人工知能学会論文誌
Online ISSN : 1346-8030
Print ISSN : 1346-0714
ISSN-L : 1346-0714
35 巻 , 5 号
選択された号の論文の7件中1~7を表示しています
一般論文
原著論文
  • 新井 美桜, 金子 正弘, 小町 守
    原稿種別: 原著論文
    2020 年 35 巻 5 号 p. A-K23_1-9
    発行日: 2020/09/01
    公開日: 2020/09/01
    ジャーナル フリー

    Example sentences retrieval systems display examples that match the input query. They help language learner’swriting. Learners can use the pattern of the examples in their writing. If the query word is correct, learners can findexamples in a native corpus. However, if the query word is incorrect, it is impossible to find appropriate examplesusing ordinary search engines. Existing example retrieval systems do not include grammatically incorrect examples,or only present a few examples, if any. Even if a retrieval system has a wide coverage of incorrect examples alongwith the correct counterparts, learners need to know whether their query includes errors. Considering the usability ofretrieving incorrect examples, our proposed method uses a large-scale learner corpus and presents correct expressionsalong with incorrect expressions using a grammatical error detection system so that the learner does not need to beaware of how to search for examples. Learners can recognize the wrong part in the input query and know how torevise the wrong part when they check correct expressions along with incorrect expressions. Evaluations indicate thatour method improves the accuracy of example sentence retrieval and the quality of a learner’s writing. To the bestof our knowledge, our system is the first incorrect example sentence retrieval system using neural grammatical errordetection.

  • 林 勝悟, 谷本 啓, 鹿島 久嗣
    原稿種別: 原著論文
    2020 年 35 巻 5 号 p. B-K33_1-9
    発行日: 2020/09/01
    公開日: 2020/09/01
    ジャーナル フリー

    The recent rapid and significant increase of big data in our society has led to major impacts of machine learningand data mining technologies in various fields ranging from marketing to science. On the other hand, there still existareas where only small-sized data are available for various reasons, for example, high data acquisition costs or therarity of targets events. Machine learning tasks using such small data are usually difficult because of the lack ofinformation available for training accurate prediction models. In particular, for long-term time-series prediction, thedata size tends to be small because of the unavailability of the data between input and output times in training. Suchlimitations on the size of time-series data further make long-term prediction tasks quite difficult; in addition, thedifficulty that the far future is more uncertain than the near future.

    In this paper, we propose a novel method for long-term prediction of small time-series data designed in theframework of generalized distillation. The key idea of the proposed method is to utilize the middle-time data betweenthe input and output times as “privileged information,” which is available only in the training phase and not in thetest phase. We demonstrate the effectiveness of the proposed method on both synthetic data and real-world data. Theexperimental results show the proposed method performs well, particularly when the task is difficult and has highinput dimensions.

  • 機能・振舞い・構造の観点に基づく問題解決過程のモデル
    古池 謙人, 東本 崇仁, 堀口 知也, 平嶋 宗
    原稿種別: 原著論文
    2020 年 35 巻 5 号 p. C-J82_1-17
    発行日: 2020/09/01
    公開日: 2020/09/01
    ジャーナル フリー

    Recently, the demand for programming education is increasing worldwide. Enhancing intelligent tutoringsystems (ITSs) in programming education is therefore very important. For a computer to intelligently support suchlearning, it is desirable that it be adaptive to individual learning. In ITS research, learning effectiveness is enhancedby (A) controlling features of the question or problem to be asked by indexing based on characteristics of targetdomains, or by (B) making appropriate interventions such as feedback by grasping problem-solving processes basedon explainable problem-solving models.

    It is important to reuse knowledge acquired through problem-solving in programming. To reuse knowledge, itis effective to first understand differences between knowledge items and then to organize that knowledge. In programming,requirements become a problem to be solved. Requirements are defined separately in the software engineeringfield as functional requirements and non-functional requirements. Functional requirements are requirements for whatis satisfied, while non-functional requirements are characteristics for satisfying the functional requirements such asinterface or security. The purpose of this study is to organize the knowledge related to this process by regarding theachievement of functional requirements as problem-solving in programming.

    Assuming that problem-solving is directed toward acquisition of knowledge required for a solution, descriptionsof the programming knowledge itself lead to indexing of the problem. Some studies have utilized function–behavior–structure aspects, combining each aspect to handle knowledge in parts and using them for knowledge descriptions.We have considered that the problem-solving process in this programming can be explained according tothe definition of function–behavior–structure aspects. Therefore, we proposed a model of parts based on function–behavior–structure aspects. And, we further proposed a model of the problem-solving process of parts.

    In order to verify the effectiveness of feedback by the proposed models, an evaluation experiment was performedin comparison with the feedback by our previous system. Feedback by the proposed models is that can begenerated based on “parts management” function and “grasp behavior of structure” function of the ITS functions thatcan be realized by the proposed model.

    Experiment results are suggested that the proposed models can provide more appropriate feedback that can berealized in the system, suggesting that effective support can be realized through learning of parts under the proposedmodels.

    In this research, by defining programming knowledge as parts, we approach various elements related to programmingthat have previously been considered tacit and clarify and organize each element independently of theprogramming language used. In this way, we try to construct a model of the problem-solving process using partsfrom the viewpoint of learning and formalize tacit knowledge.

  • 井上 昂治, 原 康平, ララ ディベッシュ, 中村 静, 高梨 克也, 河原 達也
    原稿種別: 原著論文
    2020 年 35 巻 5 号 p. D-K43_1-10
    発行日: 2020/09/01
    公開日: 2020/09/01
    ジャーナル フリー

    A spoken dialogue system that plays the role of an interviewer for job interviews is presented. In this work, ourgoal is to implement an automated job interview system where candidates can use it as practice before the real interview.Conventional job interview systems ask only pre-defined questions, which make the dialogue monotonous andfar from human-human interviews. We propose follow-up question generation based on the assessment of candidateresponses and keyword extraction. This model was integrated into the dialogue system of the autonomous androidERICA to conduct subject experiments. The proposed job interview system was compared with the baseline systemthat did not generate any follow-up questions and selected among pre-defined questions. The experimental resultsshow that the proposed system is significantly better in subjective evaluations regarding impressions of job interviewpractice, the quality of questions, and the presence of the interviewer.

  • 林 勝悟, 河原 吉伸, 鹿島 久嗣
    原稿種別: 原著論文
    2020 年 35 巻 5 号 p. E-JA10_1-10
    発行日: 2020/09/01
    公開日: 2020/09/01
    ジャーナル フリー

    Change-point detection is a problem to find change in data. Basically, it assumes that the data is passivelygiven. When the cost of data acquisition is not ignorable, it is desirable to save resources by actively selecting effectivedata for change-point detection. In this paper, we introduce Active Change-Point Detection (ACPD), a novel activelearning problem for efficient change-point detection in situations where the cost of data acquisition is expensive.At each round of ACPD, the task is to adaptively determine the next input, in order to detect the change-point ina black-box expensive-to-evaluate function, with as few evaluations as possible. We propose a novel frameworkthat can be generalized for different types of data and change-points, by utilizing an existing change-point detectionmethod to compute change scores and a Bayesian optimization method to determine the next input. We demonstratethe efficiency of our proposed framework in different settings of datasets and change-points, using synthetic data andreal-world data, such as material science data and seafloor depth data.

  • 福馬 智生, 鳥海 不二夫
    原稿種別: 原著論文
    2020 年 35 巻 5 号 p. F-JC3_1-9
    発行日: 2020/09/01
    公開日: 2020/09/01
    ジャーナル フリー

    Latent factor models such as Matrix Factorization have become the default choice for recommender systems due to their performance and scalability. However, such algorithms have two disadvantages. First, these models suffer from data sparsity. Second, they fail to account for model uncertainty. In this paper, we exploit a meta learning strategy to address these problems. The key idea behind our method is to learn predictive distributions conditioned on context sets of arbitrary size of user/item interaction information. Our proposed framework has the advantages of being easy to implement and applicable to any existing latent factor models, providing uncertainty capabilities. We demonstrate the significant superior performance of our model over previous state-of-the-art methods, especially for sparse data in the top-N recommendation task.

  • コミュニケーション場のメカニズムにおける制約活用による人工知能活用
    中林 一貴, 益井 博史, 谷口 忠大
    原稿種別: 原著論文
    2020 年 35 巻 5 号 p. G-K31_1-10
    発行日: 2020/09/01
    公開日: 2020/09/01
    ジャーナル フリー

    Communication-field mechanism design is to design a mechanism including rules and incentives to indirectly control a group of people having communication, e.g., discussion, debate, meeting, and consultation. A communication-field mechanism is expected to give constraints to the actual communications. We hypothesized that such constraints are beneficial for the application of technologies based on artificial intelligence. In this paper, we evaluate this concept by taking an automatic speech recognition system and dealing rights to speak (DRS) as an example of proof of concept. An experiment shows that the simple introduction of DRS improves the performance of speech recognition.

feedback
Top