of our proposal method, and consider the effectiveness of our method, through experimental results.

Example-Based Machine Translation can be applied to languages whose resources like dictionaries, reliable syntactic analyzers are hardly available because it can learn from new translation examples. However, difficulties still remain in translation of sentences which are not fully covered by the matching sentence. To solve that problem, we present in this paper a translation method which recursively divides a sentence and translates each part separately. In addition, we evaluate an analogy-based word-level alignment method which predicts word correspondences between source and translation sentences of new translation examples. The translation method was implemented in a French-Japanese machine translation system and spoken language text were used as examples. Promising translation results were earned and the effectiveness of the alignment method in the translation was confirmed.