Field Of Information Retrieval Research Articles

Multilingual Scene Text Detection Using Gradient Morphology

Text detection in natural scene images is an interesting problem in the field of information retrieval. Several methods have been proposed over the past few decades for scene text detection. However, the robustness and efficiency of these methods are downgraded due to high sensitivity towards various complexities of an image. Also, in multi-lingual environment where texts may occur in multiple languages, a method may not be suitable for detecting scene texts in certain languages. To counter these challenges, a gradient morphology-based method is proposed in this paper that proves to be robust against image complexities and efficiently detects scene texts irrespective of their languages. The method is validated using low quality images from standard multi-lingual datasets like MSRA-TD500 and MLe2e. The performance of the method is compared with that of some state-of-the-art methods, and comparably better results are observed.

Unveiling the Hierarchical Structure of Music by Multi-Resolution Community Detection

Human perception of musical structure is supposed to depend on the generation of hierarchies, which is inherently related to the actual organisation of sounds in music. Musical structures are indeed best retained by listeners when they form hierarchical patterns, with consequent implications on the appreciation of music and its performance. The automatic detection of musical structure in audio recordings is one of the most challenging problems in the field of music information retrieval, since even human experts tend to disagree on the structural decomposition of a piece of music. However, most of the current music segmentation algorithms in literature can only produce flat segmentations, meaning that they cannot segment music at different levels in order to reveal its hierarchical structure. We propose a novel methodology for the hierarchical analysis of music structure that is based on graph theory and multi-resolution community detection. This unsupervised method can perform both the tasks of boundary detection and structural grouping, without the need of particular constraints that would limit the resulting segmentation. To evaluate our approach, we designed an experiment that allowed us to compare its segmentation performance with that of the current state of the art algorithms for hierarchical segmentation. Our results indicate that the proposed methodology can achieve state of the art performance on a well-known benchmark dataset, thus providing a deeper analysis of musical structure.

Automatska ekstrakcija ključnih riječi iz teksta standardnim računalnim postupcima

Automatska ekstrakcija ključnih riječi iz teksta aktualan je istraživački problem u području računalne analize prirodnog jezika i pretraživanja informacija. Iako su razvijene brojne metode za ekstrakciju ključnih riječi iz teksta, njihova učinkovitost ovisna je o brojnim faktorima poput pristupa kojim su konstruirane, domene na koju su prilagođene, vrste jezika ili zadataka za koji su konstruirane i sl., a samim time prostor za napredak u smislu nadogradnje i poboljšanja, uvijek postoji. U ovom radu objašnjene su i rekonstruirane dvije postojeće metode – RAKE i MAUI, a koje su standardni predstavnici nenadzirane i nadzirane skupine metoda. Eksperimentalno je ispitano mogu li metode uspješno ekstrahirati ključne riječi iz tekstova pisanih na talijanskom jeziku, na kojem do sada nisu usporedno testirane. Za potrebe eksperimenta prikupljeni su i ručno označeni talijanski tekstovi. Efikasnost MAUI metode pokazala se perspektivnijom u odnosu na RAKE metodu što je već ranije potvrđeno u eksperimentu ekstrakcije ključnih riječi iz tekstova pisanih na engleskom jeziku.

Using Ontology-Based Approach to Improved Information Retrieval Semantically for Historical Domain

Searching and retrieving documents from large historical archives prove to be challenging for the information retrieval (IR) field as historians typically employ their knowledge, experience, and intuition. There are several works done on the application of IR in historical documents. As such, the conventional IR model is mostly used a simple Bag-of-Word (BOW) approach and usually unable to support precise document retrieval for the domain of history. We proposed an ontology-based approach to semantically index and ranked rich historical documents. The historical documents relating to the Vietnam War were chosen for this study. Several existing ontologies have been reviewed to identify the most suitable concepts and properties which contain rich information pertaining to relevant entities such as an event, time, and people. The domain ontology was developed by utilizing the existing Simple News and Press (SNaP) ontology and extended with concepts related to the Vietnam War. The ontology was then semantically mapped with concepts found in a collection of 133 documents relating to the Vietnam war. In this paper, we also proposed a simple ontology-based weighting mechanism derived from the classic tf-idf scoring scheme. Finally, 20 SPARQL queries are implemented to do the evaluation. The evaluation shows that the proposed ontological-based approach achieved better results as compared to the base-line BM-25 probabilistic retrieval model in terms of precision and recall metrics. The use of the ontology-based approach in document retrieval can compete with the keyword-based approach.

Assessing ranking metrics in top-N recommendation

The evaluation of recommender systems is an area with unsolved questions at several levels. Choosing the appropriate evaluation metric is one of such important issues. Ranking accuracy is generally identified as a prerequisite for recommendation to be useful. Ranking metrics have been adapted for this purpose from the Information Retrieval field into the recommendation task. In this article, we undertake a principled analysis of the robustness and the discriminative power of different ranking metrics for the offline evaluation of recommender systems, drawing from previous studies in the information retrieval field. We measure the robustness to different sources of incompleteness that arise from the sparsity and popularity biases in recommendation. Among other results, we find that precision provides high robustness while normalized discounted cumulative gain offers the best discriminative power. In dealing with cold users, we also find that the geometric mean is more robust than the arithmetic mean as aggregation function over users.

Content-Based Document Image Retrieval Based on Document Modeling

Recently, language models have gained importance in the field of information retrieval. In this paper, we propose a generic language model to improve a content-based document retrieval system. In this approach, character images are extracted, clustered, and analyzed to form high-level semantic terms using a statistical document model. This model simulates the long-term relationships between characters. Documents are then indexed according to these terms, and a query document is proposed to retrieve the relevant documents. The query document can be a single keyword, or it can be synthesized from a text string. The aim is to generate a semantic representation from low-level image pixels through pattern matching and document modeling. The conventional approach of generating semantic terms in document retrieval includes every possible symbol sequence in the feature representation. Comparatively, our approach can considerably reduce the dimensions of the feature space while producing retrieval results comparable to those of the conventional and state-of-the-art approaches.

Dagstuhl seminar 19461 on conversational search

In the week of November 10--15, 2019, 44 researchers from the fields of information retrieval and Web search, natural language processing, human computer interaction, and dialogue systems met for the Dagstuhl Seminar 19461 "Conversational Search" to share the latest development in the area of conversational search and discuss its research agenda and future directions. The clear signal from the seminar is that research opportunities to advance conversational search are available to many areas and that collaboration in an interdisciplinary community is essential to achieve the goals. This report overviews the program and selected findings of the working groups.

Towards optimize-ESA for text semantic similarity: A case study of biomedical text

Explicit Semantic Analysis (ESA) is an approach to measure the semantic relatedness between terms or documents based on similarities to documents of a references corpus usually Wikipedia. ESA usage has received tremendous attention in the field of natural language processing NLP and information retrieval. However, ESA utilizes a huge Wikipedia index matrix in its interpretation by multiplying a large matrix by a term vector to produce a high-dimensional vector. Consequently, the ESA process is too expensive in interpretation and similarity steps. Therefore, the efficiency of ESA will slow down because we lose a lot of time in unnecessary operations. This paper propose enhancements to ESA called optimize-ESA that reduce the dimension at the interpretation stage by computing the semantic similarity in a specific domain. The experimental results show clearly that our method correlates much better with human judgement than the full version ESA approach.

Feature Selection us ing Normalized Weight Method f or Tamil Text Classification

The Feature Selection process simplify the Tamil text classification work at present we are in the information age, in this period all the applications has great growth in the domain of World Wide Web, so regional language like Tamil materials such as web pages, e-mails, e-books, and digital data has grown enormously so the retrieval of this Tamil digital document is more wanted by Tamil Document searcher. For quick retrieval of needed Tamil digitized documents among the millions of Tamil web documents, these documents should be classified by content according to their classes. The Tamil Text classification is a background work for many Tamil NLP applications such as query response, information extraction, information summarization, etc. the implementation of text categorization is very important in the information retrieval field. The text categorization assigns a document an appropriate category from a predefined group of categories. Tamil Text Classification classifies the documents based on Tamil text in a Document. Tamil language words are very rich in morphology and hence Tamil language consists of very large set of word forms. So it is important to reduce the features of Tamil text. This paper discusses about Feature selection Using Normalized weight from the huge set of key words from the preprocessed corpus. The Feature selection done by Term Weighting (TF*IDF) normalized method is reducing the size of the key word list which is very useful for training and testing Tamil text classification algorithms.

Survey on Information Retrieval Accuracy in Biomedical Systems

In recent years, research articles in the biomedicine domain have increased rapidly which makes it difficult for medical personnel to manually gather all the information required. To meet this information, need better, Information Retrieval (IR) techniques have been employed for accurate and timely search of the information. A special lexicon set identifies the medical domain and hence requires specialized processing steps to handle complex medical terminologies. The query and result mapping may need extra attention in order to give best results. The queries may be in the form of short questions, medical case reports, medical case narratives, verbose medical queries, community questioning, semi-structured queries, etc. The paper intends to explore the work done in the field of medical text information retrieval. Models developed or employed, indexing and ranking is surveyed and comparison is done among them.

Approximate matching-based unsupervised document indexing approach: application to biomedical domain

Document indexing is considered as a crucial phase in the information retrieval field because textual information is constantly increasing. With this accumulation of documents, the satisfaction of user needs becomes more and more complex. For these reasons, several information retrieval systems have been designed in order to respond to user requests. The main contribution of the current work resides in the suggestion of a novel hybrid approach for biomedical document indexing. We improve the estimation of the correspondence between a document and a given concept using two methods: vector space model (VSM) and description logics (DL). VSM performs partial matching between documents and external resource terms. DL allows representing knowledge in a relevant manner for better matching. The proposed contribution reduces the limitation of exact matching. It serves to index documents by exploiting medical subject headings (MeSH) thesaurus services with approximate matching. The latter partially matches document terms with biomedical vocabularies to extract other morphological variants in that resource. It also generates irrelevant concepts. The filtering step solves this problem and grants the selection of the most important concepts by exploiting the knowledge provided by MeSH. The experiments, carried out on different corpora, show encouraging results of around 25% improvement in average accuracy compared to other approaches studied in the literature.

Optimization of hierarchical reinforcement learning relationship extraction model

PurposeEntity relation extraction is an important research direction to obtain structured information. However, most of the current methods are to determine the relations between entities in a given sentence based on a stepwise method, seldom considering entities and relations into a unified framework. The joint learning method is an optimal solution that combines relations and entities. This paper aims to optimize hierarchical reinforcement learning framework and provide an efficient model to extract entity relation.Design/methodology/approachThis paper is based on the hierarchical reinforcement learning framework of joint learning and combines the model with BERT, the best language representation model, to optimize the word embedding and encoding process. Besides, this paper adjusts some punctuation marks to make the data set more standardized, and introduces positional information to improve the performance of the model.FindingsExperiments show that the model proposed in this paper outperforms the baseline model with a 13% improvement, and achieve 0.742 in F1 score in NYT10 data set. This model can effectively extract entities and relations in large-scale unstructured text and can be applied to the fields of multi-domain information retrieval, intelligent understanding and intelligent interaction.Originality/valueThe research provides an efficient solution for researchers in a different domain to make use of artificial intelligence (AI) technologies to process their unstructured text more accurately.

A Framework for Group Converging Pattern Mining using Spatiotemporal Trajectories

A group event such as human and traffic congestion can be very roughly divided into three stages: converging stage before congestion, gathered stage when congestion happens, and dispersing stage that congestion disappears. It is of great interest in modeling and identifying converging behaviors before gathered events actually happen, which helps to proactively predict and handle potential public incidents such as serious stampedes. However, most of existing literature put too much emphasis on the second stage, only a few of them is dedicated to the first stage. In this paper, we propose a novel group pattern, namely converging, which refers to a group of moving objects converging from different directions during a certain period before gathered. To discover efficiently such converging patterns, we develop a framework for converging pattern mining (CPM) by examining how moving objects form clusters and the process of the “cluster containment”. The framework consists of three phases: snapshot cluster discovery phase, cluster containment join phase, and converging detection phase. As cluster containment mining is the key step, we develop three algorithms to discover cluster containment matches: a containment-join-algorithm, called SSCCJ, by using spatial proximity; a signature tree-based cluster-containment-join-algorithm, called STCCJ, which takes advantage of the cluster containment relations and signature techniques to filter enormous unqualified candidates in an efficient and effective way; and third, to keep the advantages of the above algorithms while avoiding their flaws, we further propose a signature quad-tree based cluster-containment-join algorithm, called SQTCCJ, which can identify efficiently matches by considering cluster spatial proximity as well as containment relations simultaneously. To assess the proposed methods, we redefine two evaluation metrics based on the concept of “Precision and Recall” in the field of information retrieval and the characteristics of converging patterns. We also propose a new indicator for measuring the duration of the converging stage in a group event. Finally, the effectiveness of the CPM and the efficiency of the mining algorithms are evaluated using three types of trajectory datasets, and the results show that the SQTCCJ algorithm demonstrates a superior performance.

Recognizing Indonesian Acronym and Expansion Pairs with Supervised Learning and MapReduce

During the previous decades, intelligent identification of acronym and expansion pairs from a large corpus has garnered considerable research attention, particularly in the fields of text mining, entity extraction, and information retrieval. Herein, we present an improved approach to recognize the accurate acronym and expansion pairs from a large Indonesian corpus. Generally, an acronym can be either a combination of uppercase letters or a sequence of speech sounds (syllables). Our proposed approach can be computationally divided into four steps: (1) acronym candidate identification; (2) acronym and expansion pair collection; (3) feature generation; and (4) acronym and expansion pair recognition using supervised learning techniques. Further, we introduce eight numerical features and evaluate their effectiveness in representing the acronym and expansion pairs based on the precision, recall, and F-measure. Furthermore, we compare the k-nearest neighbors (K-NN), support vector machine (SVM), and bidirectional encoder representations from transformers (BERT) algorithms in terms of accurate acronym and expansion pair classification. The experimental results indicate that the SVM polynomial model that considers eight features exhibits the highest accuracy (97.93%), surpassing those of the SVM polynomial model that considers five features (90.45%), the K-NN algorithm with k = 3 that considers eight features (96.82%), the K-NN algorithm with k = 3 that considers five features (95.66%), BERT-Base model (81.64%), and BERT-Base Multilingual Cased model (88.10%). Moreover, we analyze the performance of the Hadoop technology using various numbers of data nodes to identify the acronym and expansion pairs and obtain their feature vectors. The results reveal that the Hadoop cluster containing a large number of data nodes is faster than that with fewer data nodes when processing from ten million to one hundred million pairs of acronyms and expansions.

A Novel on Classification Techniques of News with Help of Sentiment Detection

The problem of data classification is an important topic in the field of machine learning and information retrieval. This has been widely studied and has been applied in many fields. There are multiple models which are proposed for the classification, like tree-structured classifiers, genetic algorithms, Bayesian classification, neural networks etc. These have a large range of applications in different areas like, fraud/spam detection, Customer Segmentation, Medical Diagnosis, Credit approval, weather prediction etc. This project tries to aim at a particular subclass of classification, namely sentiment analysis. Hybrid techniques should be applied in this field of study as each of the existing models have brought about some new expertise and their improvements need to be combined to give higher performance and accuracy. The sentiment analysis problem requires to take as input a block of text and correctly predict the sentiment of the writer or the speaker of the text. We have sufficient data to build a system that uses hybrid techniques like Naïve Bayes and combines the existing models to perform sentiment analysis on a dataset and study its results. The hybrid approach using Naïve Bayes to this problem gives promising results.

2SRM: Learning social signals for predicting relevant search results

Search systems based on both professional meta-data (e.g., title, description, etc.) and social signals (e.g., like, comment , rating, etc.) from social networks is the trending topic in information retrieval (IR) field. This paper presents 2SRM (Social Signals Relevance Model), an approach of IR which takes into account social signals (users' actions) as an additional information to enhance a search. We hypothesize that these signals can play a role to estimate a priori social importance (relevance) of the resource (document). In this paper, we first study the impact of each such signal on retrieval performance. Next, some social properties such as popularity, reputation and freshness are quantified using several signals. The 2SRM combines the social relevance, estimated from these social signals and properties, with the conventional textual relevance. Finally, we investigate the effect of the social signals on the retrieval effectiveness using state-of-the-art learning approaches. In order to identify the most effective signals, we adopt feature selection algorithms and the correlation between the signals. We evaluated the effectiveness of our approach on both IMDb (Internet Movie Databese) and SBS (Social Book Search) datasets containing movies and books resources and their social characteristics collected from several social networks. Our experimental results are statistically significant, and reveal that incorporating social signals in retrieval model is a promising approach for improving the retrieval performance.

QAOC: Novel query analysis and ontology-based clustering for data management in Hadoop

Bottleneck issues handled in the field of information retrieval are analysis of query and management of data storage. Hadoop is a large scale environment that is supported with larger storage and faster processing. Even though, it suffers from these challenging issues while the number of information requesters is higher. This paper addresses these two bottleneck issues in Hadoop by retrieving the information with the design of Query Analysis and Ontology-based Clustering (QAOC) architecture. In QAOC architecture, the components involved are query manager, scheduler and data management. Initially the query manager consolidates the query if they are similar; hereby the searching time is effectively minimized. Then the user query is scheduled in neuro-fuzzy by computing query arrival time, query length and query expiry time. The data management in the back-end is operated by weighted ontology-based clustering method to cluster the data based on their relevancy. The scheduled user query is searched in the ontology based balanced binary tree and lastly the relevant results are ranked using Okapi BM25 and delivered to user. This QAOC architecture is experimented on Hadoop 2.7 and the results are compared in terms of execution time, processing speed and memory consumption.

Feature Weighting in Finding Feedback Documents for Query Expansion in Biomedical Document Retrieval

Finding good feedback documents for query expansion is a well-known problem in the field of information retrieval. This paper describes a novel approach for finding relevant documents for feedback in query expansion for biomedical document retrieval. The proposed approach relies on a small amount of human intervention to find good feedback documents and tries to learn the relation between query and documents in terms of usefullness of document for query expansion. This proposed approach uses an NLP-based feature weighting technique with classification and clustering method on the documents and identifies relevant documents for feedback. The documents are represented using term frequency and inverse document frequency (TF–IDF) features and these features are weighted according to the type of query and type of the terms. The experiments performed on CDS 2014, 2015 and 2016 datasets show that the feature weighting in finding feedback documents for query expansion approach gives good results as compared to the results of pseudo-relevance feedback, relevance feedback and the results of TF–IDF features without weighting.

DL-VSM based document indexing approach for information retrieval

Textual information is constantly increasing. With this accumulation of documents, the satisfaction of user needs becomes more and more complex. For that, several information retrieval systems have been designed in order to respond to user requests. Document indexing is considered as a crucial phase in the information retrieval field. The main contribution of the current work resides in the suggestion of a novel hybrid approach for biomedical document indexing. We improve the estimation of the correspondence between a document and a given concept using two methods: vector space model (VSM) and description logics (DL). VSM performs partial matching between documents and external resource terms. DL allows representing knowledge in a relevant manner for better matching. The proposed contribution reduces the limitation of exact matching. It serves to index documents by exploiting medical subject headings (MeSH) thesaurus services with approximate matching. The latter partially matches document terms with biomedical vocabularies to extract other morphological variants in that resource. It also generates irrelevant concepts. The filtering step solves this problem and grants the selection of the most important concepts by exploiting the knowledge provided by MeSH. The experiments, carried out on different corpora, show encouraging results (+ 25% improvement in average accuracy compared to other approaches in the literature).

Extraction Algorithm of English Text Information From Color Images Based on Radial Wavelet Transform

The Chinese and English text information of color image reflects some important contents of color image to a certain extent, and these texts are extracted automatically, which has important application value in the fields of information retrieval, digital library, web page retrieval and intelligent transportation. In this paper, various signal extraction techniques based on radial wavelet transform modulus Maxima are analyzed, and it is found that these techniques have poor ability to extract weak signals, and the higher requirements for directionality lead to pseudo-boundary phenomena in two-dimensional image extraction results. Based on the correlation denoising method of radial wavelet coefficients, the radial wavelet entropy is introduced into the field of signal extraction, and the complex Morlet radial wavelet is selected as the basis function. A complex Morlet radial wavelet entropy extraction algorithm suitable for extracting the English text characteristics of weak signal color images is proposed. In addition, a method of scene text recognition is proposed. Based on D-SIFT local features and spatio-temporal histogram, this method vectorizes text samples in the framework of Bo Ws model. Because the spatio-temporal histogram can flexibly model the structural information of the text, this method can effectively describe the scene text. On this basis, the selective ensemble learning method is used to improve the performance of the text extractor. In order to expand the application range of the algorithm and improve the running efficiency of the integrated extractor, a model compression method based on color image English text samples and integrated extractor is proposed. The integrated extractor which takes up more space and runs slowly is compressed into an equivalent and more efficient one. In order to reduce the number of pseudo-samples needed in the process of model compression, a method of training local extractor based on color image English text samples is proposed, which greatly reduces the number of pseudo-samples needed and improves the efficiency of the model compression method.