Domain-specific Contents Research Articles

Strategies of Building AI Agents for Multimodal Productivity with Contemporary Large Language Models

The past few years have witnessed significant advancements in generative artificial intelligence (AI) led by large language models (LLMs), applications demonstrating capabilities in traditionally unattainable tasks. Numerous efforts are being initiated exploring a prospect all the more exciting, to employ LLMs not just as language processors, but as a starting point toward AI agents that can adapt to diverse tasks and complex scenarios. In this paper, a survey is offered on the state-of-the-art strategies of deploying such models for the generation of both text-based domain-specific contents and multimodal outputs embodied in interactions with web applications, industrial software and ultimately the physical world. Two approaches to implementing multimodality are delineated: Direct embedding of multimodal data, and conversion of multimodal data to text. Both types have seen extensive use in areas of research focus, such as image processing, embodied action and software automation. Representative cases from these categories are reviewed with a focus on their input/output modalities, methods of processing multimodal data and output quality.

Health-Related Quality of Life and Its Related Factors in Survivors of Stroke in Rural China: A Large-Scale Cross-Sectional Study.

BackgroundStroke is a major health threat and the leading cause of mortality and disability in China. The aims of this study were to identify the possible influencing factors of health-related quality of life (HRQoL) and its domain-specific contents in stroke patients in rural areas in China.MethodsA total of 1,709 stroke patients aged 36–79 years from the baseline data of Henan Rural Cohort study (n = 39,259) were included in the cross-sectional study. The Chinese version of the European Quality of Life Five Dimension (including mobility, self-care, usual activities, pain/discomfort, and anxiety/depression) Five Level Scale (EQ-5D-5L) and visual analog scale (VAS) were used to evaluate HRQoL in stroke patients. Tobit regression models, generalized linear models and binary logistic regression models were constructed to determine potential influencing factors of the EQ-5D utility index, as well as influencing factors of each domain and VAS score.ResultsThe mean utility index and VAS scores of stroke patients were 0.885 (SD, 0.204), and 68.39 (SD, 17.31), respectively. Pain/discomfort (PD, 35.2%) and mobility (MO, 30.4%) were the most frequently reported issues. Regression models revealed that illiterate; a low monthly income; low physical activity intensity; and diabetes, anxiety, depression, or poor sleep quality were significantly associated with lower utility index and VAS scores among stroke patients. In addition, patients with stroke who were older, female, drinking, smoking, and consuming a high-fat diet, had a higher BMI, and lived with a stroke for a longer time, were also significantly associated with different dimensions of the EQ-5D.ConclusionPatients with stroke in rural areas in China had a low HRQoL. Factors associated with the EQ-5D utility index as well as each domain and VAS score, need to be considered by health providers in rural areas. Patients with stroke in rural areas need to be included in national basic public medical services and managed systematically by medical institutions.

Integrated Service Composition Approach Based on Transparent Access to Heterogeneous IoT Networks Using Multiple Service Providers

The Internet of Things (IoT) enables the number of connected devices to be increased rapidly based on heterogeneous technologies such as platforms, frameworks, libraries, protocols, and standard specifications. Based on the connected devices, various applications can be developed by integrating domain-specific contents using the service composition for providing improved services. The management of the information including devices, contents, and composite objects is necessary to represent the physical objects on the Internet for accessing the IoT services transparently. In this paper, we propose an integrated service composition approach based on multiple service providers to provide improved IoT services by combining various service objects in heterogeneous IoT networks. In the proposed IoT architecture, each service provider provides web services based on Representational State Transfer (REST) Application Programming Interface (API) that delivers information to the clients as well as other providers for integrating the information to provide new services. Through the REST APIs, the integration management provider combines the service result of the IoT service provider to other contents to provide improved services. Moreover, the interworking proxy is proposed to bridge heterogeneous IoT networks for enabling transparent access in the integrated services through proving protocol translating on the entry of the device networks. Therefore, the interworking proxy is deployed between the IoT service provider and device networks to enable clients to access heterogeneous IoT devices through the composited services transparently.

Emotional Connotations of Musical Instrument Timbre in Comparison With Emotional Speech Prosody: Evidence From Acoustics and Event-Related Potentials.

Music and speech both communicate emotional meanings in addition to their domain-specific contents. But it is not clear whether and how the two kinds of emotional meanings are linked. The present study is focused on exploring the emotional connotations of musical timbre of isolated instrument sounds through the perspective of emotional speech prosody. The stimuli were isolated instrument sounds and emotional speech prosody categorized by listeners into anger, happiness and sadness, respectively. We first analyzed the timbral features of the stimuli, which showed that relations between the three emotions were relatively consistent in those features for speech and music. The results further echo the size-code hypothesis in which different sound timbre indicates different body size projections. Then we conducted an ERP experiment using a priming paradigm with isolated instrument sounds as primes and emotional speech prosody as targets. The results showed that emotionally incongruent instrument-speech pairs triggered a larger N400 response than emotionally congruent pairs. Taken together, this is the first study to provide evidence that the timbre of simple and isolated musical instrument sounds can convey emotion in a way similar to emotional speech prosody.

Symptoms and beyond: Self-concept among sexually assaulted women.

The unique characteristics of sexual assault (SA)-a toxic mix of an interpersonal harm, a violent exploitation of one's body, and a transformation of an act of connectedness into an act of submission-are postulated to negatively affect the self-concept. We sought to deepen the understanding of self-concept impairments among sexually assaulted women with varying levels of posttraumatic distress. To this end, we compared women with a main trauma of SA to women with a main trauma of motor-vehicle accident (MVA) and to nontraumatized (NT) women on several self-concept aspects. Our main hypotheses were (a) sexually assaulted women without PTSD exhibit impaired self-concept as compared with NT women and (b) SA is related to greater self-concept impairments as compared with MVA, even when posttraumatic distress is statistically controlled. Women (N = 235: NT = 69, MVA = 87, SA = 79) completed a web-based survey including measures designed to assess the global and domain-specific contents and structure of the self-concept as well as background and clinical questionnaires. Sexually assaulted women without PTSD reported impaired self-concept as compared with NT women. Furthermore, SA was related to greater self-concept impairments as compared with MVA, even when considering participants' levels of posttraumatic distress. SA is related to unique self-concept impairments that extend beyond symptoms, emphasizing the need to assess and address self-concept impairments in sexually assaulted women. The importance of adopting a multifaceted conceptualization of the self to gain a deeper understanding of the aftermath of trauma is highlighted. (PsycINFO Database Record

Discovering Hidden Tasks and Process Structure through Email Logs for DSM

The design structure matrix (DSM) method is a powerful network modelling tool to highlight the system architecture via presenting the interactions between the elements comprising a system. In DSM, measuring the interaction strength between elements is the first and crucial step. However, since the system elements are generally pre-identified by experts, their interaction relationships are often obtained from experts’ evaluation, which is time consuming and error-prone. To address this critical issue, in this paper, we aim to discover the hidden tasks of a process and their interactions from email logs with domain-specific contents. To this purpose, we have proposed a three-stage numerical DSM modeling approach, where text mining and clustering techniques are studied to discover hidden tasks. Furthermore, a metric concerning the measurement of interaction strength based on the overlapping time and the volume of exchanged information is presented. A case study is provided based on the emails collected from a design project.

Dynamic Reputation Rating Mechanism for Social Content Curation Services

Recently various social curation mechanisms have been developed to organize and suggest digital contents around one or more particular themes or topics for online users on Social Network Services (SNS). Collaborative filtering method can be used to improve efficiency of automated social curation systems, and so we have already applied this method to enhance credibility of curators in previous research, but these approaches have problem in extracting user preferences for users who have not evaluated many contents. In this study, we use dynamic curator groups which are automatically formed to recommend and organize domain specific contents. The group members have dynamic reputation value depending on their evaluation performance. Social curations over online digital contents are very effective to find relevant information in a specific domain. In addition, we show simulation results to evaluate the reliability enhancement of the proposed dynamic curator model for automated curation services of social content.

Term extraction from sparse, ungrammatical domain-specific documents

Existing term extraction systems have predominantly targeted large and well-written document collections, which provide reliable statistical and linguistic evidence to support term extraction. In this article, we address the term extraction challenges posed by sparse, ungrammatical texts with domain-specific contents, such as customer complaint emails and engineers’ repair notes. To this aim, we present ExtTerm, a novel term extraction system. Specifically, as our core innovations, we accurately detect rare (low frequency) terms, overcoming the issue of data sparsity. These rare terms may denote critical events, but they are often missed by extant TE systems. ExtTerm also precisely detects multi-word terms of arbitrarily lengths, e.g. with more than 2 words. This is achieved by exploiting fundamental theoretical notions underlying term formation, and by developing a technique to compute the collocation strength between any number of words. Thus, we address the limitation of existing TE systems, which are primarily designed to identify terms with 2 words. Furthermore, we show that open-domain (general) resources, such as Wikipedia, can be exploited to support domain-specific term extraction. Thus, they can be used to compensate for the unavailability of domain-specific knowledge resources. Our experimental evaluations reveal that ExtTerm outperforms a state-of-the-art baseline in extracting terms from a domain-specific, sparse and ungrammatical real-life text collection.

User-author centered multimedia building blocks

The advances of multimedia models and tools popularized the access and production of multimedia contents: in this new scenario, there is no longer a clear distinction between authors and end-users of a production. These user-authors often work in a collaborative way. As end-users, they collectively participate in interactive environments, consuming multimedia artifacts. In their authors' role, instead of starting from scratch, they often reuse others' productions, which can be decomposed, fusioned and transformed to meet their goals. Since the need for sharing and adapting productions is felt by many communities, there has been a proliferation of standards and mechanisms to exchange complex digital objects, for distinct application domains. However, these initiatives have created another level of complexity, since people have to define which share/ reuse solution they want to adopt, and may even have to resort to programming tasks. They also lack effective strategies to combine these reused artifacts. This paper presents a solution to this demand, based on a user-author centered multimedia building block model--the digital content component (DCC). DCCs upgrade the notion of digital objects to digital components, as they homogenously wrap any kind of digital content (e.g., multimedia artifacts, software) inside a single component abstraction. The model is fully supported by a software infrastructure, which exploits the model's semantic power to automate low level technical activities, thereby freeing user-authors to concentrate on creative tasks. Model and infrastructure improve recent research initiatives to standardize the means of sharing and reuse domain specific digital contents. The paper's contributions are illustrated using examples implemented in a DCC-based authoring tool, in real life situations.