Catálogo de publicaciones - libros

We combined the mutual information score and TF × IDF score (IR score) in order to select the best keyword translation in our transitive translation. The transitive translation used bilingual dictionaries to translate Indonesian query into Japanese keywords. The Japanese keywords are then used as the input to retrieve Japanese documents. The keyword selection is done in two steps. The first step is to sort translation candidates according to their mutual information scores calculated from a monolingual target language corpus. The second step is to select the best candidate set among 5 top mutual information scores based on their TF × IDF scores. The experiment against NTCIR-3 Web Retrieval Task data shows that the keyword selection based on this combination achieved higher IR score than a direct translation method using original Indonesian-Japanese dictionary and also higher than the machine translation result using Kataku (Indonesian-English) and Babelfish (English-Japanese) engines.

The research goal of this paper is to develop an advanced product search agent framework where personalized agents can meet consumer’s information needs more effectively and accurately based on the Web Services, Semantic Web technologies and AI techniques. These days, one of the major bottlenecks in E-commerce is that it is not easy for consumers to find the relevant information about the products they want. Such a situation is caused mainly by inaccurate representation of consumer’s search intent, and absence of appropriate product information filtering and retrieval mechanism. To resolve these problems, we developed an ontology-based personalized product search query representation methodology, an information extracting methodology specialized for semantic web-based product information, and a multi-attribute-based product scoring methodology. Furthermore, we implemented the proposed methodologies as a prototype system and validated its performance by connecting our system to the well-known Amazon.com and Buy.com.

This paper proposes an automatic code classification for Korean census data by using information retrieval technique and memoory-based learning technique. The purpose of the proposed system is to convert natural language responses on survey questionnaires into corresponding numeric codes according to standard code book from the Census Bureau. The system was trained by memory based learning and experimented with 46,762 industry records and occupation 36,286 records. It was evaluated by using 10-fold cross-validation method. As experimental results, the proposed system showed 99.10% and 92.88% production rates for level 2 and level 5 codes respectively.

The automatic detection of novelty, or newness, as part of an information retrieval system would greatly improve a searcher’s experience by presenting “documents” in order of how much extra information they add to what is already known instead of how similar they are to a user’s query. In this paper we present a novelty detection system evaluated on the AQUAINT text collection as part of our TREC 2004 Novelty Track experiments. Subsequent to participation in TREC, the algorithm has been evaluated on another collection with its parameters optimized and we present those results here. We also discuss how we are extending the text-only approach to novelty detection to also include input from video analysis.

Our approach to solve the problem of Korean named entity classification adopted a co-training method called DL-CoTrain. We use only a part-of-speech tagger and a simple noun phrase chunker instead of a full parser to extract the contextual features of a named entity. We will discuss the linguistic features in Korean which are valuable for named entity classification and experimentally show how large a labeled corpus and which unlabeled corpus is necessary for the better performance and portability of a named entity classifier. With only about a quarter of the labeled corpus, our method can compete with its supervised counterpart.

Arabic numerals show a high occurrence-frequency and deliver significant senses, especially in scientific or informative texts. The problem, how to convert Arabic numerals to phonemes with ambiguous classifiers in Korean, is not easily resolved. In this paper, the ambiguities of Arabic numerals combined with homographic classifiers are analyzed and the resolutions for their sense disambiguation based on KorLex (ean ico-Semantic Network) are proposed. Words proceeding or following the Arabic Numerals are categorized into 54 semantic classes based on the lexical hierarchy in KorLex 1.0. The semantic classes are trained to classify the meaning and the reading of Arabic Numerals using a decision tree. The proposed model shows 87.3% accuracy which is 14.1% higher than the baseline.

Interest point matching is widely used for image indexing. In this paper we introduce a new distance measure between two local descriptors instead of conventional Mahalanobis distance to improve matching accuracy. From experiments with synthetic images we show that the error distribution of local jet is gaussian but the distribution of the descriptors derived from local jet is not gaussian. Based on the observation, we design a new distance measure between two local descriptors and improve accuracy of point matching. We also reduce the number of candidate points and reduce the computational cost by taking into account the characteristic scale ratio. Experimental results confirm the validity of our method.

In recent years, the employment of user feedback information to improve the image retrieval precision has become a hot subject in the research field. But in traditional relevance feedback methods, both relevant and irrelevant user assigned information was required for the retrieval system. For the sake of practicality and convenience, the present paper advances that users only need to choose their inquired image files, which generate a new index vector as relevant information. Through the feature vector space transformation, the index is moved towards the user’s inquiry intention. Meanwhile, the analysis of the user’s inquiry intention together with relevant forecast of index target in the database make it possible for the less similar vectors to get closer to the demanding vectors and thus increasing index precision. In this paper, a prototype system is introduced of image database and experimental illustration to 51138 image files. Compared with the traditional relevance feedback technique, the suggested method is shown to obviously improve the retrieval function.

Content-based image retrieval is a focused problem in current multimedia domain. To obtain better searching results more efficiently in some applications, a proper indexing structure is indispensable. This paper reviews the typical indexing structures in content-based image retrieval at first. Then based on the comparison of their different performance, the paper uncovers the problems in those structures and points out the development direction to improve the performance of CBIR in the future.

Document similarity search is to find documents similar to a query document in a text corpus and return a ranked list of documents to users, which is widely used in recommender systems in library or web applications. The popular approach to similarity search is to calculate the similarities between the query document and documents in the corpus and then rank the documents. In this paper, we investigate the use of document summarization techniques to improve the effectiveness of document similarity search. In the proposed summary-based approach, the query document is summarized and similarity searches are performed with the new query of the produced summary instead of the original document. Different retrieval models and different summarization methods are investigated in the experiments. Experimental results demonstrate the higher effectiveness of the summary-based similarity search.