Earticle

초거대 언어모델(LLM)의 활용이 확산되며, 산업 현장에서는 오픈소스 LLM과 RAG 기술을 접목해 비용 효율적인 AI 적용을 모색하고 있다. 본 연구는 관세 분야의 핵심 요소인 HS code 판별에 RAG 기반 LLM 시스템을 적용하고, 사이즈가 다른 모델 간의 성능 차이를 비교·분석하였다. 다양한 품목분류 데이터를 기반으로 구축한 RAG 시스템이 통관 업무의 자동화 및 효율화에 기여할 수 있음을 실험을 통해 확인하였으며, 이는 민간 기업의 관세 실무에 활용 가능한 기초 자료로서의 가치를 가진다.

Retrieval of scholarly articles about a specific research issue is a routine job of researchers to cross-validate the evidence about the issue. Two articles that focus on a research issue should share similar terms in their core contents, including their goals, backgrounds, and conclusions. In this paper, we present a technique CCSE ($\underline{C}ore$ $\underline{C}ontent$ $\underline{S}imilarity$ $\underline{E}stimation$) that, given an article a, recommends those articles that share similar core content terms with a. CCSE works on titles and abstracts of articles, which are publicly available. It estimates and integrates three kinds of similarity: goal similarity, background similarity, and conclusion similarity. Empirical evaluation shows that CCSE performs significantly better than several state-of-the-art techniques in recommending those biomedical articles that are judged (by domain experts) to be the ones whose core contents focus on the same research issues. CCSE works for those articles that present research background followed by main results and discussion, and hence it may be used to support the identification of the closely related evidence already published in these articles, even when only titles and abstracts of the articles are available.

Purpose. The purpose of this paper is to analyze and evaluate discovery layer search tools for retrieval of legal information in Indian law libraries. This paper covers current practices in legal information retrieval with special reference to Indian academic law libraries, and analyses its importance in the domain of law.Design/Methodology/Approach. A web survey and observational study method are used to collect the data. Data related to the discovery tools were collected using email and further discussion held with the discovery layer/ tool /product developers and their representatives.Findings. Results show that most of the Indian law libraries are subscribing to bundles of legal information resources such as Hein Online, JSTOR, LexisNexis Academic, Manupatra, Westlaw India, SCC web, AIR Online (CDROM), and so on. International legal and academic resources are compatible with discovery tools because they support various standards related to online publishing and dissemination such as OAI/PMH, Open URL, MARC21, and Z39.50, but Indian legal resources such as Manupatra, Air, and SCC are not compatible with the discovery layers. The central index is one of the important components in a discovery search interface, and discovery layer services/tools could be useful for Indian law libraries also if they can include multiple legal and academic resources in their central index. But present practices and observations reveal that discovery layers are not providing facility to cover legal information resources. Therefore, in the present form, discovery tools are not very useful; they are an incomplete and half solution for Indian libraries because all available Indian legal resources available in the law libraries are not covered.Originality/Value. Very limited research or published literature is available in the area of discovery layers and their compatibility with legal information resources.

The study examined information retrieval tools as predictors for information resources utilization, four research questions, and four hypotheses were made to guide the study. A descriptive survey was adopted for the study. Random sampling technique was used to select sample of 393 from a population of 557 academic staff registered in the University of Port Harcourt library. The questionnaire was adopted as a data collection instrument titled Information retrieval as predictors for information resources utilization (IRPIRUQ). Data were analyzed using both simple and multiple regression while analysis of variance (ANOVA) associate with regression was used for testing the hypotheses at 0.05 alpha level. The study revealed that information resources are under utilized as the OPAC and Online Databases are not easily accessed. Further findings showed that the academic staff made use of internet search engines more often than the OPAC and online databases. It was recommended among others that a new library software be installed in place of KOHA for wider connectivity and adequate distribution of software that will aid usage of the online databases and OPAC.

Communication is a process involving a group of interrelated elements working together for the purpose of information transfer. This paper discusses information retrieval as a communication process in the 21st century library. The difficulties associated with access to recorded knowledge through bibliographic control devices have been exacerbated by the interposition of additional encoding processes in the library and further decoding by the users. In addition, the innovation of internet/web has revolutionized the means and mode of communication process in the library by flooding information seekers with information and creating an illusion of self-sufficiency in many users. With these changes in information seeking behaviour and pattern, a cybernetic approach to information retrieval has emerged emphasizing adaptive control mechanisms and feedback processes. This paper argues that libraries should strive to continuously remain relevant by keeping abreast with changes in the behavior of information users. To this end, this paper proposes apomediatic-cybernetic model of communication, which illustrates information retrieval processes for the 21st-century library.

Given a query (a health question), retrieval of relevant frequently asked questions (FAQs) is essential as the FAQs provide both reliable and readable information to healthcare consumers. The retrieval requires the estimation of the semantic similarity between the query and each FAQ. The similarity estimation is challenging as semantic structures of Chinese healthcare FAQs are quite different from those of the FAQs in other domains. In this paper, we propose a conceptual model for Chinese healthcare FAQs, and based on the conceptual model, present a technique ECA that estimates conceptual similarities between FAQs. Empirical evaluation shows that ECA can help various kinds of retrievers to rank relevant FAQs significantly higher. We also make ECA online to provide services for FAQ retrievers.

Evaluation of retrieval systems is mostly limited to laboratory settings and rarely considers changes of performance over time. This article presents an evaluation of retrieval systems for internal Web site search systems between the years 2006 and 2011. A holistic evaluation methodology for real Web sites was developed which includes tests for functionality, search quality, and user interaction. Among other sites, one set of 20 Web site search systems was evaluated three times in different years and no substantial improvement could be shown. It is surprising that the communication between site and user still leads to very poor results in many cases. Overall, the quality of these search systems could be improved, and several areas for improvement are apparent from our evaluation. For a comparison, Google’s site search function was also tested with the same tasks.

In this study, we present a Retrieval-Augmented Generation (RAG)-based pipeline designed to extract key values from scientific literature on Silicon Carbide (SiC) crystal growth using the Physical Vapor Transport (PVT) method. To improve the relevance and completeness of the retrieved context, we implemented a hybrid retrieval strategy that combines dense retrieval via FAISS with sparse retrieval using BM25. We employed two distinct prompting approaches for key value extraction. The first approach addresses interactive user queries by utilizing the retrieved context to generate informed responses. The second approach, intended for bulk extraction, follows a two-step process: a binary classification prompt first checks for the presence of relevant information related to a query. If relevant information is confirmed, a subsequent prompt extracts the value under strict constraints-requiring exact phrasing without guessing or explanation. This binary pre-check significantly enhances the identification of true negative cases, thereby reducing irrelevant or missing data. For the generative component of our pipeline, we evaluated three large language models (LLMs): Llama 8B, Gemma 7B, and Mistral 7B, all operating on a local multi-GPU environment using FP16 precision. The results reveal differences in the efficiency of these models within our customized RAG system, particularly in their performance in extracting over 156 targeted technical key-value pairs from 13 benchmark papers.

Despite the rapid strides in content-based image retrieval, a notable disparity persists between the visual features of images and the semantic features discerned by humans. Hence, image retrieval based on the association of semantic similarities recognized by humans with visual similarities is a difficult task for most image-retrieval systems. Our study endeavors to bridge this gap by refining image semantics, aligning them more closely with human perception. Deep learning techniques are used to semantically classify images and retrieve those that are semantically similar to personalized images. Moreover, we introduce a keyword-based image retrieval, enabling automatic labeling of images in mobile environments. The proposed approach can improve the performance of a mobile device with limited resources and bandwidth by performing retrieval based on the visual features and keywords of the image on the mobile device.

The combination texture feature extraction approach for texture image retrieval is proposed in this paper. Two kinds of low level texture features were combined in the approach. One of them was extracted from singular value decomposition (SVD) based dual-tree complex wavelet transform (DTCWT) coefficients, and the other one was extracted from multi-scale local binary patterns (LBPs). The fusion features of SVD based multi-directional wavelet features and multi-scale LBP features have short dimensions of feature vector. The comparing experiments are conducted on Brodatz and Vistex datasets. According to the experimental results, the proposed method has a relatively better performance in aspect of retrieval accuracy and time complexity upon the existing methods.

The paper introduces the research area of interactive information retrieval (IIR) from a historical point of view. Further, the focus here is on evaluation, because much research in IR deals with IR evaluation methodology due to the core research interest in IR performance, system interaction and satisfaction with retrieved information. In order to position IIR evaluation, the Cranfield model and the series of tests that led to the Cranfield model are outlined. Three iconic user-oriented studies and projects that all have contributed to how IIR is perceived and understood today are presented: The MEDLARS test, the Book House fiction retrieval system, and the OKAPI project. On this basis the call for alternative IIR evaluation approaches motivated by the three revolutions (the cognitive, the relevance, and the interactive revolutions) put forward by Robertson & Hancock-Beaulieu (1992) is presented. As a response to this call the 'IIR evaluation model' by Borlund (e.g., 2003a) is introduced. The objective of the IIR evaluation model is to facilitate IIR evaluation as close as possible to actual information searching and IR processes, though still in a relatively controlled evaluation environment, in which the test instrument of a simulated work task situation plays a central part.

We propose a content-based image retrieval system that uses a combination of completed local binary pattern (CLBP) and color autocorrelogram. CLBP features are extracted on a multi-resolution multi-direction filtered domain of value component. Color autocorrelogram features are extracted in two dimensions of hue and saturation components. Experiment results revealed that the proposed method yields a lot of improvement when compared with the methods that use partial features employed in the proposed method. It is also superior to the conventional CLBP, the color autocorrelogram using R, G, and B components, and the multichannel decoded local binary pattern which is one of the latest methods.

Shape description is an important and fundamental issue in content-based image retrieval (CBIR), and a number of shape description methods have been reported in the literature. For shape description, both global information and local contour variations play important roles. In this paper a new included-angular ternary pattern (IATP) based shape descriptor is proposed for shape image retrieval. For each point on the shape contour, IATP is derived from its neighbor points, and IATP has good properties for shape description. IATP is intrinsically invariant to rotation, translation and scaling. To enhance the description capability, multiscale IATP histogram is presented to describe both local and global information of shape. Then multiscale IATP histogram is combined with included-angular histogram for efficient shape retrieval. In the matching stage, cosine distance is used to measure shape features' similarity. Image retrieval experiments are conducted on the standard MPEG-7 shape database and Swedish leaf database. And the shape image retrieval performance of the proposed method is compared with other shape descriptors using the standard evaluation method. The experimental results of shape retrieval indicate that the proposed method reaches higher precision at the same recall value compared with other description method.

In this paper, we present a new texture image retrieval method which combines color and texture features extracted from images by a set of multi-resolution multi-direction (MRMD) filters. The MRMD filter set chosen is simple and can be separable to low and high frequency information, and provides efficient multi-resolution and multi-direction analysis. The color space used is HSV color space separable to hue, saturation, and value components, which are easily analyzed as showing characteristics similar to the human visual system. This experiment is conducted by comparing precision vs. recall of retrieval and feature vector dimensions. Images for experiments include Corel DB and VisTex DB; Corel_MR DB and VisTex_MR DB, which are transformed from the aforementioned two DBs to have multi-resolution images; and Corel_MD DB and VisTex_MD DB, transformed from the two DBs to have multi-direction images. According to the experimental results, the proposed method improves upon the existing methods in aspects of precision and recall of retrieval, and also reduces feature vector dimensions.

In this paper, a texture feature extraction method using local energy and local correlation of Gabor transformed images is proposed and applied to an image retrieval system. The Gabor wavelet is known to be similar to the response of the human visual system. The outputs of the Gabor transformation are robust to variants of object size and illumination. Due to such advantages, it has been actively studied in various fields such as image retrieval, classification, analysis, etc. In this paper, in order to fully exploit the superior aspects of Gabor wavelet, local energy and local correlation features are extracted from Gabor transformed images and then applied to an image retrieval system. Some experiments are conducted to compare the performance of the proposed method with those of the conventional Gabor method and the popular rotation-invariant uniform local binary pattern (RULBP) method in terms of precision vs recall. The Mahalanobis distance is used to measure the similarity between a query image and a database (DB) image. Experimental results for Corel DB and VisTex DB show that the proposed method is superior to the conventional Gabor method. The proposed method also yields precision and recall 6.58% and 3.66% higher on average in Corel DB, respectively, and 4.87% and 3.37% higher on average in VisTex DB, respectively, than the popular RULBP method.

Healthcare professionals have edited many health questions (HQs) and their answers for healthcare consumers on the Internet. The HQs provide both readable and reliable health information, and hence retrieval of those HQs that are relevant to a given question is essential for health education and promotion through the Internet. However, retrieval of relevant HQs needs to be based on the recognition of the intention of each HQ, which is difficult to be done by predefining syntactic and semantic rules. We thus model the intention recognition problem as a text classification problem, and develop two techniques to improve a learning-based text classifier for the problem. The two techniques improve the classifier by location-based and area-based feature weightings, respectively. Experimental results show that, the two techniques can work together to significantly improve a Support Vector Machine classifier in both the recognition of HQ intentions and the retrieval of relevant HQs.

To manipulate large video contents, effective video indexing and retrieval are required. A large number of video indexing and retrieval algorithms have been presented for frame-wise user query or video content query whereas a relatively few video sequence matching algorithms have been proposed for video sequence query. In this paper, we propose an efficient algorithm that extracts key frames using color histograms and matches the video sequences using edge features. To effectively match video sequences with a low computational load, we make use of the key frames extracted by the cumulative measure and the distance between key frames, and compare two sets of key frames using the modified Hausdorff distance. Experimental results with real sequence show that the proposed video sequence matching algorithm using edge features yields the higher accuracy and performance than conventional methods such as histogram difference, Euclidean metric, Battachaya distance, and directed divergence methods.

The big challenge in current content-based image retrieval systems is to reduce the semantic gap between the low level-features and high-level concepts. In this paper, we have proposed a novel framework for efficient image retrieval to improve the retrieval results significantly as a means to addressing this problem. In our proposed method, we first extracted a strong set of image features by using the dual-tree rotated complex wavelet filters (DT-RCWF) and dual tree-complex wavelet transform (DT-CWT) jointly, which obtains features in 12 different directions. Second, we presented a relevance feedback (RF) framework for efficient image retrieval by employing a support vector machine (SVM), which learns the semantic relationship among images using the knowledge, based on the user interaction. Extensive experiments show that there is a significant improvement in retrieval performance with the proposed method using SVMRF compared with the retrieval performance without RF. The proposed method improves retrieval performance from 78.5% to 92.29% on the texture database in terms of retrieval accuracy and from 57.20% to 94.2% on the Corel image database, in terms of precision in a much lower number of iterations.

의미 포토 검색은 포토의 간단한 시각화 특성과 적합한 의미를 분류하는데 있어서의 갭을 간소화시키는 데 중요한 역할을 한다. 의미 검색을 이용한 효과적인 포토 검색은 포토 검색에 있어서 매우 중요한 과제중의 하나이다. 따라서 우리는 사용자 인터페이스의 포토 주석을 이용한 새로운 이벤트 의미 포토 검색 기법을 제안한다. 본 논문에서는 순수 XML 데이터베이스와 MPEG-7표준을 기반으로 포토 관리 및 의미 검색이 쉬운 포토 앨범 관리 시스템을 설계 및 구현하였다.

Semantic photo retrieval has been an important role in reducing the semantic gap between the simple visual features and the abundant semantics delivered by a photo. Effective photo retrieval using semantics is one of the major challenges in photo retrieval. And we propose a new event semantic photo retrieval method by using photo annotation user interface. In this paper, A photo album management system that facilitates photo management and semantic retrieval, which fully relies on the MPEG-7 standard as an information base and a native XML database, has been designed and implemented.

In this paper, we propose new semantic contents modeling using individual features, associative relations and visual features for efficiently supporting browsing and retrieval of video semantic contents. And we implement and design a browsing and retrieval system based on the semantic contents modeling. The browsing system supports annotation based information, keyframe based visual information, associative relations, and text based semantic information using a tree based browsing technique. The retrieval system supports text based retrieval, visual feature and associative relations according to the retrieval types of semantic contents.