Text Normalization Research Articles

Enhancing the Fake News Classification Model Using Find-Tuning Approach

In recent years, fake news on social media has become a pressing concern, posing significant threats to individuals, organizations, and society as a whole. we present a novel strategy to enhance the accuracy of fake news classification models through fine-tuning. Our proposed model involves adding new layers and freezing some layers in the BEART model, resulting in improved performance. To facilitate our research, we constructed a comprehensive fake news dataset by combining real and fake datasets obtained from secondary sources. Firstly, the dataset underwent rigorous pre-processing, including data cleaning, text normalization, tokenization, stop word removal, and other techniques, ultimately enabling binary classification. Subsequently, the proposal model (DCNN) was trained on this dataset to classify news articles as either real or fake. Notably, experimental results demonstrate that our approach outperforms several recent studies in detecting fake news, achieving high accuracy. To evaluate the effectiveness of our proposed model, we employed various evaluation methods. Firstly, we utilized the Tag Cloud technique, which visually represents the most frequently used words in the text or documents, enabling us to distinguish between real and fake news. Additionally, we employed the Classification report, which provides precision, F1, recall, and support scores, to comprehensively assess the model's performance. Furthermore, we employed the confusion matrix, a tabular layout that effectively visualizes the classification algorithm's performance, thereby enabling a clear interpretation based on known true values. Therefore, the proposed model was trained to classify news articles as real or fake, and the experiments on the dataset show that this approach performs better than several recent studies for detecting fake news, achieving high accuracy.

More extreme duplication in FDA Adverse Event Reporting System detected by literature reference normalization and fuzzy string matching.

Literature reports of adverse drug events can be replicated across multiple companies, resulting in extreme duplication (defined as a majority of reports being duplicates) in the FDA Adverse Event Reporting System (FAERS) database because they can escape legacy duplicate detection algorithms routinely deployed on that data source. Literature reference field, added to in 2014, could potentially be utilized to identify replicated reports. FAERS does not enforce adherence to the Vancouver referencing convention, thus the same article may be referenced differently leading to duplication. The objective of this analysis is to determine if variations of the same literature references observed in FAERS can be resolved with text normalization and fuzzy string matching. We normalized the literature references recorded in the FAERS database through the first quarter of 2021 with a rule-based algorithm so that they better conform to the Vancouver convention. Levenshtein distance was then utilized to merge sufficiently similar normalized literature references together. Normalization of literature references increases the percentage that can be parsed into author, title, and journal from 61.74% to 93.93%. We observe that about 98% of pairs within groups do have a Levenshtein similarity of the title above the threshold. The extreme duplication ranged from 66% to 87% with a median of 72% of reports being duplicates and often involved addictovigilance scenarios. We have shown that these normalized references can be merged via fuzzy string matching to improve enumeration of all the individual case safety reports that refer to the same article. Inclusion of the PubMed ID and adherence to the Vancouver convention could facilitate identification of duplicates in the FAERS dataset. Awareness of this phenomenon may improve disproportionality analysis, especially in areas such as addictovigilance.

Usable Amharic text corpus for natural language processing applications

In this paper, we describe the preparation of a usable Amharic text corpus for different Natural Language Processing (NLP) applications. Natural language applications, such as document classification, topic modeling, machine translation, speech recognition, and others, suffer greatly from a lack of digital resources. This is especially true for Amharic, a resource-constrained, morphologically rich, and complex language. In response to this, a total of 67,739 Amharic news documents consisting of 8 different categories from online sources are collected. The collected corpus passes through a number of pre-processing steps including; data cleaning, text normalization and punctuation correction. To validate the usability of the collected corpora from different domains, a baseline document classification experiment was conducted. Experimental results show that, 84.53% accuracy is registered using deep learning in the absence of linguistic information. Finding indicated that it is possible to use the prepared corpora for different natural language applications in the absence of linguistic resources such as stemmer and dictionary despite the complexity of Amharic language. We are further working towards Amharic news document classification by incorporating a linguistic independent stop-word detection, stemming and unsupervised morphological segmentation of Amharic documents.

Transformer-Based Joint Learning Approach for Text Normalization in Vietnamese Automatic Speech Recognition Systems

In this article, we investigate the task of normalizing transcribed texts in Vietnamese Automatic Speech Recognition (ASR) systems in order to improve user readability and the performance of downstream tasks. This task usually consists of two main sub-tasks: predicting and inserting punctuation (i.e., period, comma); and detecting and standardizing named entities (i.e., numbers, person names) from spoken forms to their appropriate written forms. To achieve these goals, we introduce a complete corpus including of 87,700 sentences and investigate conditional joint learning approaches which globally optimize two sub-tasks simultaneously. The experimental results are quite promising. Overall, the proposed architecture outperformed the conventional architecture which trains individual models on the two sub-tasks separately. The joint models are furthered improved when integrated with the surrounding contexts (SCs). Specifically, we obtained 81.13% for the first sub-task and 94.41% for the second sub-task in the F1 scores using the best model.

Local search five‐element cycle optimized reLU‐BiLSTM for multilingual aspect‐based text classification

SummaryAspect‐based sentiment analysis has gained wide popularity due to its benefits of text extraction, classification, and ranking the overall sentiments of each feature extracted. However, the aspect‐based feature extraction techniques often result in acquiring more number aspects that refer to the same feature which arises the need for aspect‐based text classification. Since most of the existing techniques focus on monolingual aspect‐based sentimental analysis, we planned to develop a multilingual aspect‐based text classification for Indian languages. We perform the multilingual aspect‐based text classification on different morphologically rich and complex languages such as Hindi, Tamil, Malayalam, Bengali, Urdu, Telugu, and Sinhalese. To achieve this objective, in this article we present an optimized rectified linear unit (reLU) layer‐based bidirectional long short‐term memory (reLU‐BiLSTM) deep learning tool is developed. The parameters of the reLU‐BiLSTM architecture are optimized using the local search‐based five‐element cycle optimization algorithm (LSFECO) optimization algorithm. Initially, the proposed model preprocesses the multilingual texts obtained from the reviews using different techniques such as tokenization, special character removal, text normalization and so forth. The discrete and categorical features from the different languages are initially extracted by applying the bidirectional encoder representations from transformers (BERT) model which processes the sentences in the text in a layer‐by‐layer manner. The context learning and word embeddings (aspects) present in the text are identified using different approaches such as word mover's distance, continuous Bag‐of‐Words (CBOW), and Cosine similarity. The LSFECO optimized reLU‐BiLSTM architecture classifies the different aspects present in the embedding document to its corresponding classes (flowers, plants, animals, sports, politics, etc). The efficiency of the proposed methodology is evaluated using the text obtained from different text documents such as semantic relations from Wikipedia, Habeas Corpus (HC) Corpora, Sentiment Lexicons for 81 Languages, IIT Bombay English‐Hindi Parallel Corpus, and Indic Languages Multilingual Parallel Corpus. When compared to conventional techniques, the proposed methodology outperforms them in terms of entropy, coverage, purity, processing time, accuracy, F1‐score, recall, and precision.

Comparison of LSTM, SVM, and naive bayes for classifying sexual harassment tweets

Twitter is now a very open and extensive social media; anyone can freely express their opinion on any topic on social media. The content or discussion on Twitter is also quite diverse and unlimited. However, because it is unlimited, many misuse it for negative things. One of them is verbal sexual harassment through Twitter. This research aims to identify sexual harassment in an Indonesian tweet using sentiment analysis using the LSTM, SVM, and naive bayes methods with text normalization. In this study, 2990 tweets in the Indonesian language were tested from 4th to 6th in May 2022. The Twitter data shows that tweets included in sexual harassment are more than those not included in sexual harassment, totaling 2026 data. From the results of the evaluation of tweet data classification using text normalization with LSTM, the accuracy is 84.62%, SVM is 86.54%, and naive bayes is 85.45%. Using the SVM algorithm with text normalization gets the highest accuracy compared to LSTM and naive bayes in classifying Indonesian sexual harassment tweets.

A NAIVE BAYESIAN CLASSIFIER FOR NORMALIZATION OF TEXT: A CASE STUDY FOR K AZAKH LANGUAGE

The amount of complicated documents and texts has increased exponentially in recent years, necessitating a deeper understanding of machine learning technologies in order to effectively identify texts in numerous applications. Text normalization is one of the best decisions. It is the reduction of all words of the text to the original form. This paper investigates a layered strategy for fixing mistakes in Kazakh language literature downloaded from the Internet. Because of the widespread use of social media as a source for linguistic study, error correction is a critical issue. The goal of this research was to look at the current Naive Bayes algorithm in English, as well as the normalization of words and sentences in natural languages, in order to create a similar algorithm for the Kazakh language. The method of morphology of Kazakh words and their difference from English was considered suitable for processing words in a dictionary. As a result of the normalization system, the efficiency of this method for the Kazakh language was proved. Keywords: text normalization, Naïve Bayes algorithm, natural language, processing of text, classifier

Human–machine collaboration in transcription

As automatic speech recognition (ASR) has improved, it has become a viable tool for content transcription. Prior to the use of ASR for this task, content transcription was achieved through human effort alone. Despite improvements, ASR performance is as yet imperfect, especially in more challenging conditions (eg multiple speakers, noise, nonstandard accents). Given this, a promising way forward is a human-in-the-loop (HIL) approach. This contribution describes our work with HIL ASR on the transcription task. Traditionally, ASR performance has been measured using word error rate (WER). This measure may not be sufficient to describe the full set of errors that a speech-to-text (STT) pipeline designed for transcription can make, such as those involving capitalisation, punctuation, and inverse text normalisation (ITN). It is therefore the case that improved WER does not always lead to increased productivity, and the inclusion of ASR in HIL may adversely affect productivity if it contains too many errors. Rev.com provides a convenient laboratory to explore these questions. Originally, the company provided transcriptions of audio and video content executed solely by humans (known as Revvers). More recently, ASR was introduced in an HIL workflow where Revvers postedited an ASR first draft. We provide an analysis of the interaction between metrics of ASR accuracy and the productivity of our 72,000+ Revvers transcribing more than 15,000 hours of media every week. To do this, we utilise two measures of transcriptionist productivity: transcriber real time factor (RTF) and words per minute (WPM). Through our work, we hope to focus attention on the human productivity and quality of experience (QoE) aspects of improvements in ASR and related technologies. Given the broad scope of content transcription applications and the still elusive objective of perfect machine performance, keeping the human in the loop in both practice and mind is critical. This paper provides an overview of human and machine transcription and Rev’s marketplace, followed by an analysis of the relationship between ASR accuracy and transcriptionist productivity, and concludes with suggestions for future work.

Arabic Aspect-Based Sentiment Classification Using Seq2Seq Dialect Normalization and Transformers

Sentiment analysis is one of the most important fields of natural language processing due to its wide range of applications and the benefits associated with using it. It is defined as identifying the sentiment polarity of natural language text. Researchers have recently focused their attention on Arabic SA due to the massive amounts of user-generated content on social media and e-commerce websites in the Arabic world. Most of the research in this fieldwork is on the sentence and document levels. This study tackles the aspect-level sentiment analysis for the Arabic language, which is a less studied version of SA. Because Arabic NLP is challenging and there are few available Arabic resources and many Arabic dialects, limited studies have attempted to detect aspect-based sentiment analyses on Arabic texts. Specifically, this study considers two ABSA tasks: aspect term polarity and aspect category polarity, using the text normalization of the Arabic dialect after making the classification task. We present a Seq2Seq model for dialect normalization that can serve as a pre-processing step for the ABSA classification task by reducing the number of OOV words. Thus, the model’s accuracy increased. The results of the conducted experiments show that our models outperformed the existing models in the literature on both tasks and datasets.

Effective Preprocessing and Normalization Techniques for COVID-19 Twitter Streams with POS Tagging via Lightweight Hidden Markov Model

The major focus of this research work is to refine the basic preprocessing steps for the unstructured text content and retrieve the potential conceptual features for further enhancement processes such as semantic enrichment and named entity recognition. Although some of the preprocessing techniques such as text tokenization, normalization, and Part-of-Speech (POS) tagging work exceedingly well on formal text, it has not performed well when it is applied into informal text such as tweets and short messages. Hence, we have given the enhanced text normalization techniques to reduce the complexity persist over the twitter streams and eliminate the overfitting issues such as text anomalies and irregular boundaries while fixing the grammar of the text. The hidden Markov model (HMM) has been pervasively used to extract the core lexical features from the Twitter dataset and suitably adapt the external documents to supplement the extraction techniques to complement the tweet context. Using this Markov process, the POS tags are identified as states of the Markov process, and words are the desired results of the model. As this process is very crucial for the next stage of entity extraction and classification, the effective handling of informal text is considered to be important and therefore proposed the most effective hybrid approach to deal with the issues appropriately.

Real-Time Text Classification of User-Generated Content on Social Media: Systematic Review

The aim of this systematic review is to determine the current state of the art in the real-time classification of user-generated content from social media. Focus is on the identification of the main characteristics of data used for training and testing, the types of text processing and normalization that are required, the machine learning methods used most commonly, and how these methods compare to one another in terms of classification performance. Relevant studies were selected from subscription-based digital libraries, free-to-access bibliographies, and self-curated repositories and then screened for relevance with key information extracted and structured against the following facets: natural language processing (NLP) methods, data characteristics, classification methods, and evaluation results. A total of 25 studies published between 2014 and 2018 covering 15 types of classification algorithms were included in this review. Support vector machines (SVMs), Bayesian classifiers, and decision trees were the most commonly employed algorithms with recent emergence of neural network approaches. Domain-specific, application programming interface (API)-driven collection is the most prevalent origin of datasets. The reuse of previously published datasets as a means of benchmarking algorithms against other studies is also prevalent. In conclusion, there are consistent approaches taken when normalizing social media data for text mining and traditional text mining techniques are suited to the task of real-time analysis of social media.

Online product sentiment analysis using random evolutionary whale optimization algorithm and deep belief network

• Image and text normalization techniques are used to improve the data quality. • Used MLBP , SURF, TF-IDF and LSA techniques for feature extraction. • Used RE-WOA to reduce the dimension of the extracted feature vectors. • Used DBN classifier to classify the user's sentiments of the online products. In the recent decades, the online product sentiment analysis is an emerging research topic that assists the customers to take better decisions on purchasing the products and to achieve better sales of the products. Recently, several machine learning techniques are experimented on many datasets for analyzing the customer's sentiments through online portals. Still, the customers are struggling to obtain the aspect sentiments expressed by other customers, particularly in the amazon websites. Therefore, a novel automated model is proposed in this manuscript for an effective online product sentiment analysis. After collecting the multimodal data from the amazon websites, the image and data normalization techniques are employed for better understanding of the collected data. Further, the feature extraction is performed by utilizing Latent Semantic Analysis (LSA), Term Frequency- Inverse Document Frequency (TF-IDF), Modified Local Binary Pattern (MLBP), and Speeded Up Robust Features (SURF) descriptors for extracting the textual and visual feature vectors from the preprocessed data. Finally, the Random Evolutionary Whale Optimization Algorithm (REWOA) and Deep Belief Network (DBN) classifier are integrated for feature vector optimization and sentiment classification. By using feature optimization, the system complexity and running time of the classifier is improved. The experimental investigation states that the developed REWOA-DBN model achieved 96.86% of classification accuracy, which is better compared to other optimizers and classifiers

Graph-based Turkish text normalization and its impact on noisy text processing

User generated texts on the web are freely-available and lucrative sources of data for language technology researchers. Unfortunately, these texts are often dominated by informal writing styles and the language used in user generated content poses processing difficulties for natural language tools. Experienced performance drops and processing issues can be addressed either by adapting language tools to user generated content or by normalizing noisy texts before being processed. In this article, we propose a Turkish text normalizer that maps non-standard words to their appropriate standard forms using a graph-based methodology and a context-tailoring approach. Our normalizer benefits from both contextual and lexical similarities between normalization pairs as identified by a graph-based subnormalizer and a transformation-based subnormalizer. The performance of our normalizer is demonstrated on a tweet dataset in the most comprehensive intrinsic and extrinsic evaluations reported so far for Turkish. In this article, we present the first graph-based solution to Turkish text normalization with a novel context-tailoring approach, which advances the state-of-the-art results by outperforming other publicly available normalizers. For the first time in the literature, we measure the extent to which the accuracy of a Turkish language processing tool is affected by normalizing noisy texts before being processed. An analysis of these extrinsic evaluations that focus on more than one Turkish NLP task (i.e., part-of-speech tagger and dependency parser) reveals that Turkish language tools are not robust to noisy texts and a normalizer leads to remarkable performance improvements once used as a preprocessing tool in this morphologically-rich language.

Comparison of text preprocessing methods

AbstractText preprocessing is not only an essential step to prepare the corpus for modeling but also a key area that directly affects the natural language processing (NLP) application results. For instance, precise tokenization increases the accuracy of part-of-speech (POS) tagging, and retaining multiword expressions improves reasoning and machine translation. The text corpus needs to be appropriately preprocessed before it is ready to serve as the input to computer models. The preprocessing requirements depend on both the nature of the corpus and the NLP application itself, that is, what researchers would like to achieve from analyzing the data. Conventional text preprocessing practices generally suffice, but there exist situations where the text preprocessing needs to be customized for better analysis results. Hence, we discuss the pros and cons of several common text preprocessing methods: removing formatting, tokenization, text normalization, handling punctuation, removing stopwords, stemming and lemmatization, n-gramming, and identifying multiword expressions. Then, we provide examples of text datasets which require special preprocessing and how previous researchers handled the challenge. We expect this article to be a starting guideline on how to select and fine-tune text preprocessing methods.

Boring Problems Are Sometimes the Most Interesting

Abstract In a recent position paper, Turing Award Winners Yoshua Bengio, Geoffrey Hinton, and Yann LeCun make the case that symbolic methods are not needed in AI and that, while there are still many issues to be resolved, AI will be solved using purely neural methods. In this piece I issue a challenge: Demonstrate that a purely neural approach to the problem of text normalization is possible. Various groups have tried, but so far nobody has eliminated the problem of unrecoverable errors, errors where, due to insufficient training data or faulty generalization, the system substitutes some other reading for the correct one. Solutions have been proposed that involve a marriage of traditional finite-state methods with neural models, but thus far nobody has shown that the problem can be solved using neural methods alone. Though text normalization is hardly an “exciting” problem, I argue that until one can solve “boring” problems like that using purely AI methods, one cannot claim that AI is a success.

Performance Comparison of Similarity Measure Algorithm as Data Preprocessing Stage: Text Normalization in Bahasa

Purpose: More and more data are stored in text form due to technological developments, making text data processing more difficult. It also causes problems in the text preprocessing algorithm, one of which is when two texts are identical, but are considered distinct by the algorithm. Therefore, it is necessary to normalize the text to get the standard form of words in a particular language. Spelling correction is often used to normalize text, but for Bahasa Indonesia, there has not been much research on the spell correction algorithm. Thus, there needs to be a comparison of the most appropriate spelling correction algorithms for the normalization process to be effective.Methods: In this study, we compared three algorithms, namely Levenshtein Distance, Jaro-Winkler Distance, and Smith-Waterman. These algorithms were evaluated using questionnaire data and tweet data, which both are in Bahasa Indonesia.Result: The fastest normalization time is obtained by the Jaro-Winkler, taking an average of 31.01 seconds for questionnaire data and 59.27 seconds for tweet data. The best accuracy is obtained by the Levenshtein Distance with a value of 44.90% for the questionnaire data and 60.04% for the tweet data. Novelty: The novelty of this research is to compare the similarity measure algorithm in Bahasa Indonesia. Therefore, the most suitable similarity measure algorithm for Bahasa Indonesia will be obtained.

Towards automatic annotation of collaborative problem‐solving skills in technology‐enhanced environments

AbstractBackgroundCollaborative problem solving (CPS) is important for success in the 21st century, especially for teamwork and communication in technology‐enhanced environments. Measurement of CPS skills has emerged as an essential aspect in educational assessment. Modern research in CPS relies on theory‐driven measurements that are usually carried out as manual annotations over recorded logs of collaborative activities. However, manual annotation has limited scalability and is not conductive towards CPS assessments at scale.ObjectiveWe explore possibilities for automated annotation of actions in collaborative‐teams, especially chat messages. We evaluate two approaches that employ machine learning for automated classification of CPS events.MethodData were collected from engineering, physics and electronics students' participation in a simulation‐based task on electronics concepts, in which participants communicated via text‐chat messages. All task activities were logged and time stamped. Data have been manually classified for the CPS skills, using an ontology that includes both social and cognitive dimensions. In this article, we describe computational linguistic methods for automatically classifying the CPS skills from logged data, with a view towards automating CPS assessments.ResultsWe applied two machine learning methods to our data. A Naïve Bayes classifier has been previously used in CPS research, but it is only moderately successful on our data. We also present a k‐nearest‐neighbours (kNN) classifier that uses distributional semantic models for measuring text similarity. This classifier shows strong agreement between automated and human annotations. The study also demonstrates that automatic spelling correction and slang normalization of chat texts are useful for accurate automated annotation.ImplicationsOur results suggest that a kNN classifier can be very effective for accurate annotation of CPS events. It achieves reasonably strong results even when trained on only half of the available data. This shows a promise towards reduction of manual data annotation for CPS measurement.

Sentiment Analysis using Improved Novel Convolutional Neural Network (SNCNN)

Sentiment Analysis is an important method in which many researchers are working on the automated approach for extraction and analysis of huge volumes of user achieved data, which are accessible on social networking websites. This approach helps in analyzing the direct falls under the domain of SA. SA comprises the vast field of effective classification of user-initiated text under defined polarities. The proposed work includes four major steps for solving these issues: the first step is preprocessing which holds tokenization, stop word removal, stemming, cleaning up of unwanted text information like removing of Ads from Web pages, Text normalization for converting binary format. Secondly, the Feature extraction is based on the Bag words, Word2Vec and TF-ID which is a Term Frequency-Inverse Document Frequency. Thirdly, this feature selection includes the procedure for examining semantic gaps along with source features using teaching models and this involves target task characteristic application for Improved Novel Convolutional Neural Network (INCNN). The Feature Selection accompanies the procedure of Information Gain (IG) and PCC which is a Pearson Correlation Coefficient. Finally, the classification step INCNN gives out sentiment posts and responses for the user-based post aspects which helps in enhancing the system performance. The experimental outcome proposes the INCNN algorithm and provides higher performance rather than the existing approach. The proposed INCNN classifier results in highest accuracy.

Gilda: biomedical entity text normalization with machine-learned disambiguation as a service.

Gilda is a software tool and web service that implements a scored string matching algorithm for names and synonyms across entries in biomedical ontologies covering genes, proteins (and their families and complexes), small molecules, biological processes and diseases. Gilda integrates machine-learned disambiguation models to choose between ambiguous strings given relevant surrounding text as context, and supports species-prioritization in case of ambiguity. The Gilda web service is available at http://grounding.indra.bio with source code, documentation and tutorials available via https://github.com/indralab/gilda. Supplementary data are available at Bioinformatics Advances online.

Analysis of the Full-Size Russian Corpus of Internet Drug Reviews with Complex NER Labeling Using Deep Learning Neural Networks and Language Models

The paper presents the full-size Russian corpus of Internet users’ reviews on medicines with complex named entity recognition (NER) labeling of pharmaceutically relevant entities. We evaluate the accuracy levels reached on this corpus by a set of advanced deep learning neural networks for extracting mentions of these entities. The corpus markup includes mentions of the following entities: medication (33,005 mentions), adverse drug reaction (1778), disease (17,403), and note (4490). Two of them—medication and disease—include a set of attributes. A part of the corpus has a coreference annotation with 1560 coreference chains in 300 documents. A multi-label model based on a language model and a set of features has been developed for recognizing entities of the presented corpus. We analyze how the choice of different model components affects the entity recognition accuracy. Those components include methods for vector representation of words, types of language models pre-trained for the Russian language, ways of text normalization, and other pre-processing methods. The sufficient size of our corpus allows us to study the effects of particularities of annotation and entity balancing. We compare our corpus to existing ones by the occurrences of entities of different types and show that balancing the corpus by the number of texts with and without adverse drug event (ADR) mentions improves the ADR recognition accuracy with no notable decline in the accuracy of detecting entities of other types. As a result, the state of the art for the pharmacological entity extraction task for the Russian language is established on a full-size labeled corpus. For the ADR entity type, the accuracy achieved is 61.1% by the F1-exact metric, which is on par with the accuracy level for other language corpora with similar characteristics and ADR representativeness. The accuracy of the coreference relation extraction evaluated on our corpus is 71%, which is higher than the results achieved on the other Russian-language corpora.