Journal of Japan Society for Fuzzy Theory and Intelligent Informatics
Online ISSN : 1881-7203
Print ISSN : 1347-7986
ISSN-L : 1347-7986
Volume 30 , Issue 6
Showing 1-12 articles out of 12 articles from the selected issue
Erratum
Data Mining and Knowledge Emergence
Original papers
  • Naoki SHINO, Ryosuke YAMANISHI, Yoko NISHIHARA
    2018 Volume 30 Issue 6 Pages 779-787
    Published: December 15, 2018
    Released: December 15, 2018
    JOURNALS OPEN ACCESS

    This paper proposes a method to discover the relationships among ingredients based on various aspects. The proposed method can be used to recommend another ingredient that should be exchangeable with a given ingredient in a recipe, i.e., the alternative-ingredients. The aspects to relate the ingredients with each other are A) the affinity for the remained ingredients, B) the similarity of role in the recipe with the exchanged ingredient, and C) the similarity of texture with the exchanged ingredient. In the proposed method, aspect A) is estimated by the co-occurrence relations among ingredients in the recipe database. For aspects B) and C), the proposed method calculates the similarities of both the role in the recipe and the texture. The similarities are estimated by the ones of respective word vector obtained from both the list of the ingredients and the recipe reviews. Through the experiments, it was confirmed that the recommendations of alternative-ingredients with each aspect were similar to the human judge. It was shown that the proposed method recommends appropriate alternative-ingredients with each aspect.

    Download PDF (1036K)
  • Shin-ya SATO
    2018 Volume 30 Issue 6 Pages 788-795
    Published: December 15, 2018
    Released: December 15, 2018
    JOURNALS OPEN ACCESS

    In an exploratory search where a user is likely to be unsure about the goal of her/his search, thus needs to clarify the goal through trial and error, it is important to support the user’s understanding of “what she/he is trying to find out” (search target). In order to deepen the understanding of the target in the search process, it is necessary for the user not only to obtain more detailed information by so-called narrowing down but also to grasp the relationship with other concepts and to understand various aspects of the target concept. In this paper, we propose a method of discovering relevant information for suppoting multidimensional understanding of the target concept. The enumerations in Wikipedia articles such as tables and lists are used to find the information. We conducted a user experiment for verifying usefulness of the proposed method. Experimental results show that the proposed method can find a larger amount of useful information compared to other methods including query recommendation of search engines. It is also shown that the proportion of useful information among related information found by the proposed method exceeds 70%.

    Download PDF (633K)
  • Yoko NISHIHARA, Reona TAKAYAMA, Kensuke HISHIDA, Ryosuke YAMANISHI
    2018 Volume 30 Issue 6 Pages 796-803
    Published: December 15, 2018
    Released: December 15, 2018
    JOURNALS OPEN ACCESS

    With the rapid progress of computer Shogi and news of attractive professional players, many people are interested in Shogi. Shogi commentators on TV programs and computer software of Shogi game often indicate the position of a game (the information of superiority of a game). The indicated information is almost incomprehensible for Shogi beginners because the information includes predictions of the future situations. We guess that the beginners can understand the position and the progress of a game if the position is evaluated from the present arrangement of Komas. This paper proposes a support method for Shogi beginners in understanding Shogi games by evaluating battlefields and a king’s danger degree by using the present arrangement of Komas. We made an interface for visualization of evaluated items and had experiments with the interface. Experimental results showed that the proposed method can support the Shogi beginners to understand the position of a game. They gave more opinions about understanding of the position of a game.

    Download PDF (1368K)
  • Yuki ANDO, Yohei SEKI
    2018 Volume 30 Issue 6 Pages 804-814
    Published: December 15, 2018
    Released: December 15, 2018
    JOURNALS OPEN ACCESS

    To invite the tourist effectively, it is essential to place an advertisement in the cities where many citizens have visited the leisure venue for sightseeing. The mood of the citizens’ concerns for the cities or for the districts are different according to the cities, and the reason of the concerns is also different with their purposes as homecoming or sightseeing. In this paper, we propose a method to visualize the mood of the citizen concerns for the regional names (city names, district names, etc.) with clustering the similar objects using the Twitter data crawled for each city. At first, we collect the citizen users in Twitter using profiles to crawl the citizen users’ tweets and also select the tweets including regional names and movement verbs to regions. Then, we create distributed vector representation of words based on skip-gram using the selected tweets. We also choose two terms pair to compare the mood of citizen concerns for regional names based on movement intention and factuality. The regional names are visualized based on the difference of the normalized similarities of the regional names between two chosen terms pair using Z-score. They are classified using clustering technique to clarify the similar concerns of citizens for regional names. To verify the effectiveness of the proposed method, we conducted the experiment using Twitter users in eight cities in Japan. We confirmed that the popular sightseeing spot, residence area, and homecoming spot were visualized properly and the cities sharing tourism purposes or in short distance area tended to be classified in the same cluster.

    Download PDF (1157K)
R&D Papers
  • Pin Chieh CHENG, Kenichiro KOBAYASHI, Takehiko HASHIMOTO, Kazunori SAI ...
    Type: 実践研究論文
    2018 Volume 30 Issue 6 Pages 815-822
    Published: December 15, 2018
    Released: December 15, 2018
    JOURNALS OPEN ACCESS

    This paper describes a property search system called Tech Supplier, which is designed for helping expert searchers in a real estate company to find properties for renovation. ReTech (Real Estate Technology) or PropTech (Property Technology) is known to be one of the most promising topics following FinTech (Financial Technology). Among several potential technologies in ReTech, property search attracts a lot of attention. Different from such item as books, movies, and daily goods, properties are difficult to search for ordinary persons because of several reasons: less opportunity to buy properties, imbalance of knowledge about the properties between customers and salespersons in real estate companies. Therefore, the salesperson usually helps customers to find their relevant properties based on his / her domain knowledge. To improve the efficiency of property search for renovation, Tech Supplier has several functions: floor plan identification function, elimination of apparently irrelevant properties from a retrieved list, and properties search with ranking function. Among those functions, this paper focuses on the latter two ones. Experiments are conducted with the help of expert searchers in a real estate company in terms of search efficiency of the proposed system and effectiveness of a learning to rank. Experimental results show that relevant properties can be found within a higher rank of retrieved lists by introducing a learning to rank. It is also observed that time spent on a session (a series of searches performed for a certain customer need) by expert searchers using Tech Supplier is reduced to 55% of an existing system.

    Download PDF (1394K)
Regular
Original Papers
  • Motohiro TAKAGI, Daiki HIGURASHI, Akito SAKURAI
    2018 Volume 30 Issue 6 Pages 823-831
    Published: December 15, 2018
    Released: December 15, 2018
    JOURNALS OPEN ACCESS

    In this paper, we propose an image quality estimation method using deep learning. In the conventional image quality estimation method, image characteristics were analyzed for each distortion of the image, and a model was constructed. In recent years, image quality estimation method that learns automatically the relationship between distortion and image quality using machine learning has been proposed. Furthermore, image quality estimation method using deep learning which is frequently used for general image recognition has been proposed. In this paper, we propose image quality estimation method which using deep neural network that considers the type of distortion. We show that our method improves estimation accuracy of image quality compared with conventional CNN model.

    Download PDF (3449K)
  • Rina HAYASHI, Shohei KATO
    2018 Volume 30 Issue 6 Pages 832-839
    Published: December 15, 2018
    Released: December 15, 2018
    JOURNALS OPEN ACCESS

    In recent years, robot-assisted therapy is attracting attention which could give healing effects through interaction with robots. Many of robots used for robot-assisted therapy react by voice simply. In this paper, we proposed introducing the idea of synchrony to such robot’s voice. And we verified influence on psychological and physiological stress relief effects by synchronizing the fundamental frequency, which is most synchronized when communication is active, of the robot’s voice with that of the proximate utterance of the user. As a result, we confirmed that psychological and physiological stress relief effects, especially relieving fatigue, improving vigor, and increasing alpha waves, obtained by the voice, which fundamental frequency is synchronized with that of the proximate utterance of the user, are significantly higher. On the other hand, we also confirmed that psychological stress relief effects, especially reducing confusion, obtained by the voice, which fundamental frequency is adjusted randomly independently of that of the proximate utterance of the user, is significantly lower. These results suggest that psychological and physiological stress relief effects can be drawn out by synchronizing the fundamental frequency of the robot’s voice with that of the proximate utterance of the user instead of adjusting blindly.

    Download PDF (2246K)
Short Notes
  • Ryota NISHIMURA, Takumi NAGAO, Ayato ICHIMANDA, Norihide KITAOKA
    2018 Volume 30 Issue 6 Pages 840-845
    Published: December 15, 2018
    Released: December 15, 2018
    JOURNALS OPEN ACCESS

    In a super-aging society in recent years, declining auditory function of elderly people is regarded as a problem. If the hearing function deteriorates, communication with conversation also becomes difficult. The elderly support system using the spoken dialog system has been developed, however the synthesized speech used in such a system is difficult for the elderly to listen. In this study, we conducted listening experiments for elderly people to analyzed the auditory function of elderly people. In the listening experiment, word intelligibility tests were conducted to analyze the discrimination rate for phonemic unit (consonant part, vowel part). As a result, there were many confusion in phoneme perception between fricatives, affricates and plosives. From this result, we applied consonant emphasis processing to speech based on the auditory function of the elderly people, and then we investigated whether the processed speech is easy for the elderly to listen. The consonant emphasis processing was performed on phonemes where there were many confusion in phoneme perception, such as /k/, /s/, /t/, /h/, /ky/, /sy/, /ch/. The processing method is to amplify the amplitude of the consonant part to 400% of the original speech. In the evaluation experiment of processed speech, we conducted listening experiments for the same subjects in the word intelligibility test and compared the correct answer rates. As a result, correct answer rate for processed speech is improved in some phonemes.

    Download PDF (1705K)
feedback
Top