Document Classification

This research deals with the use of self-organising maps for the classification of text documents. The aim was to classify documents to separate classes according to their topics. We therefore constructed self-organising maps that were effective for this task and tested them with German newspaper documents. We compared the results gained to those of k nearest neighbour searching and k-means clustering. For five and ten classes, the self-organising maps were better yielding as high average classification accuracies as 88-89%, whereas nearest neighbour searching gave 74-83% and k-means clustering 72-79% as their highest accuracies.

We propose a novel bio-inspired solution for biomedical article classification. Our method draws from an existing model of T-cell cross-regulation in the vertebrate immune system (IS), which is a complex adaptive system of millions of cells interacting to distinguish between harmless and harmful intruders. Analogously, automatic biomedical article classification assumes that the interaction and co-occurrence of thousands of words in text can be used to identify conceptually-related classes of articles-at a minimum, two classes with relevant and irrelevant articles for a given concept (e.g. articles with protein-protein interaction information). Our agent-based method for document classification expands the existing analytical model of Carneiro et al. , by allowing us to deal simultaneously with many distinct T-cell features (epitomes) and their collective dynamics using agent based modeling. We already extended this model to develop a bio-inspired spam-detection system . Here we develop our agent-base model further, and test it on a dataset of publicly available full-text biomedical articles provided by the BioCreative challenge . We study several new parameter configurations leading to encouraging results comparable to state-of-the-art classifiers. These results help us understand both T-cell cross-regulation and its applicability to document classification in general. Therefore, we show that our bio-inspired algorithm is a promising novel method for biomedical article classification and for binary document classification in general.

We present and study an agent-based model of T-Cell cross-regulation in the adaptive immune system, which we apply to binary classification. Our method expands an existing analytical model of T-cell cross-regulation that was used to study the self-organizing dynamics of a single population of T-Cells in interaction with an idealized antigen presenting cell capable of presenting a single antigen. With agent-based modeling we are able to study the selforganizing dynamics of multiple populations of distinct Tcells which interact via antigen presenting cells that present hundreds of distinct antigens. Moreover, we show that such self-organizing dynamics can be guided to produce an effective binary classification of antigens, which is competitive with existing machine learning methods when applied to biomedical text classification. More specifically, here we test our model on a dataset of publicly available full-text biomedical articles provided by the BioCreative challenge . We study the robustness of our model's parameter configurations, and show that it leads to encouraging results comparable to state-of-the-art classifiers. Our results help us understand both T-cell crossregulation as a general principle of guided self-organization, as well as its applicability to document classification. Therefore, we show that our bio-inspired algorithm is a promising novel method for biomedical article classification and for binary document classification in general.

Document (text) classification is a common method in e-business, facilitating users in the tasks such as document collection, analysis, categorization and storage. Semantic analysis can help to improve the performance of document classification. Though having been considered when designing previous methods for automatic document classification, more focus should be given to semantics with the increase number of content-rich electronic documents, forum posts or blogs online, which can reduce human workload by a great margin. This paper proposes a novel semantic document classification approach aiming to resolve two types of semantic problems: (1) polysemy problem, by using a novel semantic similarity computing strategy (SSC) and ( ) synonym problem, by proposing a novel strong correlation analysis method (SCM). Experiments show that our strategies can help to improve the performance of the baseline methods.

Abstract. This paper presents a deployed semantic web application in the cultural domain: the semantic portal MuseumFinland. It is a demonstration of a community portal and a publication channel by which heterogeneous collection database contents of different museums can be published on the Semantic Web. By semantic web techniques, it is possible to make collections semantically interoperable and provide the museum visitors with intelligent content-based search and browsing services to the global collection base.

Definir l'architecture d?un systeme logiciel est une etape essentielle de son developpement. C'est la premiere etape de prise de decisions de conception. C?est en particulier la que seront prises les decisions concernant la maniere de prendre en compte les exigences non fonctionnelles dont celles de qualite. Lorsqu?il s?agit de construire un nouveau systeme, il est interessant de valoriser la connaissance acquise dans le developpement d?autres systemes comparables, c'est-a-dire appartenant au meme domaine fonctionnel. Notre propos est d?etudier comment la capitalisation de la connaissance afferant a un domaine fonctionnel donne peut permettre d'atteindre l'objectif recherche, a savoir une architecture de qualite. Nous proposons une demarche de construction d'une architecture guidee par la qualite et appliquons cette demarche dans un premier temps pour le developpement d'une architecture de reference d'un domaine fonctionnel puis pour celle d'un sy...

Automatic classification of textual content in an under-resourced language is challenging, since lexical resources and preprocessing tools are not available for such languages. Their bag-of-words (BoW) representation is usually highly sparse and noisy, and text classification built on such a representation yields poor performance. In this paper, we explore the effectiveness of lexical normalization of terms and statistical feature pooling for improving text classification in an under-resourced language. We focus on classifying citizen feedback on government services provided through SMS texts which are written predominantly in Roman Urdu (an informal forward transliterated version of the Urdu language). Our proposed methodology performs normalization of lexical variations of terms using phonetic and string similarity. It subsequently employs a supervised feature extraction technique to obtain category-specific highly discriminating features. Our experiments with classifiers reveal that significant improvement in classification performance is achieved by lexical normalization plus feature pooling over standard representations.

This paper presents a new reduction algorithm which employs Constraint Satisfaction Techniques for removing redundant literals of a clause efficiently. Inductive Logic Programming (ILP) learning algorithms using a generate and test approach produce hypotheses with redundant literals. Since the reduction is known to be a co-NP-complete problem, most algorithms are incomplete approximations. A complete algorithm proposed by Gottlob and Fermüller is optimal in the number of θ-subsumption calls. However, this method is inefficient since it exploits neither the result of the θ-subsumption nor the intermediary results of similar θ-subsumption calls. Recently, Hirata has shown that this problem is equivalent to finding a minimal solution to a θ-subsumption of a clause with itself, and proposed an incomplete algorithm based on a θ-subsumption algorithm of Scheffer. This algorithm has a large memory consumption and performs many unnecessary tests in most cases. In this work, we overcome this problem by transforming the θ-subsumption problem in a Constraint Satisfaction Problem, then we use an exhaustive search algorithm in order to find a minimal solution. The experiments with artificial and real data sets show that our algorithm outperforms the algorithm of Gottlob and Fermüller by several orders of magnitude, particularly in hard cases.

This paper presents a new reduction algorithm which employs Constraint Satisfaction Techniques for removing redundant literals of a clause efficiently. Inductive Logic Programming (ILP) learning algorithms using a generate and test approach produce hypotheses with redundant literals. Since the reduction is known to be a co-NP-complete problem, most algorithms are incomplete approximations. A complete algorithm proposed by Gottlob and Fermüller is optimal in the number of θ-subsumption calls. However, this method is inefficient since it exploits neither the result of the θ-subsumption nor the intermediary results of similar θ-subsumption calls. Recently, Hirata has shown that this problem is equivalent to finding a minimal solution to a θ-subsumption of a clause with itself, and proposed an incomplete algorithm based on a θ-subsumption algorithm of Scheffer. This algorithm has a large memory consumption and performs many unnecessary tests in most cases. In this work, we overcome this problem by transforming the θ-subsumption problem in a Constraint Satisfaction Problem, then we use an exhaustive search algorithm in order to find a minimal solution. The experiments with artificial and real data sets show that our algorithm outperforms the algorithm of Gottlob and Fermüller by several orders of magnitude, particularly in hard cases.

In this paper we describe the two classification approaches (i.e. categorization and clustering) and their preceding steps. For each approach we give a brief description of the underlying theory and outline the advantages and disadvantages of the different methods. Finally we specify potential application areas in accordance with knowledge management and illustrate exemplarily one topic in detail. We describe the enhancement of queries for illustration purposes because it is a common research problem with respect to information retrieval and, thus, t o knowledge management.

This paper introduces a novel paradigm to impute missing data that combines a decision tree with an auto-associative neural network (AANN) based model and a principal component analysis-neural network (PCA-NN) based model. For each model, the decision tree is used to predict search bounds for a genetic algorithm that minimize an error function derived from the respective model. The models' ability to impute missing data is tested and compared using HIV sero-prevalance data. Results indicate an average increase in accuracy of 13% with the AANN based model's average accuracy increasing from 75.8% to 86.3% while that of the PCA-NN based model increasing from 66.1% to 81.6%.

Text document classification is an important research topic in the field of information retrieval, and so it is how we represent the information extracted from the documents to be classified. There exists document classification methods and techniques based on the vector space model, which doesn't capture the relation between words, which is considered of importance to make a better comparison and therefore classification. For this reason, two significant contributions were made, the first one is the way to create the feature vector for document comparison, which uses adapted concepts of non-rigid 3D models comparison and graphs as a data structure to represent such documents. The second contribution is the classification method itself, which uses the representative feature vectors of each category to classify new documents.

The increasing procedural demand in judicial institutions has caused a workload overload, impacting the efficiency of the legal system. This scenario, exacerbated by limited human resources, highlights the need for technological solutions to streamline the processing and analysis of documents. In light of this reality, this work proposes a pipeline for automating the classification of these documents, evaluating four methods of representing legal texts at the pipeline’s input: original text, summaries, centroids, and document descriptions.

The pipeline was developed and tested at the Public Defender’s Office of the State of Goiás (DPE-GO). Each approach implements a specific strategy to structure the input texts, aiming to enhance the models’ ability to interpret and classify legal documents. A new Portuguese dataset was introduced, specifically designed for this application, and the performance of Large Language Models (LLMs) was evaluated in classification tasks.

The analysis results demonstrate that the use of summaries improves classification accuracy and maximizes the F1-score, optimizing the use of LLMs by reducing the number of tokens processed without compromising precision. These findings highlight the impact of textual representations of documents and the potential of LLMs for the automatic classification of legal documents, as in the case of DPE-GO.

The contributions of this work indicate that the application of LLMs, combined with optimized textual representations, can significantly increase the productivity and quality of services provided by judicial institutions, promoting advancements in the overall efficiency of the legal system.

Effective decision making is based on accurate and timely information. However, human decision makers are often overwhelmed by the huge amount of electronic data these days. The main contribution of this paper is the development of effective information agents which can autonomously classify and filter incoming electronic data on behalf of their human users. The proposed information agents are innovative because they can quickly classify electronic documents solely based on the short titles of these documents. Moreover, supervised learning is not required to train the classification models of these agents. Document classification is based on information inference conducted over a high dimensional semantic information space. What is more, a belief revision mechanism continuously maintains a set of user preferred information categories and filter documents with respect to these categories. Preliminary experimental results show that our document classification and filtering mechanism outperforms the Support Vector Machines (SVM) model which is regarded as one of the best performing classifiers.

As businesses move toward smarter automation and AI-driven processes, understanding structured documents like invoices, forms, receipts, and reports has become essential. This is where LayoutLM plays a groundbreaking role. Developed by Microsoft, LayoutLM is a document AI model that combines visual layout, text content, and positional data to extract information with impressive accuracy. It is specifically designed for understanding documents in which structure and format are just as important as the text itself. Traditional natural language processing models treat text in a linear way, focusing only on the words and their sequence. But in structured documents, layout and positioning matter significantly. For example, a date printed in the corner of an invoice or a total amount in a specific table row carries semantic weight based on where it is placed. LayoutLM incorporates this spatial layout by combining OCR text with visual features and token positions on the document image. This spatial awareness allows the model to understand the meaning of content not just by what is said, but also by where it appears.

Text based Mining is the process of analyzing a document or set of documents to understand the content and meaning of the information they contain. Text Mining enhances human's ability to Process massive quantities of information and it has high Commercial values. Text mining, sometimes alternately referred to as text data mining, roughly, process of deriving highquality information from text. High-quality information is typically derived through the devising of patterns and trends through means such as statistical pattern learning. It usually involves the process of structuring the input text deriving patterns within the structured data, and finally evaluation and interpretation of the output. 'High quality' in text mining usually refers to some combination of relevance, novelty, and interestingness. Typical text mining tasks include text categorization, text clustering, concept/entity extraction, analysis and entity relation modeling (i.e., learning relations between named entities).

This paper describes the design and collection of NameDat, a database containing English proper names spoken by native Norwegians. The database was designed to cover the typical acoustic and phonetic variations that appear when Norwegians pronounce English names. The intended use of the database is acoustic and lexical modeling of these phonetic variations. The English names in the database have been enriched with several annotation tiers. The recorded names were selected according to three selection criteria: the familiarity of the name, the expected recognition performance and the coverage of non-native phonemes. The validity of the manual annotations was verified by means of an automatic recognition experiment of non-native names. The experiment showed that the use of the manual transcriptions from NameDat yields an increase in recognition performance over automatically generated transcriptions.

Recent analysis have identified a significant rise in transactional fraud, where bad actors seek to deceive individuals or firms into unauthorized financial actions. Traditional fraud detection systems frequently struggle to effectively identify such activities, leading to financial damages and security breaches. Addressing this problem requires employing sophisticated machine learning techniques specially designed to detect transactional fraud. This study presents a novel fraud detection method called "filter", aimed at uncovering misleading transactional behaviours. By employing tailored features to reveal fraudulent patterns and activities, our filter achieves a remarkable accuracy of over 99.01% in distinguishing fraudulent transactions from legitimate ones, while maintaining a low false positive rate. Our approach was evaluated with a dataset comprising 746 instances of fraudulent transactions and 4822 instances of legitimate transactions. The results underscore the superior performance of our filter compared to existing methods, particularly in accurately detecting fraudulent transactions. Moreover, our hybrid NB-ANN model achieves the highest accuracy of 99.01%, outperforming both Naïve Bayes (98.57%) and Artificial Neural Network (98.12%) techniques. This highlights the effectiveness of the hybrid method in boosting detection accuracy for transactional fraud. Implementing our filter and leveraging the hybrid NB-ANN model, organizations can greatly improve their ability to detect and prevent fraudulent activities thereby protecting their financial assets and maintaining customer trust.

The quick spurt in online transactions has also brought with it a parallel surge in fraud. With digital payments, the scope for fraud remains very high. Credit card fraud, among others, can cause heavy losses to customers and erode the confidence of consumers in Internet transactions. Furthermore, the detection of fraudsters in the online world poses big challenges. For one thing, there is an imbalance in data: fraud transactions are very few compared to genuine transactions. In this research paper, we intend to classify fraud through a supervised machine learning method. SVM is utilized for classification purposes. Based on the available dataset, we performed data analysis to obtain valuable information concerning fraud detection. To solve the problem of data imbalance, we first preprocessed the raw data by randomly choosing some legitimate transactions and normalizing the features. We also used feature selection and scaling methods to improve the accuracy of the model. After training the SVM on the sanitized dataset, we tested the model using performance measures like accuracy, precision, recall, and F1-score. These measures are especially important when working with skewed data. The findings show that the SVM model can detect fraudulent transactions with high precision and a good rate of recall. This shows that it can assist in reducing false alarms while being able to detect most fraud cases. In addition, we touch upon the tradeoff between false negatives and false positives because both pose very significant implications within financial institutions.

This paper describe a methodology for semi-automatic classification schema definition (a classification schema is a taxonomy of categories useful for automatic document classification). The methodology is based on: (i) an extensional approach useful to create a typology starting from a document base, and (ii) an intensional approach to build the classification schema starting from the typology. The extensional approach uses clustering techniques to group together documents on the basis of a similarity measure, whereas the intensional approach uses different operations (aggregation, reduction, generalization specialization) to define classes.

This paper describes our approach towards the ECML/PKDD Discovery Challenge 2010. The challenge consists of three tasks: (1) a Web genre and facet classification task for English hosts, (2) an English quality task, and (3) a multilingual quality task (German and French). In our approach, we create an ensemble of three classifiers to predict unseen Web hosts whereas each classifier is trained on a different feature set. Our final NDCG on the whole test set is 0.537 for Task 1, 0.844 for Task 2, and 0.823 (French) and 0.793 (German) for Task 3, which ranks fourth place in the ECML/PKDD Discovery Challenge 2010.

Year. SchoolNet provides global Internet access to secondary schools (grades 7-12) throughout Thailand. By using technology to improve our education system, this project supports the human-resource-development emphasised in the 8th National Economic and Social Development Plan and National Information Technology Committee (NITC) IT-2000 Plan. The purpose of SchoolNet is to bring Thai students ’ education up to a higher standard and level of quality as well as to bridge the information gap between students in the urban and rural areas by using information technology. Starting with secondary schools, students benefit from nation-wide access to a free web-hosting service and Internet accounts. The strength of a nation depends on the quality of its citizens. In order to remain economically competitive in the age of an increasingly globalized digital economy; Thailand needs a well-educated workforce adept at the use of Information Technology. Unfortunately, Thailand's existing educa...

Various platforms, including patent systems and repositories like GitHub and arXiv, support knowledge dissemination across domains. As knowledge increasingly spans multiple disciplines, there is a need to track innovations that intersect various fields. Despite available data, a comprehensive knowledge taxonomy for effectively tracking innovations across domains is lacking. Developing such a taxonomy and employing automated classification methods will enhance the ability to track shared knowledge. In this work, we first developed a knowledge taxonomy based on the CPC schema. We formulated the classification of textual data into defined knowledge fields as a multi-label problem. Then, we evaluated the effectiveness of the classification models by fine-tuning pre-trained transformer language models. The multilabel framework enables the tracking of knowledge trends at the intersection of various disciplines.

Huge number of documents are increasing rapidly, therefore, to organize it in digitized form text categorization becomes an challenging issue. A major issue for text categorization is its large number of features. Most of the features are noisy, irrelevant and redundant, which may mislead the classifier. Hence, it is most important to reduce dimensionality of data to get smaller subset and provide the most gain in information. Feature selection techniques reduce the dimensionality of feature space. It also improves the overall accuracy and performance. Hence, to overcome the issues of text categorization feature selection is considered as an efficient technique. Therefore, we proposed a multistage feature selection model to improve the overall accuracy and performance of classification. In the first stage document preprocessing part is performed. Secondly, each term within the documents are ranked according to their importance for classification using the information gain. Thirdly rough set technique is applied to the terms which are ranked importantly and feature reduction is carried out. Finally a document classification is performed on the core features using Naive Bayes and KNN classifier. Experiments are carried out on three UCI datasets, Reuters 21578, Classic 04 and Newsgroup 20. Results show the better accuracy and performance of the proposed model.

Some of the key questions in paleography are those of classification, namely trying to ascertain when and where a given manuscript was written, and — if possible — by whom. Paleographers bring many skills and tools to bear on these questions in what is often a complicated and laborious task requiring reference to paleographic, linguistic and archaeological data, among others. Because it is difficult to quantify the degree of certainty in the final readings and assessments, or even to articulate the arguments underlying these readings, experts have begun to develop computer-based methods for paleographic research in which the description of the various findings is made explicit (Ciula, 2005; Stokes, 2008; Aussems and Brink, 2009; Hofmeister et al., 2009). Some of these computer-based approaches involve little or no human intervention. However, others require manual selection of regions in the image or manual recording of descriptors, that is, of features in the handwriting which are ...

Documents are one of the most common methods for maintaining data and records. Everyday a lot of documents/files are generated with lot of data for future research purposes or for business analytics. These files/documents must should be stored effectively so that it can be retrieved whenever needed. Organizing large documents can be a tedious task as the internal content of the files are not known. Manually organizing each and every file is not practically possible as it may take hours to categorize a file based on its contents and also the accuracy of classification cannot be guaranteed. In the fields like Library Science a huge amount of files are required to be maintained, which can be helpful in future for business decisions or for research purpose. To make this task easier Text Document Classifier can be used. Text Document Classifier can classify a given document based on the contents inside the document and label the document from the pre-defined classes. Unlike traditional classification Techniques in Machine Learning like Support Vector Machine, term frequency-identification and Naïve Bayes Classifier, Neural Networks has better analytical results. Traditional Classification Methods has limitations in terms of effective feature extraction and the dimensionality problem, these limitations can be solved by Convolutional Neural Networks.

At the preceding GREC, we have proposed to use a "bag of symbols" formalism (similar to the bag of words approach) for the indexing of a graphical document image database. In this paper, we extend the proposed approach through the introduction of a rejection stage in the system. This rejection is based on the use of an original One Class Classifier. Some preliminary results are proposed.

An unprecedented rise in Unified Payment Interface (UPI) for the transactions in rural areas make them prone to fraud. In this paper, authors present a new method for detecting UPI fraud based on the name of the bank account, transaction ID, and the amount of money that has been transacted. Three machine learning algorithms-Random Forest, K-Nearest Neighbor (KNN), and Decision Tree-are employed to classify transactions into two categories: Some of the messages so designed were 'Fraud: Unusual Activity Detected' and 'No Fraud: Details Verified and Processed'. Also, Random Forest provides high accuracy level in comparison to possible overfitting, KNN has a further advantage in comparison to other labeled datasets, and Decision Tree contributes to better understanding of an algorithm for making decisions. The idea of the new system is to enhance the problem of fraud in UPI transactions in rural areas to avoid unauthorized transactions.

Document similarity is an important part of Natural Language Processing and is most commonly used for plagiarism-detection and text summarization. Thus, finding the overall most effective document similarity algorithm could have a major positive impact on the field of Natural Language Processing. This report sets out to examine the numerous document similarity algorithms, and determine which ones are the most useful. It addresses the most effective document similarity algorithm by categorizing them into 3 types of document similarity algorithms: statistical algorithms, neural networks, and corpus/knowledge-based algorithms. The most effective algorithms in each category are also compared in our work using a series of benchmark datasets and evaluations that test every possible area that each algorithm could be used in.

Document similarity is an important part of Natural Language Processing and is most commonly used for plagiarism-detection and text summarization. Thus, finding the overall most effective document similarity algorithm could have a major positive impact on the field of Natural Language Processing. This report sets out to examine the numerous document similarity algorithms, and determine which ones are the most useful. It addresses the most effective document similarity algorithm by categorizing them into 3 types of document similarity algorithms: statistical algorithms, neural networks, and corpus/knowledge-based algorithms. The most effective algorithms in each category are also compared in our work using a series of benchmark datasets and evaluations that test every possible area that each algorithm could be used in.

Credit card fraud detection remains a significant challenge for financial institutions and consumers globally, prompting the adoption of advanced data analytics and machine learning techniques. In this study, we investigate the methodology and performance evaluation of various machine learning algorithms for credit card fraud detection, emphasizing data preprocessing techniques and model effectiveness. Through thorough dataset analysis and experimentation using cross-validation approaches, we assess the performance of logistic regression, decision trees, random forest classifiers, Naïve Bayes classifiers, K-nearest neighbors (KNN), and artificial neural networks (ANN-DL). Key performance metrics such as accuracy, sensitivity, specificity, and F1-score are compared to identify the most effective models for detecting fraudulent transactions. Additionally, we explore the impact of different folds in cross-validation on model performance, providing insights into the classifiers' robustness and stability. Our findings contribute to the ongoing efforts to develop efficient fraud detection systems, offering valuable insights for financial institutions and researchers striving to combat credit card fraud effectively.

The detection of outliers in text documents is a highly challenging task, primarily due to the unstructured nature of documents and the curse of dimensionality. Text document outliers refer to text data that deviates from the text found in other documents belonging to the same category. Mining text document outliers has wide applications in various domains, including spam email identification, digital libraries, medical archives, enhancing the performance of web search engines, and cleaning corpora used in document classification. To address the issue of dimensionality, it is crucial to employ feature selection techniques that reduce the large number of features without compromising their representativeness of the domain. In this paper, we propose a hybrid density-based approach that incorporates mutual information for text document outlier detection. The proposed approach utilizes normalized mutual information to identify the most distinct features that characterize the target doma...

We propose a new approach denoted BNDI (Bayesian Network for Document Indexing) for indexing biomedical documents with controlled biomedical vocabulary based on a Bayesian Network. BNDI uses the probability inference to extract descriptors from the document. The main contribution of our approach is that it takes into account the structure of biomedical terminologies specially The Medical Subject Heading (MeSH) thesaurus. We experimented our approach on a subset of the OSHUMED collection.

In this paper, we present an application using the SUMMA-LSA platform developed by Baier, Lehnard, Hoffmann & Schneider (this volume). SUMMA-LSA was used to evaluate biology knowledge of 7th and 8th grades students dealing with The human body energy requirements . Student knowledge has been measured by means of classical and "evidential" multiple choices questions (MCQs) as well as open questions. The highly significant correlations between student answers and LSA cosine values opened encouraging perspectives in the development of e-learning systems that allows self-evaluations in real time and adaptation to learners' knowledge. R ÉSUM É. Dans cet article, nous présentons une application utilisant la plateforme SUMMA de développement et de test de l'Analyse de la Sémantique Latente développée par Baier, Lenhard, Hoffmann & Schneider (ce numéro). SUMMA-LSA a été utilisé pour évaluer les connaissances d'élèves de 5 • et de 4 • dans le domaine de la biologie, le fonctionnement du corps humain et ses besoins en énergie , à l'aide de QCM classiques et évidentiels et de questions Studia Informatica Universalis. ouvertes. Les corrélations hautement significatives entre les performances des élèves et les valeurs de cosinus fournies par LSA ouvrent des perspectives prometteuses dans le développement de systèmes d'apprentissage en ligne, adaptés au niveau de connaissances des apprenants, et permettant des (auto)-évaluations en temps réel.

In text mining, information retrieval, and machine learning, text documents are commonly represented through variants of sparse Bag of Words (sBoW) vectors (e.g. TF-IDF [1]). Although simple and intuitive, sBoW style representations suffer from their inherent over-sparsity and fail to capture word-level synonymy and polysemy. Especially when labeled data is limited (e.g. in document classification), or the text documents are short (e.g. emails or abstracts), many features are rarely observed within the training corpus. This leads to overfitting and reduced generalization accuracy. In this paper we propose Dense Cohort of Terms (dCoT), an unsupervised algorithm to learn improved sBoW document features. dCoT explicitly models absent words by removing and reconstructing random sub-sets of words in the unlabeled corpus. With this approach, dCoT learns to reconstruct frequent words from co-occurring infrequent words and maps the high dimensional sparse sBoW vectors into a low-dimensional dense representation. We show that the feature removal can be marginalized out and that the reconstruction can be solved for in closed-form. We demonstrate empirically, on several benchmark datasets, that dCoT features significantly improve the classification accuracy across several document classification tasks.

We present an efficient document representation learning framework, Document Vector through Corruption (Doc2VecC). Doc2VecC represents each document as a simple average of word embeddings. It ensures a representation generated as such captures the semantic meanings of the document during learning. A corruption model is included, which introduces a data-dependent regularization that favors informative or rare words while forcing the embeddings of common and non-discriminative ones to be close to zero. Doc2VecC produces significantly better word embeddings than Word2Vec. We compare Doc2VecC with several state-of-the-art document representation learning algorithms. The simple model architecture introduced by Doc2VecC matches or out-performs the state-of-the-art in generating high-quality document representations for sentiment analysis, document classification as well as semantic relatedness tasks. The simplicity of the model enables training on billions of words per hour on a single machine. At the same time, the model is very efficient in generating representations of unseen documents at test time.

Feature selection is an essential preprocessing step for classifiers with high dimensional training sets. In pattern recognition, feature selection improves the performance of classification by reducing the feature space but preserving the classification capabilities of the original feature space. Image classification using frequent approximate subgraph mining (FASM) is an example where the benefits of features selections are needed. This is due using frequent approximate subgraphs (FAS) leads to high dimensional representations. In this paper, we explore the use of feature selection algorithms in order to reduce the representation of an image collection represented through FASs. In our results we report a dimensionality reduction of over 50% of the original features and we get similar classification results than those reported by using all the features.

In recent years, kNN algorithm is paid attention by many researchers and is proved one of the best text categorization algorithms. Text categorization is according to training set which is assigned class label to decide a new document which is not assigned class label belongs to some kind of document. Until now, kNN algorithm has still some issues to need to study further. Such as: improvement of decision rule; selection of k value; selection of dimensions (i.e. feature set selection); problems of multiclass text categorization; the algorithm's executive efficiency (time and space) etc. In this paper, we mainly focus on improvement of decision rule and dimension selection. We design an adaptive fuzzy kNN text classifier. Here the adaptive indicate the adaptive of dimension selection. The experiment results show that our algorithm is effective and feasible.

Frequent Itemset Mining (FIM) is a critical data mining operation, but the computational overhead increases in the presence of large data sets and deteriorates the efficiency of FIM computation. In this paper, we explore algorithms based on CPU, GPU, and FPGA to take advantage of the parallelism in FIM computation. However, because of data transfer and memory overhead, these methods can bring negative performance. It is crucial to select the best algorithm for the particular hardware. We analyze five different methods to tackle these problems. GMiner [1] modifies Apriori algorithm [2] to minimize the memory requirements while using GPUs. To deal with exa-scale data mining operations, SGMiner [3] uses a combination of GPU and SSD to counter the memory requirements. GFPG-LLMA [4] modifies the FP-Growth algorithm [5] to make it more suitable for use on a GPU by addressing attributes of the algorithm that make it memory latency intensive. GSS [6] focuses on reducing interblock communication within GPU. The FPGA-based approaches such as Compact Architecture and Extended Architecture [7] are modifications over Eclat algorithm [8], focusing on optimizing resource utilization and speedup respectively. We observe that there is a trade-off between speedup, memory requirement, and complexity thus it is important to choose the hardware and algorithm based on the use case.

Recently the development of high performance, automatic text classification as well as feature selection has very challenging and tedious process because of many problems have been still unresolved. This paper provide the work done so far for the development of most efficient technique on the feature selection and document classification using support vector machine (SVM). Feature selection improves the efficiency and accuracy of text classification algorithm by removing redundant and irrelevant terms. Instead of the work done yet there is need to achieve huge level accuracy by reducing the computational time.

The concept of example credibility evaluates how much a classifier can trust an example when building a classification model. It is given by a credibility function, which is application dependent and estimated according to a series of factors that influence the credibility of the examples. Here we deal with automatic document classification and study the credibility of a document according to three factors: content, authorship and citations. We propose a genetic programming algorithm to estimate the credibility of training examples, and then add this estimation to a credibility-aware classifier. For that, we model the authorship and citation data as a complex network, and select a set of structural metrics that can be used to estimate credibility. These metrics are then merged with other content-related ones, and used as terminals for the GP. The GP was tested in a subset of the ACM-DL, and results showed that the credibility-aware classifier obtained results of micro and macroF1 from 5% to 8% better than the traditional classifiers.

Classification algorithms usually assume that any example in the training set should contribute equally to the classification model being generated. However, this is not always the case. This paper shows that the contribution of an example to the classification model varies according to many factors, which are application dependent, and can be estimated using what we call a credibility function. The credibility of an entity reflects how much value it aggregates to a task being performed, and here we investigate it in Automatic Document Classification, where the credibility of a document relates to its terms, authors, citations, venues, time of publication, among others. After introducing the concept of credibility in classification, we investigate how to estimate a credibility function using information regarding documents content, citations and authorship using mainly metrics previously defined in the literature. As the credibility of the content of a document can be easily mapped to any other classification problem, in a second phase we focus on content-based credibility functions. We propose a genetic programming algorithm to estimate this function based on a large set of metrics generally used to measure the strength of term-class relationship. The proposed and evolved credibility functions are then incorporated to the Naïve Bayes classifier, and applied to four text collections, namely ACM-DL, Reuters, Ohsumed, and 20 Newsgroup. The results obtained showed significant improvements in both micro-F1 and macro-F1, with gains up to 21% in Ohsumed when compared to the traditional Naïve Bayes.

Adverse Drug Reaction (ADR) extraction is the process of identifying drug implications mentioned in social posts. Handling medical text for the identification of ADR is vital to research in terms of configuring the side effect and other medical-related entities within any medical text. However, investigating the role of such effect in the context of positive and negative is the responsibility of sentiment classification task where every medical review document would be categorized into its polarity, this is known as Medical Sentiment Analysis (MSA). Several studies have presented various techniques for MSA. Most of the recent studies have concentrated on architectures such as the Convolutional Neural Network (CNN) to get the document embedding. Yet, such architecture focuses only on the input without considering the previous or latter input. This might lead to weaker embedding for the document where some terms would not be considered. Hence, this paper proposes a new document embedding approach based on the Recurrent Neural Network (RNN) to improve the sentiment classification. Using a benchmark dataset of medical sentiments, the proposed method showed greater performance of sentiment classification accuracy. Such finding proves the effectiveness of RNN in producing document embedding.

In the last decade, a variety of topic models have been proposed for text engineering. However, except Probabilistic Latent Semantic Analysis (PLSA) and Latent Dirichlet Allocation (LDA), most of existing topic models are seldom applied or considered in industrial scenarios. This phenomenon is caused by the fact that there are very few convenient tools to support these topic models so far. Intimidated by the demanding expertise and labor of designing and implementing parameter inference algorithms, software engineers are prone to simply resort to PLSA/LDA, without considering whether it is proper for their problem at hand or not. In this paper, we propose a configurable topic modeling framework named Familia, in order to bridge the huge gap between academic research fruits and current industrial practice. Familia supports an important line of topic models that are widely applicable in text engineering scenarios. In order to relieve burdens of software engineers without knowledge of Bayesian networks, Familia is able to conduct automatic parameter inference for a variety of topic models. Simply through changing the data organization of Familia, software engineers are able to easily explore a broad spectrum of existing topic models or even design their own topic models, and find the one that best suits the problem at hand. With its superior extendability, Familia has a novel sampling mechanism that strikes balance between effectiveness and efficiency of parameter inference. Furthermore, Familia is essentially a big topic modeling framework that supports parallel parameter inference and distributed parameter storage. After being open-sourced at Github, Familia rapidly becomes the second most popular project in topic model area, and is widely used in both industrial community and academia. The utilities and necessity of Familia are demonstrated in real-life industrial applications. Familia would significantly enlarge software engineers' arsenal of topic models and pave the way for utilizing highly customized topic models in real-life problems.

Credit card fraud detection remains a significant challenge for financial institutions and consumers globally, prompting the
adoption of advanced data analytics and machine learning techniques. In this study, we investigate the methodology and
performance evaluation of various machine learning algorithms for credit card fraud detection, emphasizing data preprocessing
techniques and model effectiveness. Through thorough dataset analysis and experimentation using cross-validation approaches,
we assess the performance of logistic regression, decision trees, random forest classifiers, Naïve Bayes classifiers, K-nearest
neighbors (KNN), and artificial neural networks (ANN-DL). Key performance metrics such as accuracy, sensitivity, specificity, and
F1-score are compared to identify the most effective models for detecting fraudulent transactions. Additionally, we explore the
impact of different folds in cross-validation on model performance, providing insights into the classifiers' robustness and
stability. Our findings contribute to the ongoing efforts to develop efficient fraud detection systems, offering valuable insights
for financial institutions and researchers striving to combat credit card fraud effectively.