Information Engineering Express

Characteristics of Datasets for Fake News Detection to Mitigate Domain Bias

Linshuo Yang — Tue, 06 Feb 2024 16:13:23 +0000

“Fake news”, news intentionally containing false information, has become quite common and often causes social disruption. Many researches on automatic detection of fake have been extensively studied. The classification accuracy is improving, but a major challenge for practical application still remains: models can not work well for news in unknown fields, called “domains”, due to bias caused by different words and phrases among domains. To improve the accuracy of cross-domain fake news detection, it is crucial to mitigate the domain bias since unknown news articles to be classified can be in unknown domains. As a preliminary experiment, we trained a classifier using news articles whose noun phrases were masked because they are considered as a major source of the bias. However, contrary to expectations, masking did not improve accuracy. From the preliminary experiment, we obtained the hypothesis that pairs of fake and real news on the same topic can mitigate the domain bias. Using comparative experiments, we show that accuracy is higher when trained on paired news articles than when trained on unpaired ones.This result strongly suggests that a fake news dataset consisting of paired news could be effective for cross-domain detection.

A Method to Reduce the Burden of the Recreation Moderator by Using a Humanoid Communication Robot

Atsushi Shimoda, Shiro Itai, Toshimitsu Hamada, Maki Matsumoto — Mon, 25 Mar 2024 09:50:20 +0000

The authors proposed a method to reduce the burden on the recreation moderator by using a humanoid communication robot. This system projects quizzes on the screen, and humanoid communication robots read the explanations to proceed with the recreation. This paper presents the details of the proposed system, the method of implementing recreation using the system, the effect of recreation on participants, and the effect of reducing the burden on nursing care staff. An experiment was conducted comparing cases in which nursing care staff acted as moderators and robots acted as moderators. As a result, it became clear that the burden on nursing care staff could be reduced while recreation remained active.

Automatic URL Signature Construction and Impact Assessment

Shota Fujii, Nobutaka Kawaguchi, Tomoya Suzuki, Toshihiro Yamauchi — Mon, 20 May 2024 06:19:44 +0000

In the more recent cyberattacks and malware, the servers of the attacker (e.g., C2 servers) play an important role. It is important to use network-based signatures to block malicious communications to reduce the impact. However, the signatures must not block harmless communications during normal business operations. Therefore, signature generation requires a high level of understanding of the business, and highly depends on individual skills. It is necessary to test and ensure that the generated signatures do not interfere with benign communications, which results in high operational costs. We propose SIGMA, a system that automatically generates signatures to block malicious communication without interfering with benign communication and then automatically evaluates the impact of the signatures. SIGMA automatically extracts the common parts of malware communication destinations by clustering them and generating multiple candidate signatures. Thereafter, it automatically calculates the impact on normal communication based on business logs, etc., and presents the final signature that has the highest blockability of malicious communication and non-blockability of normal communication to the analyst. We aim to reduce the human factor in generating the signatures, reduce the cost of the impact evaluation, and support the decision of whether to apply the signatures.

In our evaluation, we showed that SIGMA can automatically generate a set of signatures that detect 100% of suspicious URLs with an over-detection rate of just 0.87%, based on the results of 14,238 malware analyses and actual business logs. This result suggests that the cost of generating signatures and evaluating their impact on business operations can be reduced; these are time-consuming and human-intensive processes.

Analysis of International Interests and Emotional Responses to the Coronavirus Pandemic

Satoshi Fukuda, Hidetsugu Nanba, Hiroko Shoji — Mon, 20 May 2024 06:23:43 +0000

The outbreak of coronavirus disease 2019 (COVID-19) in December 2019 is still exerting a global impact in 2022, with various media outlets reporting related news items on a daily basis. We analyzed the interest in, and emotional reactions to, COVID-19 of people around the world, as expressed on Twitter. As a measure of interest, we examined replies to news tweets posted by four news outlets (Yahoo! News, The Wall Street Journal, The Guardian, and The Times of India), and classified the emotional content of each reply tweet using Plutchik’s wheel of emotion. The analysis suggested that negative sentiment prevailed worldwide between January 2020 and May 2022; fear-related tweets were significantly more common from January to February 2020 than in the other months in all news reports. However, anticipation-related tweets were more common than those in all other emotion categories in October–November 2021 in Japan. We also analyzed the factors that contributed to the rise of a particular emotion by tracing the news to which tweets with the emotion replied. Our approach that used the news and reply tweets was useful in approximating the factors of the emotional reactions of people in different countries to COVID-19.

Improving the Consistency of Dialog Models Through Speaker Separation Learning

Sakuei Onishi, Takamune Onishi, Hiromitsu Shiina — Wed, 03 Apr 2024 12:09:51 +0000

In recent years, dialog systems, a type of application in the field of natural language processing, have become more prevalent in our daily lives, such as through help desk services. In dialog response generation, responses generated for a specific context may differ from those for other contexts not only grammatically but also semantically in some cases. Thus, simply applying translation technologies would cause issues with the diversity of the generated responses. Previous studies, such as VHRED and GVT, used sampled latent variables for response generation to achieve response diversity. In this study, we propose a method (extended GVTSC) for classifying dialogs before reflecting them in internal dialog processing, in addition to the characteristics of each speaker, to improve diversity while maintaining consistency.

Some Antecedents of Employee Engagement of Japanese Companies

Thu, 15 Feb 2024 06:34:26 +0000

Corporate human resource managers and institutional investors place a high value on employee engagement (EE) in Japanese companies and demand proactive disclosure of this information as human capital information (HCI). Currently, this demand for disclosure has been growing stronger in Europe and the United States. Therefore, Western HR consulting firms are providing services to Japanese companies with suggestions for improving EE. It is unclear whether this service is suitable because of the differences in employment systems between Western countries and Japan. However, EE should be appropriately enhanced in Japanese companies. This paper uses an employee survey of Japanese companies to identify EE antecedents. A multiple-indicator model was created by analyzing the covariance structure of the survey results. Four factors were extracted from this analysis. Two antecedents, “Empowerment” and “Loyalty,” were identified as constituting “Engagement.” This result is generally consistent with the “job engagement” results of Saks’ study. This consistency indicates that our research results are adequate. In the future, we will also investigate organizational engagement, which is missing in our findings, and identify more EE antecedents.

Interactive Evolutionary Computation Creating Congruent Media Content Composed of Different Media Types

Makoto Fukumoto, Taichi Miyamoto, Haoran Gan — Mon, 25 Mar 2024 11:26:18 +0000

We use multiple media content every day, and using congruent media content composed of different media types is ideal for users. However, it is still difficult to obtain congruent media content. Interactive Evolutionary Computation (IEC) is a well-known method for obtaining good media content suited to each user’s feelings as solutions to search problems. Conventional IECs were used for searching sole media type. This study proposes a new IEC that searches the congruent media content as a good combination of different types of media content. In the proposed IEC, the solution candidate contains variables corresponding to different media types. A system was constructed with a genetic algorithm, and it was used to investigate the efficiencies of the pro-posed IEC in the experiment. The target of creation was a relaxing set of music melody and scent. Twenty participants evaluated sets of music melodies and scents throughout ten generations in the search experiment. The experimental results showed a significant increase in the mean fitness and a significant decrease in the distance between solutions. No significant increase was observed in the maximum fitness values.