Search

迷惑メール対策には，メール内容から迷惑メールを判定する手法と送信者情報を用いる手法があげられる．送信者情報の送信元IPアドレスや送信者のドメイン名から受け取るべきメールかを判断できれば，判定のための処理負荷の高いメール内容によるメールフィルタの処理を軽減させることができる．本論文では，送信ドメイン認証技術を利用することで転送メールの送信元を特定し，メール転送元が受け取るべき送信元であることを示し，メール転送元を含めた正規メール送信元を収集することで許可リストを構築する手法を提案する．この手法を含めた，送信者レピュテーションの構築手法を提案し，実際に受信したメールの記録を利用して送信者レピュテーションを構築し，適用することで送信者レピュテーションの構築手法の有効性を示す．, Anti-spam measures include methods for determining unsolicited mail from the mail content and methods for using sender information. If it can be determined from the sender's IP address of sender information and the sender's domain name whether the email should be received, it is possible to reduce the processing of the email filter by the email content that has a high processing load for the determination. In this research, we identify senders of forwarded emails by using sender domain authentication technology, show that the senders should receive the emails, and collect the legitimate email senders including mail forwarders. We propose a method to build a permission list in. We propose a sender reputation construction method that includes this method, construct the sender reputation using the records of the actually received emails, and show the effectiveness of the sender reputation construction method.

近年，教科全体や科目内全体で俯瞰的な学習が求められている．特に教科・科目間の関連を図った横断的な学習が必要とされていること，多肢選択式問題が大量かつ広範囲の出題に向いており多分野の出題に対応する形式であることから，学習者と出題者の双方にとって俯瞰的な多肢選択式問題が有用であると考えられる．“俯瞰的な問題”とは，幅広い関連情報を含み全体像をとらえさせるような内容である．一方，俯瞰的な問題を人手で生成・収集することはコストがかかる．そこで本研究では，出題時に幅広い関連情報を提示することで俯瞰的な視点で問題をとらえさせるような，多肢選択式問題の自動生成手法を提案する．本手法では，多肢選択式問題を構成する問題と選択肢に対してそれぞれ要件を設定し，Linked Dataを利用した自動問題生成アプローチを考案する．Linked Dataとは，構造化されたデータどうしをリンクさせることができるグラフデータである．出題形式として，単一の正解が存在する単数回答形式，複数の正解が存在する複数回答形式，複数の問題に対して単一の正解の組合せが存在する組合せ回答形式の3種類の生成アルゴリズムを提案した．生成問題に対する評価では，被験者として教職免許状所持者と学生に出題したうえで，要件に対応して設定した評価項目を満たすことを確かめた．, In recent years, just about all subjects require students to learn in a broad perspective. Because the need exists for cross-curriculum learning aimed at relating subject areas, it is useful for multiple choice questions to include panoramic information for learners. Panoramic information means comprehensive information that gives us macro-perspective; through which us look down at the whole learning subjects. A question including panoramic information refers to content that includes transverse related information and makes respondents grasp the whole knowledge. However, it is costly to manually generate and collect appropriate multiple-choice questions for learners and exam preparers. Therefore, in this research, we propose a method for automatic generation of multiple choice questions including panoramic information using Linked Data. Linked Data is graphical data that can link structured data, and it is used as a technology for data integration and utilization. In this paper, we aim to realize a system for automatically generating three types of multiple choice questions by implementing an approach to generating questions and choices. An evaluation method for the generation of questions and choices involves setting indicators for each evaluation item, such as validity and the degree of the inclusion of panoramic information.

In recent years, drug trafficking using microblogs has risen and become a social problem. A common method of cyber patrols for cracking down on crimes, such as drug trafficking, involves searching for crime-related keywords. However, criminals who post crime-inducing messages make maximum use of "codewords" rather than keywords, such as enjo kosai, marijuana, and methamphetamine, to camouflage their criminal intentions. Research suggests that these codewords change once they become popular; therefore, searching for a specific word requires significant effort to keep track of the latest codewords. In this study, we focused on the appearance of codewords and those likely to be included in incriminating posts with aim to detect codewords with the high likelihood of inclusion in incriminating posts. We proposed new methods for detecting codewords based on differences in word usage and conducted experiments on concealed-word detection in order to evaluate method effectiveness. The results showed that the proposed method was capable of detecting concealed words other than those in the initial list and to better degree relative to baseline methods. These findings demonstrated the ability of the proposed method to rapidly and automatically detect codewords that change over time and blog posts that induce crimes, thereby potentially reducing the burden of continuous monitoring of codewords.

Recently, fake news is shared via social networks and makes wrong rumors more diffusible. This problem is serious because the wrong rumor sometimes make social damage by deceived people. Fact-checking is a solution to measure the credibility of news articles. However the process usually takes a long time and it is hard to make it before their diffusion. Automatic detection of fake news is a popular researching topic. It is confirmed that considering not only articles but also social contexts(i.e. likes, retweets, replies, comments) supports to spot fake news correctly. However, the social contexts are naturally unavailable when an article comes out, making early fake news detection by means of the social context useless. We propose a fake news detector with the ability to generate fake social contexts, aiming to detect fake news in the early stage of its diffusion where few social contexts are available. The fake context generation is based on a fake news generator model. This model is trained to generate comments using a dataset which consists of news articles and their social contexts. In addition, we also trained a classify model. This used news articles, real-posted comments, and generated comments. To measure our detector's effectiveness, we examined the performance of the generated comments for articles with real comments and generated ones by the classifying model. As a result, we conclude that considering a generated comment help detect more fake news than considering real comments only. It suggests that our proposed detector will be effective to spot fake news on social networks.

To date, existing research has attempted to extract user opinions relating to products and services by differentiating the products and services through their characteristics. However, it is difficult to obtain the characteristics of similar products and services in these studies. So-called competitive products and services are actually quite similar, though they might vary in their strengths and weaknesses across competitors. These advantages and disadvantages are essential for users when selecting products and services. On the other hand, in recent years, researchers have interpreted the output of machine learning models so that humans can understand the reason of the output. LIME and SP-LIME are typical approaches used in the literature. In this paper, we propose a new method using LIME and SP-LIME to compare the characteristics of the services of three competing hotels and investigate appropriate parameters for our method. We try to extract descriptive words for each competitor from review texts, taking a Japanese business hotel market and global game console market as examples.

人や企業は様々な条件下で最適な行動を取るのだろうか．取らないのであればそれはなぜか．その原因を求めることは，実際の個人・企業等の理解を大きく助ける．また，ゲーム理論はスポーツや経済学そしてその他の社会科学の理解に大きく関わってきた．本研究は比較的データが集めやすく混合戦略を適用できるサッカーのPK戦に注目し，独自の確率を考慮した利得表を作成した．その利得表を用いてPK戦におけるキッカーの最適戦略を求め，最適戦略と実際の戦略とのズレを明らかにした．そのズレの原因を求める為にデータセット内の各データ項目についての確率分布を比較するというアプローチをした．データはインターネット動画サイトより収集した，プロ選手による2001年〜2017年の間の世界各国のPK戦150試合（計1539人分）を使用した．実験結果として，最適戦略と実際の戦略との間にズレが存在することが分かった．またそのズレには国籍・スコア差の関与が示唆された．その結果から，サッカーPK戦における最適戦略と実際の戦略との間におけるズレの原因を推定した．本手法はスポーツ分野以外への応用も期待できる．

リーマンショック以降，欧州を中心にインターバンクネットワーク上の金融機関のシステミックリスクに対して関心が高まっている．本研究では，金融機関の破綻の影響がインターバンクネットワークにおいてどのように伝播するかについてエージェントベースシミュレーションを用いて分析した．シミュレーションでは，(1)市場性資産価格の下落率と(2)有価証券に与える損害率を変化させる2種類のシナリオを用意し，金融機関の合併の有効性・破綻リスクのモデル化を試みた．その結果，(1)金融機関の合併を実施した場合においても，多数行にわたる破綻のリスクは削減できない，(2)短期間で有価証券に損害が発生した場合，長期間にわたって損害が発生する状況よりもシステミックリスクが大きくなることが明らかになった．, The aim of the present study is to evaluate systemic risks by manipulating the declining rate of marketable asset price and damaging of securities due to merger of financial institutions. An agent-based simulation platform with purchase method of international accounting standard is developed and analyse the influence of the goodwill (Noren), produced by merging financial institutions. The research revealed the following two points: (1) merger of financial institutions cannot prevent the failures of many banks; (2) systemic risk is more increasing damaging of securities than declining rate of marketable asset price.