• The KIPS Transactions:PartD
• /
• v.12D no.5 s.101
• /
• pp.789-796
• /
• 2005

Because existing information retrieval systems, in particular library retrieval systems, use 'exact keyword matching' with user's query, they present user with massive results including irrelevant information. So, a user spends extra effort and time to get the relevant information from the results. Thus, this paper will propose SULRM a Retrieval Model using Subject Classification Table, User profile, and LSI(Latent Semantic Indexing), to provide more relevant results. SULRM uses document filtering technique for classified data and document ranking technique for non-classified data in the results of keyword-based retrieval. Filtering technique uses Subject Classification Table, and ranking technique uses user profile and LSI. And, we have performed experiments on the performance of filtering technique, user profile updating method, and document ranking technique using the results of information retrieval system of our university' digital library system. In case that many documents are retrieved proposed techniques are able to provide user with filtered data and ranked data according to user's subject and preference.

• Journal of Information Processing Systems
• /
• v.13 no.5
• /
• pp.1277-1298
• /
• 2017

Related to the maximum vector problem, a skyline query is to discover dominating tuples from a set of tuples, where each defines an object (such as a hotel) in several dimensions (such as the price and the distance to the beach). A tuple, an instance of an object, dominates another tuple if it is equally good or better in all dimensions and better in at least one dimension. Traditionally, skyline queries are defined upon single-instance data or upon objects each of which is associated with an instance. However, in some cases, an object is not associated with a single instance but rather by multiple instances. For example, on a review website, many users assign scores to a product or a service, and a user's score is an instance of the object representing the product or the service. Such data is an example of multi-instance data. Unlike most (if not all) others considering the traditional setting, we consider skyline queries defined upon multi-instance data. We define the dominance calculation and propose an algorithm to reduce its computational cost. We use synthetic and real data to evaluate the proposed methods, and the results demonstrate their utility.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.14 no.6
• /
• pp.737-744
• /
• 2004

Retrieving relevant information well corresponding to the user`s request from web is a crucial task of search engines. However, most of conventional search engines based on pattern matching schemes to queries have a limitation that is not easy to provide results corresponding to the user`s request due to the uncertainty of queries. To overcome the limitation in this paper, we propose a framework for knowledge-based semantic meta-search engines with the following five processes: (i) Query formation, (ii) Query expansion, (iii) Searching, (iv) Ranking recreation, and (v) Knowledge base. From simulation results on english-based web documents, we can see that the Proposed knowledge-based semantic meta-search engine provides more correct and better searching results than those obtained by using the Google.

• The Transactions of the Korea Information Processing Society
• /
• v.7 no.7
• /
• pp.2108-2115
• /
• 2000

The conventional Boolean retrieval systems based on vector spae model can provide the results of retrieval fast, they can't reflect exactly user's retrieval purpose including semantic information. Consequently, the results of retrieval process are very different from those users expected. This fact forces users to waste much time for finding expected documents among retrieved documents. In his paper, we designed a bayesian SOM(Self-Organizing feature Maps) in combination with bayesian statistical method and Kohonen network as a kind of unsupervised learning, then perform classifying documents depending on the semantic similarity to user query in real time. If it is difficult to observe statistical characteristics as there are less than 30 documents for clustering, the number of documents must be increased to at least 50. Also, to give high rank to the documents which is most similar to user query semantically among generalized classifications for generalized clusters, we find the similarity by means of Kohonen centroid of each document classification and adjust the secondary rank depending on the similarity.

• Journal of KIISE:Software and Applications
• /
• v.31 no.8
• /
• pp.1101-1112
• /
• 2004

In this paper, we propose a Web page weighting scheme using the user preference in each sense of query word to improve the performance of Web search. Generally search engines assign weights to a web page by using relevancy only, which is obtained by comparing the query word and the words in a web page. In the information retrieval from huge data such as the Web, simple word comparison cannot distinguish important documents because there exist too many documents with similar relevancy In this paper we implement a WordNet-based user interface that helps to distinguish different senses of query word, and constructed a search engine in which the implicit evaluations by multiple users are reflected in ranking by accumulating the number of clicks. In accumulating click counts, they are stored separately according to senses, so that more accurate search is possible. The experimental results with several keywords show that the precision of proposed system is improved compared to conventional search engines.

• KIISE Transactions on Computing Practices
• /
• v.21 no.2
• /
• pp.144-147
• /
• 2015

Question Answering (QA) services can provide exact answers to user questions written in natural language form. This research focuses on how to build a QA system for a specific domain area. Online and offline QA system architecture of targeted domain such as domain detection, question analysis, reasoning, information retrieval, filtering, answer extraction, re-ranking, and answer generation, as well as data preparation are presented herein. Test results with an official Frequently Asked Question (FAQ) set showed 68% accuracy of the top 1 and 77% accuracy of the top 5. The contribution of each part such as question analysis system, document search engine, knowledge graph engine and re-ranking module for achieving the final answer are also presented.

• Journal of Information Processing Systems
• /
• v.13 no.5
• /
• pp.1213-1228
• /
• 2017

The pitch tracking of music has been researched for several decades. Several possible improvements are available for creating a good t-distribution, using the instantaneous robust algorithm for pitch tracking framework to perfectly detect pitch. This article shows how to detect the pitch of music utilizing an improved detection method which applies a statistical method; this approach uses a pitch track, or a sequence of frequency bin numbers. This sequence is used to create an index that offers useful features for comparing similar songs. The pitch frequency spectrum is extracted using a modified instantaneous robust algorithm for pitch tracking (IRAPT) as a base combined with the statistical method. The pitch detection algorithm was implemented, and the percentage of performance matching in Thai classical music was assessed in order to test the accuracy of the algorithm. We used the longest common subsequence to compare the similarities in pitch sequence alignments in the music. The experimental results of this research show that the accuracy of retrieval of Thai classical music using the t-distribution of instantaneous robust algorithm for pitch tracking (t-IRAPT) is 99.01%, and is in the top five ranking, with the shortest query sample being five seconds long.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.3
• /
• pp.894-907
• /
• 2022

The features extracted by convolutional neural networks are more descriptive of images than traditional features, and their convolutional layers are more suitable for retrieving images than are fully connected layers. The convolutional layer features will consume considerable time and memory if used directly to match an image. Therefore, this paper proposes a feature weighting and region integration method for convolutional layer features to form global feature vectors and subsequently use them for image matching. First, the 3D feature of the last convolutional layer is extracted, and the convolutional feature is subsequently weighted again to highlight the edge information and position information of the image. Next, we integrate several regional eigenvectors that are processed by sliding windows into a global eigenvector. Finally, the initial ranking of the retrieval is obtained by measuring the similarity of the query image and the test image using the cosine distance, and the final mean Average Precision (mAP) is obtained by using the extended query method for rearrangement. We conduct experiments using the Oxford5k and Paris6k datasets and their extended datasets, Paris106k and Oxford105k. These experimental results indicate that the global feature extracted by the new method can better describe an image.

• The Transactions of the Korea Information Processing Society
• /
• v.13 no.5
• /
• pp.221-226
• /
• 2024

In information retrieval, queries come in various types, ranging from abstract queries to those containing specific keywords, making it a challenging task to accurately produce results according to user demands. Additionally, search systems must handle queries encompassing various elements such as typos, multilingualism, and codes. Reranking is performed through training suitable documents for queries using DeBERTa, a deep learning model that has shown high performance in recent research. To evaluate the effectiveness of the proposed method, experiments were conducted using the test collection of the Product Search Track at the TREC 2023 international information retrieval evaluation competition. In the comparison of NDCG performance measurements regarding the experimental results, the proposed method showed a 10.48% improvement over BM25, a basic information retrieval model, in terms of search through query error handling, provisional relevance feedback-based product title-based query expansion, and reranking according to query types, achieving a score of 0.7810.

• Journal of KIISE:Computing Practices and Letters
• /
• v.16 no.2
• /
• pp.222-226
• /
• 2010

With the continuously increasing volume of e-commerce transactions, it is now popular to buy some products and to evaluate them on the World Wide Web. The product reviews are very useful to customers because they can make better decisions based on the indirect experiences obtainable through these reviews. However, since online shopping malls do not provide ranking results, it is not easy for users to read all the relevant review documents effectively. Product reviews include subjective and emotional opinions. Thus, the review search is different from the general web search in terms of ranking strategy. In this paper, we propose an effective method of ranking the reviews that can reflect user's intention by using opinion mining techniques. The proposed method analyzes product reviews with query words, and sentimental polarity of subjective opinions. Through diverse experiments, we show that our proposed method outperforms conventional ones.