Domain-specific Knowledge Base Research Articles

AI-Enhanced Social Work: Developing and Evaluating Retrieval-Augmented Generation (RAG) Support Systems

ABSTRACT This article introduces the application of retrieval-augmented generation (RAG) technology, aiming to enhance decision making and client support through artificial intelligence (AI)–enhanced tools. RAG addresses the critical challenge of ensuring accuracy and reliability in AI-generated content by integrating generative AI with domain-specific knowledge bases. Through a case study, we explore the development and assessment of a RAG-powered tool designed to improve faculty advising, highlighting the importance of collaboration between social workers and technologists. The adaptability of RAG to various social work challenges is discussed alongside considerations for the maintenance of the knowledge base. Our work underscores the potential of RAG to revolutionize social work practices by providing accurate and reliable information to serve individuals and communities better.

An Application Programming Interface (API) Sensitive Data Identification Method Based on the Federated Large Language Model

The traditional methods for identifying sensitive data in APIs mainly encompass rule-based and machine learning-based approaches. However, these methods suffer from inadequacies in terms of security and robustness, exhibit high false positive rates, and struggle to cope with evolving threat landscapes. This paper proposes a method for detecting sensitive data in APIs based on the Federated Large Language Model (FedAPILLM). This method applies the large language model Qwen2.5 and the LoRA instruction tuning technique within the framework of federated learning (FL) to the field of data security. Under the premise of protecting data privacy, a domain-specific corpus and knowledge base are constructed for pre-training and fine-tuning, resulting in a large language model specifically designed for identifying sensitive data in APIs. This paper conducts comparative experiments involving Llama3 8B, Llama3.1 8B, and Qwen2.5 14B. The results demonstrate that Qwen2.5 14B can achieve similar or better performance levels compared to the Llama3.1 8B model with fewer training iterations.

Extract Clinical Lab Tests From Electronic Hospital Records Through Featured Transformer Model

Natural language, as a rich source of information, has been used as the foundation of the product review, the demographic trend, and the domain specific knowledge bases. To extract entities from texts, the challenge is, free text is so sparse that missing features always exist which makes the training processing incomplete. Based on attention mechanism in deep learning architecture, the authors propose a featured transformer model (FTM) which adds category information into inputs to overcome missing feature issue. When attention mechanism performs Markov-like updates in deep learning architecture, the importance of the category represents the frequency connecting to other entities and categories and is compatible with the importance of the entity in decision-making. They evaluate the performance of FTM and compare the performance with several other machine learning models. FTM overcomes the missing feature issue and performs better than other models.

A Rule-Based Approach for Mining Creative Thinking Patterns from Big Educational Data

Numerous studies have established a correlation between creativity and intrinsic motivation to learn, with creativity defined as the process of generating original and valuable ideas, often by integrating perspectives from different fields. The field of educational technology has shown a growing interest in leveraging technology to promote creativity in the classroom, with several studies demonstrating the positive impact of creativity on learning outcomes. However, mining creative thinking patterns from educational data remains a challenging task, even with the proliferation of research on adaptive technology for education. This paper presents an initial effort towards formalizing educational knowledge by developing a domain-specific Knowledge Base that identifies key concepts, facts, and assumptions essential for identifying creativity patterns. Our proposed pipeline involves modeling raw educational data, such as assessments and class activities, as a graph to facilitate the contextualization of knowledge. We then leverage a rule-based approach to enable the mining of creative thinking patterns from the contextualized data and knowledge graph. To validate our approach, we evaluate it on real-world datasets and demonstrate how the proposed pipeline can enable instructors to gain insights into students’ creative thinking patterns from their activities and assessment tasks.

Improving the Domain Adaptation of Retrieval Augmented Generation (RAG) Models for Open Domain Question Answering

Abstract Retrieval Augment Generation (RAG) is a recent advancement in Open-Domain Question Answering (ODQA). RAG has only been trained and explored with a Wikipedia-based external knowledge base and is not optimized for use in other specialized domains such as healthcare and news. In this paper, we evaluate the impact of joint training of the retriever and generator components of RAG for the task of domain adaptation in ODQA. We propose RAG-end2end, an extension to RAG that can adapt to a domain-specific knowledge base by updating all components of the external knowledge base during training. In addition, we introduce an auxiliary training signal to inject more domain-specific knowledge. This auxiliary signal forces RAG-end2end to reconstruct a given sentence by accessing the relevant information from the external knowledge base. Our novel contribution is that, unlike RAG, RAG-end2end does joint training of the retriever and generator for the end QA task and domain adaptation. We evaluate our approach with datasets from three domains: COVID-19, News, and Conversations, and achieve significant performance improvements compared to the original RAG model. Our work has been open-sourced through the HuggingFace Transformers library, attesting to our work’s credibility and technical consistency.

Towards Real-Time Warning and Defense Strategy AI Planning for Cyber Security Systems Aided by Security Ontology

Cyber security systems generally have the phenomena of passive defense and low-efficiency early warnings. Aiming at the above problems, this study proposes a real-time warning and plans an AI defense strategy for a cyber security system aided by a security ontology. First, we design a security defense ontology that integrates attack graphs, general purpose and domain-specific knowledge bases, and on this basis, we (1) develop an ontology-driven method of early warnings of real-time attacks, which supports non-intrusive scanning attack detection and (2) combine artificial intelligence planning and bounded rationality to recommend and automatically execute defense strategies in conventional defense scenarios. A case study has been performed, and the results indicate that: (1) the proposed method can quickly analyze network traffic data for real-time warnings, (2) the proposed method is highly feasible and has the ability to implement defense strategies autonomously, and (3) the proposed method performs the best, with a 5.4–11.4% increase in defense effectiveness against the state-of-the-art counterparts considering the APT29 attack. Overall, the proposed method holds the potential to increase the defense effectiveness against cyberattacks under high computing resource constraints.

Natural Language Processing-Assisted Literature Retrieval and Analysis for Combination Therapy in Cancer.

Despite advances in molecular therapeutics, few anticancer agents achieve durable responses. Rational combinations using two or more anticancer drugs have the potential to achieve a synergistic effect and overcome drug resistance, enhancing antitumor efficacy. A publicly accessible biomedical literature search engine dedicated to this domain will facilitate knowledge discovery and reduce manual search and review. We developed RetriLite, an information retrieval and extraction framework that leverages natural language processing and domain-specific knowledgebase to computationally identify highly relevant papers and extract key information. The modular architecture enables RetriLite to benefit from synergizing information retrieval and natural language processing techniques while remaining flexible to customization. We customized the application and created an informatics pipeline that strategically identifies papers that describe efficacy of using combination therapies in clinical or preclinical studies. In a small pilot study, RetriLite achieved an F1 score of 0.93. A more extensive validation experiment was conducted to determine agents that have enhanced antitumor efficacy in vitro or in vivo with poly (ADP-ribose) polymerase inhibitors: 95.9% of the papers determined to be relevant by our application were true positive and the application's feature of distinguishing a clinical paper from a preclinical paper achieved an accuracy of 97.6%. Interobserver assessment was conducted, which resulted in a 100% concordance. The data derived from the informatics pipeline have also been made accessible to the public via a dedicated online search engine with an intuitive user interface. RetriLite is a framework that can be applied to establish domain-specific information retrieval and extraction systems. The extensive and high-quality metadata tags along with keyword highlighting facilitate information seekers to more effectively and efficiently discover knowledge in the combination therapy domain.

Towards Knowledge-Based Tourism Chinese Question Answering System

With the rapid development of the tourism industry, various travel websites are emerging. The tourism question answering system explores a large amount of information from these travel websites to answer tourism questions, which is critical for providing a competitive travel experience. In this paper, we propose a framework that automatically constructs a tourism knowledge graph from a series of travel websites with regard to tourist attractions in Zhejiang province, China. Backed by this domain-specific knowledge base, we developed a tourism question answering system that also incorporates the underlying knowledge from a large-scale language model such as BERT. Experiments on real-world datasets demonstrate that the proposed method outperforms the baseline on various metrics. We also show the effectiveness of each of the question answering components in detail, including the query intent recognition and the answer generation.

Implementation and evaluation of a multivariate abstraction-based, interval-based dynamic time-warping method as a similarity measure for longitudinal medical records

We extended dynamic time warping (DTW) into interval-based dynamic time warping (iDTW), including (A) interval-based representation (iRep): [1] abstracting raw, time-stamped data into interval-based abstractions, [2] comparison-period scoping, [3] partitioning abstract intervals into a given temporal granularity; (B) interval-based matching (iMatch): matching partitioned, abstract-concepts records, using a modified DTW. Using domain knowledge, we abstracted the raw data of medical records, for up to three concepts out of four or five relevant concepts, into two interval types: State abstractions (e.g. LOW, HIGH) and Gradient abstractions (e.g. INCREASING, DECREASING). We created all uni-dimensional (State or Gradient) or multi-dimensional (State and Gradient) abstraction combinations. Tasks: Classifying 161 oncology patients records as autologous or allogenic bone-marrow transplantation; classifying 125 hepatitis patients records as B or C hepatitis; predicting micro- or macro-albuminuria in the next year for 151 Type 2 diabetes patients. We used a k-Nearest-Neighbors majority, k = an odd number from 1 to SQRT(N), N = set size. 75,936 10-fold cross-validation experiments were performed: 33,600 (Oncology), 28,800 (Hepatitis), 13,536 (Diabetes). Measures: Area Under the Curve (AUC), optimal Youden's Index. Paired t-tests compared result vectors for equivalent configurations other than a tested variable, to determine a significant mean accuracy difference (P<0.05). Mean classification and prediction using abstractions was significantly better than using only raw time-stamped data. In each domain, at least one abstraction combination led to a significantly better mean performance than raw data. Increasing feature number and using Multi-dimensional abstractions enhanced performance. Unlike when using raw data, optimal mean performance was often reached with k=5, using abstractions.

Quality Evaluation of Structural Design in Software Reverse Engineering: A Focus On Cohesion

Software reverse engineering (SRE) plays a crucial role in contemporary software environments. Software developers may implement a system first then use SRE tools to generate design content such as the Unified Modeling Language (UML) diagrams. In the literature of SRE, studies majorly focus on how precisely the conversion can reflect the system; there is, however, little or no research that further looks into the quality of the converted results. Therefore, this paper presents an online knowledge-based ontological SRE system, OntRECoh, for quality evaluation of converted UML structural models. OntRECoh features a domain-specific knowledge base that focuses on cohesion design and a rule-based inference engine for computing the cohesion scores of Java-based implemented systems and providing improvement recommendations through its Web-based interface. Furthermore, OntRECoh includes both static and dynamic cohesion measures from both the design and the implementation aspects, for the evaluation to be more comprehensive and synthetic in the SRE context.

Real-Time Tweet Analytics Using Hybrid Hashtags on Twitter Big Data Streams

Twitter is a microblogging platform that generates large volumes of data with high velocity. This daily generation of unbounded and continuous data leads to Big Data streams that often require real-time distributed and fully automated processing. Hashtags, hyperlinked words in tweets, are widely used for tweet topic classification, retrieval, and clustering. Hashtags are used widely for analyzing tweet sentiments where emotions can be classified without contexts. However, regardless of the wide usage of hashtags, general tweet topic classification using hashtags is challenging due to its evolving nature, lack of context, slang, abbreviations, and non-standardized expression by users. Most existing approaches, which utilize hashtags for tweet topic classification, focus on extracting hashtag concepts from external lexicon resources to derive semantics. However, due to the rapid evolution and non-standardized expression of hashtags, the majority of these lexicon resources either suffer from the lack of hashtag words in their knowledge bases or use multiple resources at once to derive semantics, which make them unscalable. Along with scalable and automated techniques for tweet topic classification using hashtags, there is also a requirement for real-time analytics approaches to handle huge and dynamic flows of textual streams generated by Twitter. To address these problems, this paper first presents a novel semi-automated technique that derives semantically relevant hashtags using a domain-specific knowledge base of topic concepts and combines them with the existing tweet-based-hashtags to produce Hybrid Hashtags. Further, to deal with the speed and volume of Big Data streams of tweets, we present an online approach that updates the preprocessing and learning model incrementally in a real-time streaming environment using the distributed framework, Apache Storm. Finally, to fully exploit the batch and stream environment performance advantages, we propose a comprehensive framework (Hybrid Hashtag-based Tweet topic classification (HHTC) framework) that combines batch and online mechanisms in the most effective way. Extensive experimental evaluations on a large volume of Twitter data show that the batch and online mechanisms, along with their combination in the proposed framework, are scalable, efficient, and provide effective tweet topic classification using hashtags.

EMSContExt: EMS Protocol-Driven Concept Extraction for Cognitive Assistance in Emergency Response

This paper presents a technique for automated curation of a domain-specific knowledge base or lexicon for resource-constrained domains, such as Emergency Medical Services (EMS) and its application to real-time concept extraction and cognitive assistance in emergency response. The EMS responders often verbalize critical information describing the situations at an incident scene, including patients' physical condition and medical history. Automated extraction of EMS protocol-specific concepts from responders' speech data can facilitate cognitive support through the selection and execution of the proper EMS protocols for patient treatment. Although this task is similar to the traditional NLP task of concept extraction, the underlying application domain poses major challenges, including low training resources availability (e.g., no existing EMS ontology, lexicon, or annotated EMS corpus) and domain mismatch. Hence, we develop EMSContExt, a weakly-supervised concept extraction approach for EMS concepts. It utilizes different knowledge bases and a semantic concept model based on a corpus of over 9400 EMS narratives for lexicon expansion. The expanded EMS lexicon is then used to automatically extract critical EMS protocol-specific concepts from real-time EMS speech narratives. Our experimental results show that EMSContExt achieves 0.85 recall and 0.82 F1-score for EMS concept extraction and significantly outperforms MetaMap, a state-of-the-art medical concept extraction tool. We also demonstrate the application of EMSContExt to EMS protocol selection and execution and real-time recommendation of protocol-specific interventions to the EMS responders. Here, EMSContExt outperforms MetaMap with a 6% increase and six times speedup in weighted recall and execution time, respectively.

Ontology Learning Approaches to Provide Domain-Specific Knowledge Base

Abstract Ontology learning techniques are based on extensive basic knowledge, from unstructured data to structured data. Creating an ontology manually is undoubtedly a time consuming process. In addition, manually-curated background knowledge is a scarce resource for many domains. Ontology learning is a relatively new research area that draws from related fields. However, a number of approaches are still rising up and, what is more, they vary and evolve across different aspects. Therefore, the purpose of this article is to provide comprehensive knowledge about selected groups of methods, tools and approaches supporting ontology. Presenting different approaches to learning ontology and including them in one publicly available summary, aims to provide an overview of the most appropriate methods, tools and approaches to building ontologies from various sources. Besides, the proposed solution in this article will expand the already offered ontology of learning methods with additional, more detailed functions. In addition to the domain ontology offered, this is another goal of this article. The correctness of the proposed ontology was verified by the constructed competence questions and implemented using Description Logic query mechanism.

Survey on entity linking for domain specific with heterogeneous information networks

Entity linking is a task of extracting information that links the mentioned entity in a collection of text with their similar knowledge base as well as it is the task of allocating unique identity to various entities such as locations, individuals and companies. Knowledgebase (KB) is used to optimize the information collection, organization and for retrieval of information. Heterogeneous information networks (HIN) comprises multiple-type interlinked objects with various types of relationship which are becoming increasingly most popular named bibliographic networks, social media networks as well including the typical relational database data. In HIN, there are various data objects are interconnected through various relations. The entity linkage determines the corresponding entities from unstructured web text, in the existing HIN. This work is the most important and it is the most challenge because of ambiguity and existing limited knowledge. Some HIN could be considered as a domain-specific KB. The current Entity Linking (EL) systems aimed towards corpora which contain heterogeneous as web information and it performs sub-optimally on the domain-specific corpora. The EL systems used one or more general or specific domains of linking such as DBpedia, Wikipedia, Freebase, IMDB, YAGO, Wordnet and MKB. This paper presents a survey on domain-specific entity linking with HIN. This survey describes with a deep understanding of HIN, which includes datasets,types and examples with related concepts.

Software Process Ontology: A case study of software organisations software process sub domains

In a domain like software process that is intensively knowledge driven, transforming intellectual knowledge by formal representation is an invaluable requirement. An improved use of this knowledge could lead to maximum payoff in software organisations which is key. The purpose of formal representation is to help organisations achieve success by modelling successful organisations. In this paper, Software process knowledge from successful organisations was harvested and formally modeled using ontology. Domain specific knowledge base ontology was produced for core software process subdomain, with its resulting software process ontology produced.

A Framework for Domain-Specific Natural Language Information Brokerage

Service providers — from public institutions to primary care facilities — need to constantly attend to clients’ inquiries to provide useful information and directive guidelines. Ensuring high quality service is challenging as it not only demands detailed domain-specific knowledge, but also the ability to quickly understand the clients’ issues through their diverse — and often casual — descriptions. This paper aims to provide a framework for the development of an automated information broker agent who performs the task of a helper. The main task of the agent is to interact with the client and direct them to obtain further services that cater their personalized need. To do so, the agent should accomplish a sequence of tasks that include natural language inquiry, knowledge gathering, reasoning, and giving feedback; in this way, it simulates a human helper to engage in interaction with the client. The framework combines a question-answering reasoning mechanism while utilizing domain-specific knowledge base. When the users cannot describe clearly their needs, the system tries to narrow down the possibilities by an iterative question-answering process, until it eventually identifies the target. In realizing our framework, we make a proof-of-concept project, Mandy, a primary care chatbot system created to assist healthcare staffs by automating the patient intake process. We describe in detail the system functionalities and design of the system, and evaluate our proof-of-concept on benchmark case studies.

Expert problem solving practice in commercial property valuation: an exploratory study

PurposeThe purpose of this paper is to identify the core dimensions of problem solving of experts in commercial valuation in order to provide a rich stimulus for managing current practice and enabling future development.Design/methodology/approachThe study adopted a cognitive position but emphasised understanding the everyday commercial property valuation practice in a naturalistic context and from the participants’ perspectives. Given this, a grounded theory approach was employed as a research strategy to guide the data collection and surface theoretical interpretations. Data were obtained through in-depth interviews with practicing valuers working in private real estate firms within metropolitan Birmingham, UK.FindingsThe interviews uncover four dimensions of experts’ problem-solving practice in commercial valuation: multidimensional, domain-specific knowledge base; cognitive process that is centred on analysis and reflection; collaborative problem-solving venture with colleagues; and professional practice issues awareness. A conceptual model is proposed which integrates these dimensions enabling a clearer understanding of the nature of valuation work.Research limitations/implicationsThis study was designed to be descriptive and theory generating, thus, the findings cannot be generalised as the sample was confined to one city and consists of a small number of senior practicing valuers. Therefore, the findings may not be fully applicable to other practicing valuers, other geographical locations or more widely to other types of property valuation. Nevertheless, the findings provide an important cognitive framework which can be verified by other researchers seeking to examine the practice of expert valuers.Practical implicationsThe identification of the core dimensions of expert problem solving in commercial property valuation is shown to have implications for valuation practice, education and continued research. The valuation practice environments need to develop mechanisms to provide time that would enable these multi-dimensions of professional competence to be developed. Further work is needed to expand and refine the model across expert practice in other specialty areas of valuation practice.Originality/valueThis study expands the current understanding of valuation process to areas of expertise that have received less coverage in behavioural valuation literature, that is, the central role of knowledge and cognition and how these are integrated for effective valuation problem solving and decision making.

A Study of Context Influences in Arabic-English Language Translation Technologies

Social and cultural context is largely missing in current language translation systems. Dictionary based systems translate terms in a source language to an equivalent term in a target language, but often the translation could be inaccurate when context is not taken into consideration, or when an equivalent term in the target language does not exist. Domain knowledge and context can be made explicit by using ontologies, and ontology utilization would enable inclusion of semantic relations to other terms, leading to translation results which is more comprehensive than a single equivalent term. It is proposed that existing ontologies in the domain should be utilized and combined by ontology merging techniques, to leverage on existing resources to form a basis ontology with contextual representation, and this can be further enhanced by using machine translation techniques on existing corpora to improve the basic ontology to append further contextual information to the knowledge base. Statistical methods in machine translation could provide automated relevance determination of these existing resources which are machine readable, and aid the human translator in establishing a domain specific knowledge base for translation. Advancements in communication and technologies has made the world smaller where people of different regions and languages need to work together and interact. Machine translation provides an automatic approach to make translation facilities available to anyone, at any time. The accuracy of these translations are crucial as it could lead to misunderstandings and possible conflict. While single equivalent terms in a target language can provide a gist of the meaning of a source language term, a semantic conceptualisation provided by an ontology could enable the term to be understood in the specific context that it is being used.

A Two-Stage Joint Model for Domain-Specific Entity Detection and Linking Leveraging an Unlabeled Corpus

The intensive construction of domain-specific knowledge bases (DSKB) has posed an urgent demand for researches about domain-specific entity detection and linking (DSEDL). Joint models are usually adopted in DSEDL tasks, but data imbalance and high computational complexity exist in these models. Besides, traditional feature representation methods are insufficient for domain-specific tasks, due to problems such as lack of labeled data, link sparseness in DSKBs, and so on. In this paper, a two-stage joint (TSJ) model is proposed to solve the data imbalance problem by discriminatively processing entity mentions with different degrees of ambiguity. In addition, three novel methods are put forward to generate effective features by incorporating an unlabeled corpus. One crucial feature involving entity detection is the mention type, extracted by a long short-term memory (LSTM) model trained on automatically annotated data. The other two types of features mainly involve entity linking, including the inner-document topical coherence, which is measured based on entity co-occurring relationships in the corpus, and the cross-document entity coherence evaluated using similar documents. An overall 74.26% F1 value is obtained on a dataset of real-world movie comments, demonstrating the effectiveness of the proposed approach and indicating its potentiality to be used in real-world domain-specific applications.

Linked Geoscience Data in practice: where W3C standards meet domain knowledge, data visualization and OGC standards

The geoscience community is now facing both the challenge and the opportunity caused by the vast amount of datasets that can be made available on the Web. An efficient “data environment” on the Web has the potential to enable geoscientists to conduct their research in ways that never existed before. Standards developed by the Open Geospatial Consortium have already been used widely to build data services among different subjects in geosciences. In recent years, the Linked Open Data approach initiated by the World Wide Web Consortium has received increasing attention. In this paper, the author presents a pilot study that uses a domain specific knowledge base and data visualization techniques to leverage the functionality of geoscience data services in the Web of Data. The study focuses on the topic of the geologic time scale. Detailed works such as semantic modeling and encoding, multilingual vocabularies, exploratory data visualization, web map service and processing, and the query of linked data are introduced through real-world datasets. This study faces a broad perspective of the Linked Geoscience Data and leverages the functionalities of existing standards into a new level for geoscience applications.