Re-ranking System Research Articles

Transformer‐based reranking for improving Korean morphological analysis systems

AbstractThis study introduces a new approach in Korean morphological analysis combining dictionary‐based techniques with Transformer‐based deep learning models. The key innovation is the use of a BERT‐based reranking system, significantly enhancing the accuracy of traditional morphological analysis. The method generates multiple suboptimal paths, then employs BERT models for reranking, leveraging their advanced language comprehension. Results show remarkable performance improvements, with the first‐stage reranking achieving over 20% improvement in error reduction rate compared with existing models. The second stage, using another BERT variant, further increases this improvement to over 30%. This indicates a significant leap in accuracy, validating the effectiveness of merging dictionary‐based analysis with contemporary deep learning. The study suggests future exploration in refined integrations of dictionary and deep learning methods as well as using probabilistic models for enhanced morphological analysis. This hybrid approach sets a new benchmark in the field and offers insights for similar challenges in language processing applications.

Re-Ranking System with BERT for Biomedical Concept Normalization

In recent years, various neural network architectures have been successfully applied to natural language processing (NLP) tasks such as named entity normalization. Named entity normalization is a fundamental task for extracting information in free text, which aims to map entity mentions in a text to gold standard entities in a given domain-specific ontology; however, the normalization task in the biomedical domain is still challenging because of multiple synonyms, various acronyms, and numerous lexical variations. In this study, we regard the task of biomedical entity normalization as a ranking problem and propose an approach to rank normalized concepts. We additionally employ two factors that can notably affect the performance of normalization, such as task-specific pre-training (Task-PT) and calibration approach. Among five different biomedical benchmark corpora, our experimental results show that our proposed model achieved significant improvements over the previous methods and advanced the state-of-the-art performance for biomedical entity normalization, with up to 0.5% increase in accuracy and 1.2% increase in F-score.

HPO-Shuffle: an associated gene prioritization strategy and its application in drug repurposing for the treatment of canine epilepsy.

Epilepsy is a common neurological disorder that affects mammalian species including human beings and dogs. In order to discover novel drugs for the treatment of canine epilepsy, multiomics data were analyzed to identify epilepsy associated genes. In this research, the original ranking of associated genes was obtained by two high-throughput bioinformatics experiments: Genome Wide Association Study (GWAS) and microarray analysis. The association ranking of genes was enhanced by a re-ranking system, HPO-Shuffle, which integrated information from GWAS, microarray analysis and Human Phenotype Ontology database for further downstream analysis. After applying HPO-Shuffle, the association ranking of epilepsy genes were improved. Afterward, a weighted gene coexpression network analysis (WGCNA) led to a set of gene modules, which hinted a clear relevance between the high ranked genes and the target disease. Finally, WGCNA and connectivity map (CMap) analysis were performed on the integrated dataset to discover a potential drug list, in which a well-known anticonvulsant phensuximide was included.

Enhancing Sketch-Based Image Retrieval by CNN Semantic Re-ranking.

This paper introduces a convolutional neural network (CNN) semantic re-ranking system to enhance the performance of sketch-based image retrieval (SBIR). Distinguished from the existing approaches, the proposed system can leverage category information brought by CNNs to support effective similarity measurement between the images. To achieve effective classification of query sketches and high-quality initial retrieval results, one CNN model is trained for classification of sketches, another for that of natural images. Through training dual CNN models, the semantic information of both the sketches and natural images is captured by deep learning. In order to measure the category similarity between images, a category similarity measurement method is proposed. Category information is then used for re-ranking. Re-ranking operation first infers the retrieval category of the query sketch and then uses the category similarity measurement to measure the category similarity between the query sketch and each initial retrieval result. Finally, the initial retrieval results are re-ranked. The experiments on different types of SBIR datasets demonstrate the effectiveness of the proposed re-ranking method. Comparisons with other re-ranking algorithms are also given to show the proposed method's superiority. Further, compared to the baseline systems, the proposed re-ranking approach achieves significantly higher precision in the top ten different SBIR methods and datasets.

Relevance and Diversity of Images by Using Tags

Now a day, there is a growth of digital images and video archives. Some of these are very important from user point of view. The collection of these digital images may be for personal use and may be for public use. For searching images in the database, there is the need of secure, fast and efficient techniques. Tag-based image search is an effective technique to find images in the databases and in the social websites. In this paper a reranking system for image retrieval with the consideration of images relevance and diversity is proposed. Reranking of the images have done according to their visual features and semantic features. Visual features are used for detecting duplicate images in proposed method. Each user maintains several images. First sort these images by same group user ranking system. Users that have higher contribution to the given query rank higher. Then an implementation to all-user reranking on the ranked users image set and only the most relevant image from users image set is selected. These selected images compose the final retrieval results. We build an inverted structure for the image dataset to speed up the searching process.

A Novel Indexing and Image Annotation Structure for Efficient Image Retrieval

The present-day commercial web-based search engines have adopted web-based image search to improve accuracy in image data retrieval. Though re-ranking is conventionally considered as an effective process for determining the status of web-based picture search engines, it suffers from a few deficiencies. Hence, some classification techniques, especially novel image re-ranking system, have been proposed to implement query image re-ranking with semantic signatures in web-based image data retrieval, which automatically retrieves results based on visual semantic features for different query or keyword expansions. In essence, search-based image index annotation is a process that effectively achieves more number of matched similar images with weak label data presentation to found noise and unsatisfied picture indexing and formation. In order to overcome this limitation in index-based image retrieval, it seeks to define effective label indexing for formulating web images, using machine learning approaches. To access efficient image with annotation, we introduce and implement an effective and novel computational evaluation approach, i.e., non-training-based label index refinement approach with convex optimization classification for applying large data preprocessing for weakly labeled data in image indexing. We also develop an approximation-based grouping algorithm to improve precision and recall efficiency in large web-based image retrieval tasks. Our experimental results show efficient image indexing with different experimental studies on large scale web-based images. Our experiments in search-based annotation yield scalability measure results in the range of 85–90% as compared to the existing approaches in image retrieval.

Tag-Based Image Search by Social Re-ranking

Social media sharing websites like Flickr allow users to annotate images with free tags, which significantly contribute to the development of the web image retrieval and organization. Tag-based image search is an important method to find images contributed by social users in such social websites. However, how to make the top ranked result relevant and, with diversity, is challenging. In this paper, we propose a social re-ranking system for tag-based image retrieval with the consideration of an image's relevance and diversity. We aim at re-ranking images according to their visual information, semantic information, and social clues. The initial results include images contributed by different social users. Usually each user contributes several images. First, we sort these images by inter-user re-ranking. Users that have higher contribution to the given query rank higher. Then we sequentially implement intra-user re-ranking on the ranked user's image set, and only the most relevant image from each user's image set is selected. These selected images compose the final retrieved results. We build an inverted index structure for the social image dataset to accelerate the searching process. Experimental results on a Flickr dataset show that our social re-ranking method is effective and efficient.

A Survey on User Search Goal Inferring and Re-Ranking System

A Survey on User Search Goal Inferring and Re-Ranking System

Improving Statistical Machine Translation using Syntax-based Learning-to-Rank System

Word reordering is one of the fundamental problems of machine translation. It is an important factor in the quality and efficiency of machine translations. Tackling the reordering problem can lead to significant improvements in translation quality. A new method is introduced with the objective of solving the reordering problem. It exploits sophisticated syntactic-based features for re-ranking the n -best translation candidates provided by a phrase-based statistical machine translation system. These sophisticated reordering features are based on an innovative structure named the phrasal dependency tree, which is inspired from target-side dependency relations among contiguous non-syntactic phrases. The features benefit from phrase dependencies, translation orientation, and distance. A translation candidate is modelled as a directed and weighted graph built from information provided by the reordering features and is re-scored by the proposed re-ranking system. This system markedly improves the outputs of the machine translation of two syntactically divergent language pairs. The performance is evaluated for Persian→English and German→English translation tasks using the WMT07 benchmark. The results report 0.566/0.95/0.011- and 0.75/0.97/0.024-point improvements in terms of BLEU/TER/LRSCORE metrics on Persian→English and German→English translation tasks, respectively. The superiority of the proposed system in terms of precision and recall measures is demonstrated as well.

ACM ICMR 2014 best papers in image retrieval

The ACM International Conference onMultimedia Retrieval (ICMR) is the top multimedia retrieval conference worldwide. This year it was held in Glasgow, Scotland by the general chairs: Joemon Jose and Keith van Rijsbergen. It is my honor to present in this issue the extended versions of the best image retrieval papers which were selected on the basis of the evaluations from the program committee (special thanks to the program chairs: Mohan Kankanhalli, Stefan Rueger and R. Manmatha). One of the grand challenges for multimedia information retrieval is automatic image annotation because it can provide a direct bridge over the semantic gap from low-level features to human level tags and annotations. The paper “A Sparse Kernel RelevanceModel for Automatic Image Annotation” by SeanMoran andVictor Lavrenko introduces a new form of the Continuous RelevanceModel which can perform adaptive kernel selection per feature type. Also, they describe two novel data-adaptive kernels which improve the modeling of image feature distributions. Their findings show that the adaptive kernel assignment can yield significantly higher annotation accuracy. After the initial results have been found from an image retrieval query, a fundamental problem is to re-rank the images toward providing the user withmore accurate and relevant results. The paper “Image re-ranking system based on frequent closed patterns” by Winn Voravuthikunchai, Bruno

Image re-ranking system based on closed frequent patterns

Text-based image retrieval is a popular and simple framework, which consists in using text annotations (e.g., image names, tags) to efficiently collect images relevant to a query word, from very large image collections. Even if the set of images retrieved using text annotations is noisy, it constitutes a reasonable initial set of images that can be considered as a bootstrap and improved further by analyzing image content. In this context, this paper introduces an approach for improving this initial set by re-ranking the so-obtained images, assuming that non-relevant images are scattered (i.e., they do not form clusters), unlike the relevant ones. More specifically, the approach consists in computing efficiently and on-the-fly closed frequent patterns, and in re-ranking images based on the number of patterns they contain. To do this, the paper introduces a simple but powerful new scoring function. Moreover, after the re-ranking process, we show how pattern mining techniques can also be applied for promoting diversity in the top-ranked images. The approach is validated on three different datasets for which state-of-the-art results are obtained.

A swarm-inspired re-ranker system for statistical machine translation

Recently, re-ranking algorithms have been successfully applied on statistical machine translation systems. Due to the errors in the hypothesis alignment and varying word order between the source and target sentences and also the lack of sufficient resources such as parallel corpora, decoding may result in ungrammatical or non-fluent outputs. This paper proposes a re-ranking system based on swarm algorithms, which makes the use of sophisticated non-syntactical features to re-rank the n-best translation candidates. We introduce plenty of easy-computed non-syntactical features to deal with SMT system errors plus the quantum-behaved particle swarm optimization (QPSO) algorithm to adjust the weights of features. We have evaluated the proposed approach on 2 translation tasks in different language pairs (Persian→English and German→English) and genres (news and novel books). In comparison with PSO-, GA-, Perceptron- and averaged Perceptron-style re-ranking systems, the experimental study demonstrates the superiority of the proposed system in terms of translation quality on both translation tasks. In addition, the impacts of the proposed features on the translation quality have been analyzed, and the most positive ones have been recognized. At the end, the impact of the n-best list size on the proposed system is investigated.

INFORMATION SEARCH USING TEXT AND IMAGE QUERY

An image retrieval and re-ranking system utilizing a visual re-ranking framework which is proposed in this paper the system retrieves a dataset from the World Wide Web based on textual query submitted by the user. These results are kept as data set for information retrieval. This dataset is then re-ranked using a v isual query (multiple images selected by user from the dataset) which conveys user’s intention semantically. Visual descriptors (MPEG-7) which describe image with respect to low-level fea ture like color, texture, etc are used for calculating distances. These distances are a measure of similarity between query images and m embers of the dataset. Our proposed system has been assessed on different type s of queries such as apples, Console, Paris, etc. Isignificant improvement on initial text-based search results.This system is well suitable for online shopping application. Index Terms : MPEG-7, Color Layout Descriptor (CLD), Edge Histo gram Descriptor (EHD), image retrieval and reranking system

Feedlot performance, feed efficiency reranking, carcass traits, body composition, energy requirements, meat quality and calpain system activity in Nellore steers with low and high residual feed intake

The aim was to evaluate growth, carcass traits, feed efficiency reranking, body composition, calpain system activity and meat quality in Nellore steers that were phenotypically ranked for high and low residual feed intake (RFI). Seventy-two Nellore steers (16–21 month-old, 334±19kg BW) had free access to a feedlot diet for 70d (feeding period 1, P1). Daily dry matter intake (DMI), body weight gain (ADG) and ultrasound carcass traits were measured individually. The 12 lowest and the 12 highest RFI steers were classed as low- and high-RFI groups and were fed for a second feeding period (feeding period 2, P2). Spearman's rank correlation was performed for RFI and gain-to-feed ratio (G:F) measured over P1 and P2. The carcass traits, meat quality and calpain system activity were evaluated at slaughter, and body composition was determined. In P1, low-RFI steers had greater G:F (0.159 vs. 0.134; P<0.001), lower DMI (9.30 vs. 11.1kg/d; P<0.0001), lower RFI (−0.80 vs. 0.85kg/d; P<0.0001), and tended to have lower rates of rump fat gain (4.48 vs. 6.05mm; P=0.06), but no differences were observed for ADG, BW, and other traits measured by ultrasound (P>0.05). In P2, no differences between RFI classes were observed for G:F and DMI as a percentage of BW (P>0.05), and small differences were observed for DMI (8.25 vs. 8.99kg/d, P<0.05) and RFI (−0.28 vs. +0.29kg/d, P=0.06). The ranking correlations for RFI and G:F measured consecutively in P1 and P2 were low to moderate (r=0.11–0.40). Low-RFI steers had lower requirements of metabolizable energy (ME) for maintenance (131 vs. 160Mcal/kg EBW0.75d−1, P<0.05), but no differences were observed for fat and protein gain, retained energy and efficiency of ME use for gain (P>0.05). No differences between low- and high-RFI steers were observed for carcass traits at slaughter; however, low-RFI steers had 8.1kg less gastrointestinal fat than high-RFI steers. There were no RFI effects on meat shear force and the activities of μ-calpain, m-calpain and calpastatin (P>0.05). High-RFI steers had greater MFI in 1d aged LM (53.9 vs. 40.8, P<0.05). The variation in feed efficiency between high- and low-RFI Nellore steers is related to differences in energy requirements and deposition of fat on internal organs. The ranking for feed efficiency is altered as cattle become older. The selection for improved RFI in Nellore cattle may reduce feed requirements for beef production without affecting meat tenderness and enzymatic activity of the calpain system.