Hiromitsu Shiina

Regional Information Video Searches Using Word Searches Generated by Twitter Posts

There are currently many services available on the Internet including map search sites such as Yahoo! Map, and video search sites such as YouTube. However, there is a limit to the information that can be obtained from each of these services. For example, on map search sites you can find the location of an establishment on a map, however, you cannot obtain information about the establishment itself or view videos of events being held there. On the other hand, although you can find videos on video sites, it is then difficult to find out the locations where the videos were taken in the first place. In addition, tourist information sites are able to provide information such as images and locations, but they offer few videos of events or detailed information about the establishment, limiting the overall amount of information that you are able to obtain at any one time. Our system is, therefore, based on the idea of combining multiple services to obtain many different types of useful information. However, completeness of the search results is low based on a search only using the facility name. By extracting words related to facilities from Twitter posts related to the searched facility and also using that word, many variations to the search results were given. In order to extract words related to facilities from Twitter posts, the results of a machine based learning classification were used in advance to determine whether or not the Tweet relates to a facility.

Extracting Topic Changes through Word Co-occurrence Graphs from Japanese Subtitles of VOD Lecture

A number of universities currently utilize VOD in e-Learning systems, a form of blended learning that combines WBT (Web-Based Training) with classroom instruction. Many universities also offer e-Learning lectures, which offer slides and lecture video with VOD. However, because lectures that use standard VOD focus on letting students take the lecture, most do not include advanced search methods or implement topic detection or extraction. To aid for e-Learning users, we have been developing search facilities for VOD lectures. For this purpose, this study proposes a method of extracting the topic by creating a graph of keywords and their co-occurrence words from the lectures subtitles and then extracting from variations in the word the co-occurrence graphs between sentences.

Similar Comment Analysis Using Selector Rating Estimates in Relation to Comments

Though multiple-choice analysis is widely used in questionnaire analyses, free descriptive answers are also used in conjunction with it. In this study, by looking at the individual freely-described comments from the perspective of the questionnaire as a whole and estimating what kind of selectors are selected, we are able to extract similar freely-described comments and generate comments estimated from selector selection examples. The main features of this study are the thinking surrounding the general rating of comments, the generation of feature vectors for convolutional neural networks (CNN), and, in particular, the use of CNN as a filter.

Classification of short comments by weighted tree kernels using the hierarchy of Wikipedia

With the rise of Web services for posting tourist facility and product reviews, and short comments such as Twitter, there are increasing calls for a method that alleviates the burden of creating corpus and identifying text both automatically and accurately. Further, in the lecture questionnaires used for university lectures, the free reply answers are all short text, and classification of the short text could be used in evaluating lectures. Until now, most methods have used the appearance frequency of vocabulary, based on Bag-of-Words, as the basic method of classifying text. In this study, however, we converted short comments, such as tweets, into tree structure data for category classification, and classified according to a weighted tree kernel, in which weight is further attached to the tree node sections on the tree kernel.

Analysis of Free Description in Lecture Questionnaires Using Word Rank Affiliation Probability

To enrich structural university and graduate school education, faculty development (FD) activities are conducted to improve faculty education and research guidance capabilities. In terms of the content of FD activities, classroom observation between faculty members and participation in seminars and training events at the institution level can be considered. As a part of these activities, many universities request students to fill out lecture questionnaires as a means of evaluating the facultys educational activities. However, the evaluation discrepancies between students and the faculty is a problem faced while analyzing these lecture questionnaires. In this study, we estimated the evaluation for some of the comment evaluations (sheet section) given in the free answer section of the lecture questionnaire. The evaluation was conducted manually from differing standpoints for students and faculty. Moreover, we evaluated the estimation of words included in the content, extracted useful comments, and evaluated the faculty. Furthermore, we assessed the differences between evaluators with different standpoints. In particular, in this study, an evaluation estimate of both words and comments, as well as a recursive evaluation of comments and words was performed. For the comment evaluation method, the six-stage Likert scale was used. When ranking evaluations, the evaluation adopted a contaminated normal distribution for the affiliation probability.

Evaluation of Free Answer Comment Using Machine Learning by Word Evaluation

There is a call for an increase to education quality in response to FD (Faculty Development) activities becoming mandatory due to a revision of the University Establishment Standards in 2008. A survey on lectures was carried out as part of FD activities and used for improving lectures based on lecture evaluation and satisfaction from students. However, it is difficult to give an objective evaluation of this text data since these subjects of evaluation are wide-ranging, such as expectations towards lectures or opinions on teachers. An aggregation of the opinions gained through surveys also has its limits for manual assessment because of the heavy artificial costs and time costs required. Therefore, it is difficult to evaluate all comments given in a survey. For this reason there has been extensive research done in recent years on categorizing evaluation texts and documents through evaluation expression of words and machine learning. There has been related research estimating the sentiment polarity of the entirety of the writing using both positives and negatives that appear within, as well as research that extracts elements of evaluation information and measures its sentiment polarity. Machine learning for writing categorization has also achieved a high level of accuracy, with research existing which uses the Naive Bayes Classifier and a Support Vector Machine. This report uses actual open ended responses to surveys to describe a method for re-evaluating a comment by estimating the evaluation of words within comments from a comment evaluation done through machine learning, as well as a method that uses a dictionary of emotional words to attach a polarity value to vocabulary and estimate the evaluation of the body of a comment from that value. Finally, it describes the results of teacher and lecture evaluations from each comment evaluation.

Classification of Comments by Tree Kernels Using the Hierarchy of Wikipedia for Tree Structures

Many web services posting short comments such as product reviews and Twitter have been provided. We consider that automatic and accurate text classification may lead to develop new web services and system. In the past, the frequency of appearance of words by bag-of-words have been often used for text classification as a basic technique. In contrast, we propose a technique to classify tweets using tree kernels created by the categories of Wikipedia in this study. In addition, we developed a retrieval system for tourism videos by applying the technique to tweets related to tourism.

Method of Classification through Normal Distribution Approximation Using Estimating the Adjacent and Multidimensional Scaling

A Two types of classification methods are applied performing classification using machine learning: (1) those for which there is a presumption of data distribution using kernel functions, such as in support vector machine (SVM) and (2) those for which data distribution is not presumed, such as the k-NN. With such methods, it is easy to obtain data whose values are close for the class as a whole. In addition, it is assumed that there is a high probability that data with close values apparently belong to the same class. In contrast, while it may be easy to obtain approximate values of data for each class, there exists a relatively high probability of approximate values manifesting as data of the same class. Consequently, there is a high probability of the same class of data appearing around the position of parameters of the data manifested in feature space where data parameters are obtained. This study proposes machine learning algorithms that approximates this using a density function of normal distribution. In addition, if small amounts of data of a different class exist where data of the same class is being collected, the impact of those classes will be significant. This study propose two types of algorithms. One is a method of calculating the influence of the proximity of the training data for the entire feature space calculates. The other is a method of calculating the effect on the entire feature space for each training data. Both methods utilize two parameters as the influence of each training data and the range to be used as neighborhood data. Two parameters determine the quasi-optimal solution by the steepest descent method. Furthermore, in order to reduce the density of the influence of the training data, we propose improved method that relocates the training data from the distance between the training data by multidimensional scaling as preprocessing.

Programming Lecture Support System Linked to a Lecture Towards Non-Kanji Countries

For students from countries which do not use kanji, learning the Japanese language, particularly the memorization of kanji, is quite difficult. Moreover, it is believed that an inability to read kanji makes it difficult to comprehend lectures. Therefore, in an effort to improve the level of comprehension of the kanji used in lectures and the content of the lectures themselves, we developed a learning system for kanji and programming. The system we developed is linked with the lectures, and is comprised of kanji furigana system, which obtains kanji data from the materials (slides) used in lectures for exchange students and provides materials with furigana created from the obtained kanji data, kanji learning system and kanji check system, which measure the learning state for kanji, and a programming procedure testing system based on the sequencing of Japanese sentences. The kanji check system ascertains the learning state through questions involving choosing the correct readings for words, the ratio of furigana supplied for kanji by kanji furigana system is determined for each lecture based on a furigana rate curve created by logistic function kanji furigana system and it is frequently updated based on the results of the kanji check system, thereby providing a learning experience more closely matching the learners level.

Estimating Word Difficulty Using Semantic Descriptions in Dictionaries and Web Data

Identifying of word difficulty is useful to teachers and language learners. However, since the word difficulty has not been evaluated in all dictionary entry words, to estimate the word difficulty of documents and presentations from the aspects of language use is difficult. As fundamental research, we are attempting to evaluate the word difficulty from native and non-native learners. In the estimation of the word difficulty, we use semantic description in dictionary and Web descriptions by the result of search engine to create learning parameter, then all dictionary entry words are estimated the word difficulty by using multiclass support vector machine.