人工知能学会論文誌
Online ISSN : 1346-8030
Print ISSN : 1346-0714
ISSN-L : 1346-0714
32 巻 , 6 号
選択された号の論文の3件中1~3を表示しています
原著論文
  • 田頭 幸浩, 堀田 徹, 田中 祐介, 小野 真吾, 塚本 浩司, 田島 玲
    2017 年 32 巻 6 号 p. A-H52_1-10
    発行日: 2017/11/01
    公開日: 2017/11/01
    ジャーナル フリー

    Contextual advertising is a form of textual advertising usually displayed on third party Web pages. One of the main problems with contextual advertising is determining how to select ads that are relevant to the page content and/or the user information in order to achieve both effective advertising and a positive user experience. In this study, we propose a translation method that learns the mapping of the contextual information to the textual features of ads by using past click data. The contextual information includes the user’s demographic information and behavioral information as well as page content information. The proposed method is able to retrieve more preferable ads while maintaining the sparsity of the inverted index and the performance of the ad retrieval system. In addition, it is easy to implement and there is no need to modify an existing ad retrieval system. Extensive evaluations showed the effectiveness of our approach.

  • Chao Fan, Fujio Toriumi
    2017 年 32 巻 6 号 p. B-H42_1-11
    発行日: 2017/11/01
    公開日: 2017/11/01
    ジャーナル フリー

    Many models synthesize various types of complex networks with communities. However, a network generation model that can represent high-modularity networks is rare. In this paper, we propose a high-modularity network generation model by layer aggregation based on a multilayer network. Because people belong to many communities in society, such as family, school, hobby group, and business organizations, each example is regarded as a community in a single layer of a multilayer network. However, measuring each relationship in each community is difficult. A network on social network services (SNSs) that can be observed combines all communities. That is, a social network is generated from a multilayer network. A synthesized network in our model has either a community structure or a high-modularity structure. We apply the proposed model to generate a number of networks and compare them with real-world networks. Not only did it successfully represent real-world networks but we also found that we can predict how real-world networks are generated from the model’s parameters.

  • 北村 拓也, 長谷 浩也, 前田 一誠, 林 雄介, 平嶋 宗
    2017 年 32 巻 6 号 p. C-H14_1-12
    発行日: 2017/11/01
    公開日: 2017/12/11
    ジャーナル フリー

    We have designed and developed an exercise environment to learn logical thinking through activities to assemble logic structures called “Triangle Logic Model.” This model consists of three elements, (1) data, (2) warrant, and (3) claim, corresponding to the simplified Toulmin Model. In the Triangle Logic Model, each of the three elements is placed on its respective corner of a triangle depending on the type of the element. Data, warrant, and claim are placed on the left bottom, right bottom and top corner of a triangle, respectively. The basic design of the exercise based on the model is to require learners to place provided propositions on the corner of a triangle to give the proper role of data, warrant or claim to each of them. Variations of the exercise can be designed by including a few distractors in the provided propositions and/or by previously placing one or two propositions on the triangle. The current stage of this research restricts the content of the logic structure to syllogism and the activities to selecting and assembling provided propositions. Proposed exercise environment realizes automatic answer verification and automatic feedback provision. As an evaluation of this exercise environment, we conducted an experiment with two groups, that is, (1) an experiment group with a pre-test, the exercise using the environment, and a post-test, and (2) a control group with only the pre-and the post-test. The test is the standard test for a national investigation of logical thinking ability by National Institute for Educational Policy Research in Japan. As a result, the average score of the post-test in the experimental group had a statistically significant increase (the effect size was large), whereas there was no significant difference in the control group. Moreover, in the experimental group, there was a statistically significant high correlation between the average time to complete the exercise and the scores of both the pre- and the post-test. In the exercise, because a learner was required to complete a problem to move to the next problem, the correlation suggests that the exercise requests the learner to use their logical thinking ability. These results suggest that this exercise could improve users' logical thinking ability.

feedback
Top