Search

This paper presents a novel and comprehensive solution to enhance both the robustness and efficiency of question answering (QA) systems through supervised contrastive learning (SCL). Training a high-performance QA system has become straightforward with pre-trained language models, requiring only a small amount of data and simple fine-tuning. However, despite recent advances, existing QA systems still exhibit significant deficiencies in functionality and training efficiency. We address the functionality issue by defining four key tasks: user input intent classification, out-of-domain input detection, new intent discovery, and continual learning. We then leverage a unified SCL-based representation learning method to efficiently build an intra-class compact and inter-class scattered feature space, facilitating both known intent classification and unknown intent detection and discovery. Consequently, with minimal additional tuning on downstream tasks, our approach significantly improves model efficiency and achieves new state-of-the-art performance across all tasks.

Capturing patients’ medical histories significantly influences clinical decisions. Errors in this process lead to clinical errors, which increase costs and dissatisfaction among physicians and patients. Physicians in developing countries are overloaded with patients and cannot always follow the proper history-taking procedure. The challenges have been acknowledged; however, a comprehensive understanding of the status and the remedies has remained unexplored. This paper aims to investigate the workload, history-taking challenges, and the willingness of the physicians to accept digital solutions. A cross-sectional online survey was conducted on 104 physicians across Bangladesh, featuring 22 questions regarding their professional environment, workload, digitization status of health records, challenges in history taking, and attitudes toward adopting digital solutions for managing patient histories; 92.67% of the physicians face high workloads, 88.46% struggle in medical history taking, and only 4.81% use digital medical records. About 70% struggle to complete the necessary history-taking steps, emphasizing the urgent need for solutions. A novel visualization system, the Smart Health Gantt Chart (SHGC), has been introduced for their instant feedback. A total of 93.27% of physicians expressed their willingness to use such a system. The proposed SHGC has the potential to enhance healthcare efficiency in developing nations, benefit physicians, and improve patient-centered care.

The estimation of points of interest (POI) categories is essential in several contexts, such as land use estimation, POI and itinerary recommendation in the tourism sector, and so on. Most of these approaches are based on well-known POIs and use information such as people’s mobility or check-in data. This information is not or rarely available for lesser-known POIs. However, these lesser-known POIs cannot be ignored because of this lack of information, as they may be important to local people in terms of their culture and history and worth discovering by tourists or local authorities. To address this challenge, we propose an approach based on the techniques of coupling the syntactic and semantic analysis of data via a knowledge graph using Few-shot learning (FSL) and Light Graph Convolution Network (LightGCN). The FSL technique allows us to work with very little data, which not only works with lesser-known POIs but also reduces the complexity in terms of tasks and execution time. The results show that our approach outperforms the baseline approaches and that considering the semantic aspect of the data via Linked Open Data (LOD) significantly improves the results of the approach based on the syntactic analysis alone.

The number of helpful votes on a review is an essential indicator of how much impact the review has on other customers in electronic commerce. Therefore, predicting the number of helpful votes is an important task. Regression analysis and Tobit modeling are typical methods of prediction. Those methods come from the same initial assumption that the number of helpful votes follows a normal distribution on any dataset. However, the assumption is not usually confirmed, and the distribution of the helpful votes often follows other distributions. This paper proposes a framework for investigating the feasibility of building a model that predicts the number of helpful votes according to the distribution of the number of helpful votes. On top of that, considering the review age, we propose an adaptive window size sampling method to evaluate the model on review datasets sorted chronologically. The experimental results validated that the model adapting to the best approximate distribution gives a significant improvement compared to the baseline models. In addition, model evaluation using the adaptive window size sampling method has significant impacts on the performance on large datasets.

Learning user attributes is essential for providing users with a service. In particular, for e-commerce portals which deal in variety of goods ranging from clothes to foods to home electronics, it is especially important to learn “domain-independent” attributes such as age, gender, and personality that affect people’s behavior across various domains of daily life (e.g., clothing, eating and housing) because these attributes can be used for personalization in diverse domains their service covers. Thus far, researchers have proposed approaches to learn user representation (UR) from user-item interactions, trying to embed rich information about user attributes in UR. However, very few can learn URs that are domain-independent without confounding them with domain-specific attributes (e.g., food preferences). This could consequently undermine the former’s utility for personalizing services in other domains from which the URs are not learned. To address this, we propose an approach to learn URs that exclusively reflect domain-independent attributes. Our approach introduces a novel multi-layer RNN with two types of layers: Domain Specific Layers (DSLs) for modeling behavior in individual domains and a Domain Independent Layer (DIL) for modeling attributes that affect behavior across multiple domains. By exchanging hidden states between these layers, the RNNs implement the process of domain-independent attributes affecting domain-specific behavior and makes the DIL learn URs that capture domain-independence. Our evaluation results confirmed that the URs learned by our approach have greater utility in predicting behavior in the other domains from which these URs were not learned thereby demonstrating adaptability to various domains.

Government 2.0 activities have become very attractive and popular. Using the platforms to support the activities, anyone can anytime report issues in a city on the Web and share the reports with other people. Since a variety of reports are posted, officials in the city management section have to give priorities to the reports. However, it is not easy task for the officials to judge the importance of the reports because importance judgments vary depending on the officials, and consequently the agreement rate becomes low. To remedy the low agreement rate problem of human judgment, it is necessary to create an intelligent agent which supports finding reports with high priorities. Hirokawa et al. employed the Support Vector Machine (SVM) with a word Feature Selection method (SVM+FS) to detect signs of danger from posted reports because the signs of danger is one of high priority issues to be dealt with. However they did not compare the SVM+FS method with other conventional machine learning methods and it is not clear if the SVM+FS method has better performance than the other methods. This paper explores methods for detecting the signs of danger through comprehensive experiments to develop an intelligent agent which supports officials in the city management sections. We explores conventional machine learning methods: SVM, Random Forest, Naïve Bayse using conventional word vectors, an LDA-based document vector, and word embedding by Word2Vec and compared the best method with SVM+FS. Experimental results illustrate the superiority of SVM+FS and invoke the importance of using multiple data sets when evaluating the methods of detecting signs of danger.