Search

Since the emergence of deep learning-based chatbots for knowledge services, numerous research and development projects have been conducted in various industries. A high demand for chatbots has drastically increased the global market size; however, the limited functional scalability of open-domain chatbots is a challenge to their application to industries. Moreover, as most chatbot frameworks employ English, it is necessary to create chatbots customized for other languages. To address this problem, this paper proposes KoRASA as a pipeline-optimization method, which uses a deep learning-based open-source chatbot framework to understand the Korean language. KoRASA is a closed-domain chatbot that is applicable across a wide range of industries in Korea. KoRASA’s operation consists of four stages: tokenization, featurization, intent classification, and entity extraction. The accuracy and F1-score of KoRASA were measured based on datasets taken from common tasks carried out in most industrial fields. The algorithm for intent classification and entity extraction was optimized. The accuracy and F1-score were 98.2% and 98.4% for intent classification and 97.4% and 94.7% for entity extraction, respectively. Furthermore, these results are better than those achieved by existing models. Accordingly, KoRASA can be applied to various industries, including mobile services based on closed-domain chatbots using Korean, robotic process automation (RPA), edge computing, and Internet of Energy (IoE) services.

One of the solutions to the problem of spatio-temporal data anisotropy is the use of a multilevel system of digital twins based on the corresponding industry models and the updated archive data base. The application of this approach has successfully proved itself in information systems for monitoring the parameters of the geomagnetic field and its variations, providing spatio-temporal interpolation of geomagnetic data with an accuracy of 0.81 nT in magnetically quiet periods. At the same time, the problem of information interaction between the levels of the system of digital twins remained unresolved, which is greatly aggravated by the constantly growing volume of data and their heterogeneous nature. The paper proposes a solution to the indicated problem by means of a formalized mechanism for packaging space-time information, in which the identification of data sources is performed on the basis of a hierarchical binary tokenization system. In addition, the proposed software implementation of such an approach is considered, a distinctive feature of which is the combination of traditional clientserver and innovative serverless architectures to implement a highly loaded reactive web application for working with analyzed data. The main stages of the implementation of information interaction are highlighted and programmatically formalized - from obtaining initial information from its sources to verifying data, analyzing them, processing and forming the output information flow of the system. The results of the computational experiments carried out on the example of the problem of monitoring the parameters of the Earth's magnetic field and its variations confirmed the effectiveness of the proposed solutions, expressed both in increasing the reactivity of web-based applications and in increasing the computational speed of formation and filling of information storages that aggregate information from distributed heterogeneous sources.

Machine Translation is an effort to bridge language barriers and misinterpretations, making communication more convenient through the automatic translation of languages. The quality of translations produced by corpus-based approaches predominantly depends on the availability of a large parallel corpus. Although machine translation of many Indian languages has progressively gained attention, there is very limited research on machine translation and the challenges of using various machine translation techniques for a low-resource language such as Mizo. In this article, we have implemented and compared statistical-based approaches with modern neural-based approaches for the English–Mizo language pair. We have experimented with different tokenization methods, architectures, and configurations. The performance of translations predicted by the trained models has been evaluated using automatic and human evaluation measures. Furthermore, we have analyzed the prediction errors of the models and the quality of predictions based on variations in sentence length and compared the model performance with the existing baselines.

Mobile advertising plays a vital role in the mobile app ecosystem. A major threat to the sustainability of this ecosystem is click fraud , i.e., ad clicks performed by malicious code or automatic bot problems. Existing click fraud detection approaches focus on analyzing the ad requests at the server side. However, such approaches may suffer from high false negatives since the detection can be easily circumvented, e.g., when the clicks are behind proxies or globally distributed. In this paper, we present AdSherlock, an efficient and deployable click fraud detection approach at the client side (inside the application) for mobile apps. AdSherlock splits the computation-intensive operations of click request identification into an offline procedure and an online procedure. In the offline procedure, AdSherlock generates both exact patterns and probabilistic patterns based on URL (Uniform Resource Locator) tokenization. These patterns are used in the online procedure for click request identification and further used for click fraud detection together with an ad request tree model. We implement a prototype of AdSherlock and evaluate its performance using real apps. The online detector is injected into the app executable archive through binary instrumentation. Results show that AdSherlock achieves higher click fraud detection accuracy compared with state of the art, with negligible runtime overhead.

Emotions detection in natural languages is very effective in analyzing the user's mood about a concerned product, news, topic, and so on. However, it is really a challenging task to extract important features from a burst of raw social text, as emotions are subjective with limited fuzzy boundaries. These subjective features can be conveyed in various perceptions and terminologies. In this article, we proposed an IoT-based framework for emotions classification of tweets using a hybrid approach of Term Frequency Inverse Document Frequency (TFIDF) and deep learning model. First, the raw tweets are filtered using the tokenization method for capturing useful features without noisy information. Second, the TFIDF statistical technique is applied to estimate the importance of features locally as well as globally. Third, the Adaptive Synthetic (ADASYN) class balancing technique is applied to solve the imbalance class issue among different classes of emotions. Finally, a deep learning model is designed to predict the emotions with dynamic epoch curves. The proposed methodology is analyzed on two different Twitter emotions datasets. The dynamic epoch curves are shown to show the behavior of test and train data points. It is proved that this methodology outperformed the popular state-of-the-art methods.

PurposeThe purpose of the paper is to study multiple viewpoints which are required to access the more informative similarity features among the tweets documents, which is useful for achieving the robust tweets data clustering results.Design/methodology/approachLet “N” be the number of tweets documents for the topics extraction. Unwanted texts, punctuations and other symbols are removed, tokenization and stemming operations are performed in the initial tweets pre-processing step. Bag-of-features are determined for the tweets; later tweets are modelled with the obtained bag-of-features during the process of topics extraction. Approximation of topics features are extracted for every tweet document. These set of topics features of N documents are treated as multi-viewpoints. The key idea of the proposed work is to use multi-viewpoints in the similarity features computation. The following figure illustrates multi-viewpoints based cosine similarity computation of the five tweets documents (here N = 5) and corresponding documents are defined in projected space with five viewpoints, say, v1,v2, v3, v4, and v5. For example, similarity features between two documents (viewpoints v1, and v2) are computed concerning the other three multi-viewpoints (v3, v4, and v5), unlike a single viewpoint in traditional cosine metric.FindingsHealthcare problems with tweets data. Topic models play a crucial role in the classification of health-related tweets with finding topics (or health clusters) instead of finding term frequency and inverse document frequency (TF–IDF) for unlabelled tweets.Originality/valueTopic models play a crucial role in the classification of health-related tweets with finding topics (or health clusters) instead of finding TF-IDF for unlabelled tweets.

PurposeBlockchain technology is increasingly positioned as a promising and disruptive technology. Such a promise has attracted companies to explore how blockchain technology can be used to gain significant benefits. Process models play a cardinal role when seeking to improve business processes as they are the foundation of process analysis and redesign. This paper examines how blockchain-oriented processes can be conceptually modelled with activity- (BPMN) and artifact-centric (CMMN) modelling paradigms.Design/methodology/approachThis paper discusses how commonly occurring patterns, specific to block-chain-based applications, can be modelled with BPMN and CMMN. Furthermore, the advantages and disadvantages of both notations for accurately representing blockchain-specific patterns are discussed.FindingsThe main finding of this paper is that neither BPMN nor CMMN can adequately and accurately represent certain patterns specific for blockchain-oriented processes. BPMN, while supporting most of the patterns, does not provide sufficient support to represent tokenization. CMMN, on the other hand, does not provide support to distinguish between activities executed and data stored on-chain versus off-chain.Originality/valueThe paper provides insight into the strengths and weaknesses of BPMN and CMMN for modelling processes to be supported by blockchain. This will serve to aid analysts to produce better process models for communication purposes and, thereby, facilitate development of blockchain-based solutions.

In recent years, the use of deep learning in language models has gained much attention. Some research projects claim that they can generate text that can be interpreted as human writing, enabling new possibilities in many application areas. Among the different areas related to language processing, one of the most notable in applying this type of modeling is programming languages. For years, the machine learning community has been researching this software engineering area, pursuing goals like applying different approaches to auto-complete, generate, fix, or evaluate code programmed by humans. Considering the increasing popularity of the deep learning-enabled language models approach, we found a lack of empirical papers that compare different deep learning architectures to create and use language models based on programming code. This paper compares different neural network architectures like Average Stochastic Gradient Descent (ASGD) Weight-Dropped LSTMs (AWD-LSTMs), AWD-Quasi-Recurrent Neural Networks (QRNNs), and Transformer while using transfer learning and different forms of tokenization to see how they behave in building language models using a Python dataset for code generation and filling mask tasks. Considering the results, we discuss each approach&rsquo, s different strengths and weaknesses and what gaps we found to evaluate the language models or to apply them in a real programming context.

Abstract—The number of hotel reviews are huge and growing day by day. Many travelers rely on this review to take down their decision to book a hotel. Gather valid and useful information from a huge amount of reviews are needs that must be met. Hence, a summarize tool for hotel review is built to create a representative summary. Hotel review data growing a trend of sentiment in a range of time due to the condition and improvement at that time, so the analysis of sentiment trend is done to choose the appropriate representative review data. Then, there are two method to summarize the selected review data. First, summary was obtained from extractive method with selecting most related sentence base on its Term Frequency-Inverse Document Frequency (TF-IDF) score. Second, phrase summary style is built by pairing adjective to the nearest noun and considering the polarity. From those method, obtained Recall-Oriented Understudy for Gisting Evaluation (ROUGE)-1 recall and Bilingual Evaluation Understudy (BLEU) score respectively {0.2101and 0.7820} for first method and {0.0670 and 0.03672} for the second method. All the reviews are crawled from TripAdvisor website and have been pre-processed by segmenting/tokenization, case folding, and tagging.

This paper attempts to analyze the Korean sentence classification system. Sentence classification is the task of classifying an input sentence based on predefined categories. However, spelling or space error contained in the input sentence causes problems in morphological analysis and tokenization. This paper proposes a novel approach of Integrated Eojeol (Korean syntactic word separated by space) Embedding to reduce the effect of poorly analyzed morphemes on sentence classification. The paper also proposes two noise insertion methods that further improve classification performance. Our evaluation results indicate that by applying the proposed methods on the existing sentence classifiers, the sentence classification accuracy on erroneous sentences is increased by 8% to 15%.

Automatic grading requires the adaption of the latest technologies. It has become essential especially when most of the courses became online courses (MOOCs). The objectives of the current work are (1) Reviewing the literature on the text semantic similarity and automatic exam correction systems, (2) Proposing an automatic exam correction framework (HMB-AECF) for MCQs, essays, and equations that is abstracted into five layers, (3) Suggesting equations similarity checker algorithm named “HMB-MMS-EMA”, (4) Presenting an expression matching dataset named “HMB-EMD-v1”, (5) Comparing the different approaches to convert textual data into numerical data (Word2Vec, FastText, Glove, and Universal Sentence Encoder (USE)) using three well-known Python packages (Gensim, SpaCy, and NLTK), and (6) Comparing the proposed equations similarity checker algorithm (HMB-MMS-EMA) with a Python package (SymPy) on the proposed dataset (HMB-EMD-v1). Eight experiments were performed on the Quora Questions Pairs and the UNT Computer Science Short Answer datasets. The best-achieved highest accuracy in the first four experiments was 77.95% without fine-tuning the pre-trained models by the USE. The best-achieved lowest root mean square error (RMSE) in the second four experiments was 1.09 without fine-tuning the used pre-trained models by the USE. The proposed equations similarity checker algorithm (HMB-MMS-EMA) reported 100% accuracy over the SymPy Python package which reported 71.33% only on “HMB-EMD-v1”.

Data in the Cloud, which applies to data as a cloud service provider (CSP), transmits stores, or manages it. The company will enforce the same definition of data usage while the data is resident within the enterprise and thus extend the required cryptographic security criteria to data collected, exchanged, or handled by CSP. The CSP Service Level Agreements cannot override the cryptographic access measures. When the data is transferred securely to CSP, it can be securely collected, distributed, and interpreted. Data at the rest position applies to data as it is processed internally in organized and in the unstructured ways like databases and file cabinets. The Data at the Rest example includes the use of cryptography for preserving the integrity of valuable data when processed. For cloud services, computing takes multiple forms from recording units, repositories, and many unstructured items. This paper presents a secure model for Data at rest. The TF-Sec model suggested is planned for use with Slicing, Tokenization, and Encryption. The model encrypts the given cloud data using AES 256 encryption, and then the encrypted block is sliced into the chunks of data fragments using HD-Slicer. Then it applies tokenization algorithm TKNZ to each chunk of data, applies erasure coding technique to tokens, applies the data dispersion technique to scramble encrypted data fragments, and allocates to storage nodes of the multiple CSP. In taking the above steps, this study aims to resolve the cloud security problems found and to guarantee the confidentiality of their data to cloud users due to encryption of data fragments would be of little benefit to a CSP.

Sentiment Analysis concerns the analysis of ideas, emotions, evaluations, values, attitudes and feelings about products, services, companies, individuals, tasks, events, titles and their characteristics. With the increase in applications on the Internet and social networks, Sentiment Analysis has become more crucial in the field of text mining research and has since been used to explore users’ opinions on various products or topics discussed on the Internet. Developments in the fields of Natural Language Processing and Computational Linguistics have contributed positively to Sentiment Analysis studies, especially for sentiments written in non-structured or semi-structured languages. In this paper, we present a literature review on the pre-processing task on the field of sentiment analysis and an analytical and comparative study of different researches conducted in Arabic social networks. This study allowed as concluding that several works have dealt with the generation of stop words dictionary. In this context, two approaches are adopted: first, the manual one, which gives rise to a limited list, and second, the automatic, where the list of stop words is extracted from social networks based on defined rules. For stemming two, algorithms have been proposed to isolate prefixes and suffixes from words in dialects. However, few works have been interested in dialects directly without translation. The Moroccan dialect in particular is considered as the 5th dialect studied among Arabic dialects after Jordanian, Egyptian, Tunisian and Algerian dialects. Despite the significant lack in studies carried out on Arabic dialects, we were able to extract several conclusions about the difficulties and challenges encountered through this comparative study, as well as the possible ways and tracks to study in any dialects sentiment analysis pre-processing solution.

Findings of the Association for Computational Linguistics: EMNLP 2020, In traditional NLP, we tokenize a given sentence as a preprocessing, and thus the tokenization is unrelated to a target downstream task. To address this issue, we propose a novel method to explore a tokenization which is appropriate for the downstream task. Our proposed method, optimizing tokenization (OpTok), is trained to assign a high probability to such appropriate tokenization based on the downstream task loss. OpTok can be used for any downstream task which uses a vector representation of a sentence such as text classification. Experimental results demonstrate that OpTok improves the performance of sentiment analysis and textual entailment. In addition, we introduce OpTok into BERT, the state-of-the-art contextualized embeddings and report a positive effect.

Cyberbullying is disturbing and troubling online misconduct. It appears in various forms and is usually in a textual format in most social networks. Intelligent systems are necessary for automated detection of these incidents. Some of the recent experiments have tackled this issue with traditional machine learning models. Most of the models have been applied to one social network at a time. The latest research has seen different models based on deep learning algorithms make an impact on the detection of cyberbullying. These detection mechanisms have resulted in efficient identification of incidences while others have limitations of standard identification versions. This paper performs an empirical analysis to determine the effectiveness and performance of deep learning algorithms in detecting insults in Social Commentary. The following four deep learning models were used for experimental results, namely: Bidirectional Long Short-Term Memory (BLSTM), Gated Recurrent Units (GRU), Long Short-Term Memory (LSTM), and Recurrent Neural Network (RNN). Data pre-processing steps were followed that included text cleaning, tokenization, stemming, Lemmatization, and removal of stop words. After performing data pre-processing, clean textual data is passed to deep learning algorithms for prediction. The results show that the BLSTM model achieved high accuracy and F1-measure scores in comparison to RNN, LSTM, and GRU. Our in-depth results shown which deep learning models can be most effective against cyberbullying when directly compared with others and paves the way for future hybrid technologies that may be employed to combat this serious online issue.

In the derived approach, an analysis is performed on Twitter data for World Cup soccer 2014 held in Brazil to detect the sentiment of the people throughout the world using machine learning techniques. By filtering and analyzing the data using natural language processing techniques, sentiment polarity was calculated based on the emotion words detected in the user tweets. The dataset is normalized to be used by machine learning algorithms and prepared using natural language processing techniques like word tokenization, stemming and lemmatization, part-of-speech (POS) tagger, name entity recognition (NER), and parser to extract emotions for the textual data from each tweet. This approach is implemented using Python programming language and Natural Language Toolkit (NLTK). A derived algorithm extracts emotional words using WordNet with its POS (part-of-speech) for the word in a sentence that has a meaning in the current context, and is assigned sentiment polarity using the SentiWordNet dictionary or using a lexicon-based method. The resultant polarity assigned is further analyzed using na&iuml, ve Bayes, support vector machine (SVM), K-nearest neighbor (KNN), and random forest machine learning algorithms and visualized on the Weka platform. Na&iuml, ve Bayes gives the best accuracy of 88.17% whereas random forest gives the best area under the receiver operating characteristics curve (AUC) of 0.97.

Medical images are playing an important role in the medical domain. A mature medical visual question answering system can aid diagnosis, but there is no satisfactory method to solve this comprehensive problem so far. Considering that there are many different types of questions, we propose a model called CGMVQA, including classification and answer generation capabilities to turn this complex problem into multiple simple problems in this paper. We adopt data augmentation on images and tokenization on texts. We use pre-trained ResNet152 to extract image features and add three kinds of embeddings together to deal with texts. We reduce the parameters of the multi-head self-attention transformer to cut the computational cost down. We adjust the masking and output layers to change the functions of the model. This model establishes new state-of-the-art results: 0.640 of classification accuracy, 0.659 of word matching and 0.678 of semantic similarity in ImageCLEF 2019 VQA-Med data set. It suggests that the CGMVQA is effective in medical visual question answering and can better assist doctors in clinical analysis and diagnosis.

The blockchain technology emerged in 2008 as a distributed peer to peer network structure, capable of ensuring security for transactions made using the Bitcoin digital currency, without the need for third party intermediaries to validate them. Although its beginning was linked to cryptocurrencies, its use has diversified over the recent years. There are various projects using the blockchain technology to perform document validation, electronic voting, tokenization of non-perishable goods, and many others. With its increasing use, concern arises with possible attacks that could threaten the integrity of the consensus of the chain. One of the well-known attacks to the blockchain consensus mechanism is the selfish mining attack, in which malicious nodes can deflect their behavior from the standard pattern by not immediately disclosing their newly mined blocks. This malicious behavior can result in a disproportionate share of rewards for those nodes, especially if they have a significant processing power. The goal of this paper is to present a simple heuristic to detect the presence of selfish mining attack (and variants) in blockchain networks that use the proof-of-work (PoW) consensus algorithm. The proposal is to signal when the blockchain fork height deviates from the standard, indicating when the network is under the influence of such attacks.

The processing of speech as an explicit sequence of events is common in automatic speech recognition (linguistic events), but has received relatively little attention in paralinguistic speech classification despite its potential for characterizing broad acoustic event sequences. This paper proposes a framework for analyzing speech as a sequence of acoustic events, and investigates its application to depression detection. In this framework, acoustic space regions are tokenized to ‘words’ representing speech events at fixed or irregular intervals. This tokenization allows the exploitation of acoustic word features using proven natural language processing methods. A key advantage of this framework is its ability to accommodate heterogeneous event types: herein we combine acoustic words and speech landmarks, which are articulation-related speech events. Another advantage is the option to fuse such heterogeneous events at various levels, including the embedding level. Evaluation of the proposed framework on both controlled laboratory-grade supervised audio recordings as well as unsupervised self-administered smartphone recordings highlight the merits of the proposed framework across both datasets, with the proposed landmark-dependent acoustic words achieving improvements in F1(depressed) of up to 15% and 13% for SH2-FS and DAIC-WOZ respectively, relative to acoustic speech baseline approaches.

The increasing usage of the Internet and social networks has produced a significant amount of online textual data. These online textual data led to information overload and redundancy. It is important to eliminate the information redundancy and preserve the time required for reading these online textual data. Thus, there is a persistent need for an automatic text summarization system, which extract the relevant and salient information from a collection of documents, that sharing the same or related topics. Then, presenting this extracted information in a condensed form to preserve the main topics. This paper proposes an automatic, generic, and extractive Arabic multi-document summarization system. The proposed system employs the clustering-based and evolutionary multi-objective optimization methods. The clustering-based method discovers the main topics in the text, while the evolutionary multi-objective optimization method optimizes three objectives based on coverage, diversity/redundancy, and relevancy. The performance of the proposed system is evaluated using TAC 2011 and DUC 2002 datasets. The experimental results are compared using ROUGE evaluation measure. The obtained results showed the effectiveness of the proposed system compared to other peer systems. The proposed system outperformed other peer systems for all ROUGE metrics using TAC 2011. We achieved an F-measure of 38.9%, 17.7%, 35.4%, and 15.8% for Rouge-1, Rouge-2, Rouge-L, and Rouge-SU4, respectively. In addition, the proposed system with DUC 2002 dataset achieved an F-measure of 47.1%, 23.7%, 47.1%, 20.4% for Rouge-1, Rouge-2, Rouge-L, and Rouge-SU4, respectively.

This article presents a methodology for the analysis of data on the Internet, combining techniques of Big Data analytics, NLP and semantic web in order to find knowledge about large amounts of information on the web. To test the effectiveness of the proposed method, webpages about cyberterrorism were analyzed as a case study. The procedure implemented a genetic strategy in parallel, which integrates (Crawler to locate and download information from the web; to retrieve the vocabulary, using techniques of NLP (tokenization, stop word, TF, TFIDF), methods of stemming and synonyms). For the pursuit of knowledge was built a dataset through the description of a linguistic corpus with semantic ontologies, considering the characteristics of cyber-terrorism, which was analyzed with the algorithms, Random Forests (parallel), Boosting, SVM, neural network, K-nn and Bayes. The results reveal a percentage of the 95.62% accuracy in the detection of the vocabulary of cyber-terrorism, which were approved through cross validation, reaching 576% time savings with parallel processing.

This article proposes novel frameworks of SentiVerb and Spell Checker system, which extracts the reaction, mood, and opinion of users from social media text (SMT). The opinion of users is extracted from their written text on social media such as comments, tweets, blogs, feedbacks etc. and are classified as positive or negative opinion based on sentiment score of SMT using dictionary-based approach and a binary classifier. The dictionary-based approach uses opinion verb dictionary (OVD) to extract the sentiment of opinion verbs present in SMT. This OVD contain only opinion verbs along with their sentiment score. The various steps of the framework such as lower-case conversion, tokenization, spell checker, Part-of-Speech tagging, stop word elimination, stemming, sentiment score calculation, and classification of SMT has been discussed. A new concept of threshold negative parameter is first time introduced in this article. In the experiment, the proposed SentiVerb system’s performance is evaluated on three datasets such as Facebook comments on goods and services tax (GST) implementation in India, tweets on the debate between former president of USA Mr. Barack Obama and Mr. John McCain, and the movie reviews. Consequently, the implementation of the proposed SentiVerb system using rule-based classifier (RBC) gives the best performance result in term of accuracy with 82.5% on GST comments and 79.18% on Obama-McCain debate, which is better than the existing algorithms on the social issues related domain dataset(s). Also, system performance (accuracy of 71.3%) is better than others results on standard movie dataset.

The texts often express the emotions of the writers or cause emotions in the readers. In recent years, the development of the social networks has made emotional analysis of texts into an attractive topic for research. A sentiment analysis system for automatic detection of fine-grained emotions in text consists of three main parts of preprocessing, feature extraction and classification. The main focus of this paper is on presenting a novel ensemble classifier that is consisted of 1500 of k-Nearest Neighbor, Multilayer Perceptron and Decision Tree basic classifiers, which is able to systematically distinguish different fine-grained emotions between regular and irregular sentences with a proper accuracy. Moreover, Tree-structured Parzen Estimator is employed to tune parameters of the basic classifiers. The preprocessing and feature extraction operations are performed by natural language processing tools (Tokenization and Lemmatization) and Doc2Vector algorithm, respectively. Three different sets of ISEAR, OANC and CrowdFlower are used to evaluate the proposed method, which consists of regular and irregular sentences. The evaluation results show that accuracies of the proposed ensemble classifier are 99.49 and 88.49% in the detection of regular and irregular sentences, respectively.

Purpose The purpose of this research study is to extract and identify named entities from Hadith literature. Named entity recognition (NER) refers to the identification of the named entities in a computer readable text having an annotation of categorization tags for information extraction. NER is an active research area in information management and information retrieval systems. NER serves as a baseline for machines to understand the context of a given content and helps in knowledge extraction. Although NER is considered as a solved task in major languages such as English, in languages such as Urdu, NER is still a challenging task. Moreover, NER depends on the language and domain of study; thus, it is gaining the attention of researchers in different domains. Design/methodology/approach This paper proposes a knowledge extraction framework using finite-state transducers (FSTs) – KEFST – to extract the named entities. KEFST consists of five steps: content extraction, tokenization, part of speech tagging, multi-word detection and NER. An extensive empirical analysis using the data corpus of Urdu translation of Sahih Al-Bukhari, a widely known hadith book, reveals that the proposed method effectively recognizes the entities to obtain better results. Findings The significant performance in terms of f-measure, precision and recall validates that the proposed model outperforms the existing methods for NER in the relevant literature. Originality/value This research is novel in this regard that no previous work is proposed in the Urdu language to extract named entities using FSTs and no previous work is proposed for Urdu hadith data NER.

In today's digital information age, the conflict between the public's growing awareness of their own data protection and the data owners' inability to obtain data ownership has become increasingly prominent. The emergence of blockchain provides a new direction for data protection and data tokenization. Nonetheless, existing cryptocurrencies such as Bitcoin using Proof-of-Work are particularly energy intensive. On the other hand, classical protocols such as Byzantine agreement do not work efficiently in an open environment. Therefore, in this paper, we propose a permission-less blockchain with a novel double-DAG (directed acyclic graph) architecture called DLattice, where each account has its own Account-DAG and all accounts make up a greater Node-DAG structure. DLattice parallelizes the growth of each account's Account-DAG, each of which is not influenced by other accounts' irrelevant transactions. DLattice uses a new DPoS-BA-DAG(PANDA) protocol to reach consensus among users only when the forks are observed. Based on proposed DLattice, we introduce a process of data tokenization, including data assembling, data anchoring, and data authorization. We implement DLattice and evaluate its performance on 25 ECS virtual machines, simulating up to 500 nodes. The experimental results show that DLattice reaches a consensus in 10 seconds, achieves desired throughput, and incurs almost no penalty for scaling to more users.

Natural Language Processing for Requirements Engineering (NLP4RE) is an area of research and development that seeks to apply natural language processing (NLP) techniques, tools, and resources to the requirements engineering (RE) process, to support human analysts to carry out various linguistic analysis tasks on textual requirements documents, such as detecting language issues, identifying key domain concepts, and establishing requirements traceability links. This article reports on a mapping study that surveys the landscape of NLP4RE research to provide a holistic understanding of the field. Following the guidance of systematic review, the mapping study is directed by five research questions, cutting across five aspects of NLP4RE research, concerning the state of the literature, the state of empirical research, the research focus, the state of tool development, and the usage of NLP technologies. Our main results are as follows: (i) we identify a total of 404 primary studies relevant to NLP4RE, which were published over the past 36 years and from 170 different venues; (ii) most of these studies (67.08%) are solution proposals, assessed by a laboratory experiment or an example application, while only a small percentage (7%) are assessed in industrial settings; (iii) a large proportion of the studies (42.70%) focus on the requirements analysis phase, with quality defect detection as their central task and requirements specification as their commonly processed document type; (iv) 130 NLP4RE tools (i.e., RE specific NLP tools) are extracted from these studies, but only 17 of them (13.08%) are available for download; (v) 231 different NLP technologies are also identified, comprising 140 NLP techniques, 66 NLP tools, and 25 NLP resources, but most of them—particularly those novel NLP techniques and specialized tools—are used infrequently; by contrast, commonly used NLP technologies are traditional analysis techniques (e.g., POS tagging and tokenization), general-purpose tools (e.g., Stanford CoreNLP and GATE) and generic language lexicons (WordNet and British National Corpus). The mapping study not only provides a collection of the literature in NLP4RE but also, more importantly, establishes a structure to frame the existing literature through categorization, synthesis and conceptualization of the main theoretical concepts and relationships that encompass both RE and NLP aspects. Our work thus produces a conceptual framework of NLP4RE. The framework is used to identify research gaps and directions, highlight technology transfer needs, and encourage more synergies between the RE community, the NLP one, and the software and systems practitioners. Our results can be used as a starting point to frame future studies according to a well-defined terminology and can be expanded as new technologies and novel solutions emerge.

Developing Text Normalization (TN) systems for Text-to-Speech (TTS) on new languages is hard. We propose a novel architecture to facilitate it for multiple languages while using data less than 3% of the size of the data used by the state of the art results on English. We treat TN as a sequence classification problem and propose a granular tokenization mechanism that enables the system to learn majority of the classes and their normalizations from the training data itself. This is further combined with minimal precoded linguistic knowledge for other classes. We publish the first results on TN for TTS in Spanish and Tamil and also demonstrate that the performance of the approach is comparable with the previous work done on English. All annotated datasets used for experimentation will be released at https://github.com/amazon-research/proteno., Accepted to NAACL 2021

In this paper, a feature extraction method based on neural networks for Vietnamese text classification is developed. The Vietnamese online news documents are initially preprocessed to transform to lower-case documents as well as remove punctuation and stop-words. Then, the tokenization applying the combination of uni-gram and bi-gram models is conducted to generate a list of tokens for each document. As a result, polyphonic linguistic problems in Vietnamese language are diminished. The statistical term frequency - inverse document frequency model is employed to represent the lists of tokens as real value vectors. Instead of applying conventional feature selection algorithms, neural networks are conducted for dimensionality reduction. Therefore, not only the size of the term frequency - inverse document frequency vectors are reduced, but also distinctive feature vectors are created for text classification tasks. Support vector machines are utilized in the classification step. The results attained in the experiment have shown that our work using neural networks for the feature extraction outperforms other traditional methods.

Online product reviews play a major role in the success or failure of an E-commerce business. Before procuring products or services, the shoppers usually go through the online reviews posted by previous customers to get recommendations of the details of products and make purchasing decisions. Nevertheless, it is possible to enhance or hamper specific E-business products by posting fake reviews, which can be written by persons called fraudsters. These reviews can cause financial loss to E-commerce businesses and misguide consumers to take the wrong decision to search for alternative products. Thus, developing a fake review detection system is ultimately required for E-commerce business. The proposed methodology has used four standard fake review datasets of multidomains include hotels, restaurants, Yelp, and Amazon. Further, preprocessing methods such as stopword removal, punctuation removal, and tokenization have performed as well as padding sequence method for making the input sequence has fixed length during training, validation, and testing the model. As this methodology uses different sizes of datasets, various input word-embedding matrices of n-gram features of the review’s text are developed and created with help of word-embedding layer that is one component of the proposed model. Convolutional and max-pooling layers of the CNN technique are implemented for dimensionality reduction and feature extraction, respectively. Based on gate mechanisms, the LSTM layer is combined with the CNN technique for learning and handling the contextual information of n-gram features of the review’s text. Finally, a sigmoid activation function as the last layer of the proposed model receives the input sequences from the previous layer and performs binary classification task of review text into fake or truthful. In this paper, the proposed CNN-LSTM model was evaluated in two types of experiments, in-domain and cross-domain experiments. For an in-domain experiment, the model is applied on each dataset individually, while in the case of a cross-domain experiment, all datasets are gathered and put into a single data frame and evaluated entirely. The testing results of the model in-domain experiment datasets were 77%, 85%, 86%, and 87% in the terms of accuracy for restaurant, hotel, Yelp, and Amazon datasets, respectively. Concerning the cross-domain experiment, the proposed model has attained 89% accuracy. Furthermore, comparative analysis of the results of in-domain experiments with existing approaches has been done based on accuracy metric and, it is observed that the proposed model outperformed the compared methods.

Most of the people express their ideas, views and opinions over social media. These feedbacks or comments carry an emotion in them. This data can be either straight forward or sarcastic. This text has to be analyzed and make the reviewers understand the exact intent of the writer. Sentiment analysis is used to analyze the perspective of text. Sarcasm can also be present in the text which is a bitter way of conveying the information. Selection of the dataset is the initial task. Dataset is retrieved from Amazon datasets. The next task is preprocessing of data which includes tokenization of data, polarity identification, stemming and lemmatization. Later, feature extraction is done, which includes term frequency, Inverse document frequency and n-gram. The classification algorithms are used such as Support Vector Machine (SVM), K Nearest Neighbors and Random forest are implemented. Further the calculation of results is evaluated by the parameter accuracy.

Social network becomes an effective method to engage with our friends. It enables the users to extract a number of in-formation and its usage are increasing day by day. Amidst all the social networking sites, Twitter is one of the interactive social networking services. To change the authorized user accounts, many spammers are utilized a vast amount of spam. Machine Learning (ML) technique is utilized for spam detection system in social sites and for the detection of spammer. Data collection is usually done from H-Spam 14 site with the help of pre-processing mechanism, the data is transforming into lowercase. After the first step, pre-processed data comes under feature extraction phase, which utilizes tokenization process to breakdown each sentence into word group in order to extract the best feature from the raw data. The optimization algorithm referred as Artificial Bee Colony (ABC) is used to pick the optimized value from extracted set of features. It is also utilized to get the optimal sets of features from spam and non-spam data. At the end, performance measure criterion and comparing the existing and proposed work in order to look over the progress of the proposed work. In this work, spam detection system is having higher accuracy, precision, recall, and F-measure as compares to classifiers used previously such as, Naive Bayes and Support Vector Machine (SVM)

In videos, description and keywords play an important role in the choosing process of the right video to watch. The main idea of the proposed approach is to generate descriptions and timestamps for videos automatically. Our approach plays an essential role in reducing the time consumed searching for the proper video. It aims to save time for users watching wrong unwanted videos and saves their time using timestamps. Timestamps would help to find and watch only the desired part of the video. One of the main goals of our approach is actual keyword extraction. Extracted keywords help finding videos with the significant video's keywords. The summarizing of the video depends on frames, emotions and speech. Firstly the video content appears in the frame and output a summarized text for the video content. Secondly, emotion and how it changes during a specific period merged with the outputted summarization of the frames. Thirdly, the audio transcribing into text occurs and output an abstractive summarization of the audio track. Finally, the fusion happens between all summarizations (audio, video, emotion) using natural language processing techniques. Techniques such as tokenization, sentence segmentation and lemmatization & stemming, and then abstractive summarization. Video summarization occurs to get a meaningful accurate description of the video. Having an accurate description helps finding the inquired content matching the description. The implemented experiment showed that on average 87% of the participants found generated text well representing the video.

Unstructured data is growing rapidly due to the increase in various social media that allow individuals to express and write reviews in either a formal or informal style. Thus, in sentiment analysis, it becomes difficult to identify and analyze both positive and negative reviews. Thai is a low-resource language with few resources with which to conduct NLP research and the lack of a sentiment corpus. The main objective of this paper is to present a framework in which to construct a Thai sentiment corpus and sentiment polarity classification utilizing the cosine similarity technique. The proposed framework consists of three main steps: data collection; data preprocessing; and sentiment similarity measurement. Initially, data collection generates a sub-step that is a manual sentiment that classifies polarity into positive and negative. Data preprocessing is then applied. Moreover, we also created a special database in which to store text tokenization, convert abbreviations, check spelling errors, and conduct stop-word removal. Lastly, sentiment similarity measurement was conducted between two reviews with a combination of TF-IDF and the cosine similarity technique. We evaluated our framework by employing training data incorporating 3,129 reviews and testing data comprised of 1,000 reviews. The experiment results demonstrated that the proposed framework achieved an accuracy of 81.2%. We further observed that the data preprocessing step herein significantly affected the accuracy of the sentence similarity measurements of the two reviews.

Most of the recent Natural Language Processing(NLP) studies are based on the Pretrain-Finetuning Approach (PFA), but in small and medium-sized enterprises or companies with insufficient hardware there are many limitations to servicing NLP application software using such technology due to slow speed and insufficient memory. The latest PFA technologies require large amounts of data, especially for low-resource languages, making them much more difficult to work with. We propose a new tokenization method, ONE-Piece, to address this limitation that combines the morphology-considered subword tokenization method and the vocabulary method used after probing for an existing method that has not been carefully considered before. Our proposed method can also be used without modifying the model structure. We experiment by applying ONE-Piece to Korean, a morphologically-rich and low-resource language. We derive an optimal subword tokenization result for Korean-English machine translation by conducting a case study that combines the subword tokenization method, morphological segmentation, and vocabulary method. Through comparative experiments with all the tokenization methods currently used in NLP research, ONE-Piece achieves performance comparable to the current Korean-English machine translation state-of-the-art model.

The development of blockchain technology and some of its peculiarities such as smart contracts and asset tokenization have important implications in the auditing environment. This paper aims to evaluate the current stage of blockchain technology application on auditing area, analyzing scientific publications since the emergence of this distributed ledger technology, identifying the main implications what is already reality and the potential effects of its improvements in the audit professionals’ activities performance, based on the proposals and suggestions on the main research indexed by Scopus and Web of Science databases.

One of the most popular guidelines used to write and evaluate the exam questions is Bloom's Taxonomy (BT). Since the past recent years, many researchers have worked on the automation of exam question classification into BT. This paper gives a through overview of the standard approaches and techniques used in question classification (QC) into BT cognitive levels including the preprocessing, the dimensionality reduction, the classification and the evaluation stages. A comparison of some recent works related to the subject is also presented in this study.

Urdu is among the most widely used languages in the world for verbal and written communication. Due to lack of optimized and user friendly native Urdu-script support on various platforms, it is mostly written in Romanized script in soft form. In our research, we have developed a refined Urdu lexicon using tokens with the highest frequency of occurrence in the data set. This data set is basically a raw corpus of colloquial Urdu written in Romanized script. The corpus was collected from volunteer participants who used this language as a mode of communication on the Internet and text massaging. The raw corpus is passed through a series of steps such as Prepossessing, Tokenization and Annotation before passing it to computationally extensive subsequent steps. Edit Distance and K-means Clustering techniques are used for identification of candidate tokens and their potential selection/ inclusion in the refined lexicon. We have also identified most commonly used tokens, candidate tokens and other lingual attributes from the data collected. Based on analysis, we have proposed a computational model for refined colloquial Romanized Urdu lexicon development.

Automatic assessment of short answers is one of the Computer Assisted Test works that can assess answers in natural language. Several methods have been used to create a system capable of assessing short answers that are close to human markings. In Indonesian, it might be easy to use string-based similarity methods by matching keywords, as has been done in previous studies. However, short answers have characteristics that focus on content, question type, and answer length, which cannot be accommodated only by string-based methods. This study aims to implement a hybrid method using corpus and string-based similarities. The Semantic Text Similarity (STS) method was used in this study to assess short answers in Indonesian. The STS method consists of three combinations of similarity methods, namely Normalized and Modified Longest Common Subsequence, Second Order Co-occurrence Pointwise Mutual Information, and Common Word Order Similarity. We also use a dynamic corpus with the advantage of being relatively small and adaptable to the learning domain. The Gensim Module is used to generate a dynamic corpus. The dynamic corpus uses the top five answers from students obtained from the Gensim module. The STS method is compared with the Cosine Similarity method since Cosine Similarity is the most commonly used method to assess answers in Indonesian. The results show that the STS method can outperform the Cosine Similarity method based on the Mean Absolute Error value, but still not outperformed in terms of correlation.

Pre-processing and post-processing are significant aspects of natural language processing (NLP) application software. Pre-processing in neural machine translation (NMT) includes subword tokenization to alleviate the problem of unknown words, parallel corpus filtering that only filters data suitable for training, and data augmentation to ensure that the corpus contains sufficient content. Post-processing includes automatic post editing and the application of various strategies during decoding in the translation process. Most recent NLP researches are based on the Pretrain-Finetuning Approach (PFA). However, when small and medium-sized organizations with insufficient hardware attempt to provide NLP services, throughput and memory problems often occur. These difficulties increase when utilizing PFA to process low-resource languages, as PFA requires large amounts of data, and the data for low-resource languages are often insufficient. Utilizing the current research premise that NMT model performance can be enhanced through various pre-processing and post-processing strategies without changing the model, we applied various decoding strategies to Korean&ndash, English NMT, which relies on a low-resource language pair. Through comparative experiments, we proved that translation performance could be enhanced without changes to the model. We experimentally examined how performance changed in response to beam size changes and n-gram blocking, and whether performance was enhanced when a length penalty was applied. The results showed that various decoding strategies enhance the performance and compare well with previous Korean&ndash, English NMT approaches. Therefore, the proposed methodology can improve the performance of NMT models, without the use of PFA, this presents a new perspective for improving machine translation performance.

In the present time Chatbot is an essential tool used by many organizations to provide services to their targeted customers round the clock. This research focuses on a domain-specific Chatbot that can be helpful for educational institutes. This Chatbot will be a virtual (representation) to the admission seekers. It will provide answers regarding the university, its departments, admission fees and other admission related FAQ. For the sake of the research, frequently asked questions of a university were collected and an unsupervised learning model along with natural language processing techniques was deployed to answer the questions of the admission candidates. Tokenization, stop words removal followed by vectorization were implemented for preprocessing the training data. User’s inputs were similarly processed and then tf-idf based cosine similarity applied to retrieve the best answer. Later, a user-centric evaluation metric was used to evaluate the model and as per the metric, our current model showed approximately 80% accuracy.

The General Data Protection Regulation (GDPR) has been enforced since May 2019 and became a disruptive issue to every organization due to its severe penalties in the data breaches or use of personal data for illegal purposes, e.g., lack of the consent of data subject. Therefore, the data Pseudonymization and Anonymization are one of the employed techniques to protect and reduce the privacy risks from the data breach. Unfortunately, they also destroy the pattern of the data, which represents the fact that it could be analyzed or monetized to gain useful insights by data analytics or data science approaches. This paper focuses on optimizing the privacy and insight method that the data could be useful for analyzing and also compliance with the GDPR. This paper proposes the guideline consists of three techniques: tokenization, suppression, and generalization to protect personal data by calculating risk scores from two methods: data classification and data uniqueness. The criteria in the guideline are experimented to achieve the optimized classification performance in protected data compared with five original open data by analyzing with three data mining algorithms with the hyperparameter tuning process. The results show that the protected data by the proposed guideline can protect adequate information and achieve insignificant classification performance when compared to the unprotected data.

The work here presented moves from theacknowledgment that, even if blockchain technology has beenaround for more than ten years, the knowledge about itseconomic and business implications is fragmented andheterogeneous. In the first place, it is analyzed the shift fromeconomics to tokenomics and the central role of the token withinblockchain-based ecosystems. Subsequently, a generalizeddefinition of the token is proposed. Diving into the requirementsfor a comprehensive description of tokens, that takes intoaccount their wide variety, a comparative assessment of tokenclassification frameworks available in the literature isperformed. This analysis is leveraged to propose a new andcomprehensive token classification framework, based on amorphological analysis representation. The proposedframework will be further refined with an empirical anditerative approach in future works. See the presentation HERE:https://easychair.org/smart-slide/slide/99Fs#

In the digital payment system, the transactions and their data about clients are very sensitive, so the security and privacy of personal information of the client is a big concern. The confirmation towards security necessities prevents the data from a stolen and unauthorized person over the digital transactions, So the stronger authentication methods required, which must be based on cryptography. Initially, in the payment ecosystem, they were using the Kerberos protocol, but now different approaches such as Challenge-Handshake Authentication Protocol (CHAP), Tokenization, Two-Factor Authentication(PIN, MPIN, OTP), etc. such protocols are being used in the payment system. This paper presents the use cases of different authentication protocols. Further, the use of these protocols in online payment systems to verify each individual are explained.

Smart contracts vulnerability auditing is vitally critical to ensure transaction execution in normal on blockchain. The current data-driven approaches normally tokenize smart contracts into a series of sequences according to only one tokenization standard for vulnerability detection purpose, resulting some of the semantic contexts could not be reflected within restricted sequence length. To address this limitation, we generate sequences from smart contracts in three tokenization standards for which we utilize n-gram language model to capture semantic contexts respectively, and finally exploiting our effective combination strategy of Intersection or Union to integrate the audited results from multiple semantic contexts. In order to evaluate the proposed approach, we applied it on over 7200 Ethereum smart contract samples. Experimental result shows our proposed method is capable of detecting vulnerabilities and competitive with the baseline in test sets, with improved precision of over 44% when Intersection is applied in their results, as well as improved Recall measure up by over 300% and F-measure up by 220% when Union is applied. Our proposed method for smart contract vulnerability detection, an important tool for developing quality decentralized software applications, is able to analyze multiple semantic contexts and successfully detects more true vulnerabilities with high precision, outperforming that of the baseline approaches.

International audience; In the context of data tokenization, we model a token as a vector of a finite dimensional metric space E and given a finite subset of E, called the token set, we address the problem of deciding whether a given token is in a small neighborhood of an other token. We derive conditions to characterize thenearest token of a given one and show that these conditions are fulfilled asymptotically as the dimension of E tends to infinity. Whereas the classical nearest neighbor search is inefficient to solve such problem, we propose a new probabilistic algorithm, which becomes efficient if the dimension of E is large enough.

Stack Overflow is one of the most widely used platforms for asking questions and queries on topics related to computer science, software development and general computer programming. Tagging of the questions is particularly useful for indexing information based on the tags. Currently, a user enters the tag manually for a question asked by him/her. The question should contain at least one tag manually typed by the user. It can be seen that most of the questions asked should either have more tags associated with it or aren’t tagged accurately and appropriately. Since there are a huge number of tags, the process of searching through all the tags manually and find relevant ones can be cumbersome and is therefore overlooked by most of the users asking the questions. This research is focused on exploring methods for developing an autonomous tagging system using Machine learning methods like k-Nearest Neighbors and Random Forest along with some crucial data preprocessing steps like Stemming, Tokenization and removing Stop words. The dataset for the above research is taken from kaggle.com which has a 10% Stackoverflow question dataset open for all. The results of the following proposed system for automatic tagging were satisfactory. Random Forest gave an average percentage accuracy of 70% across all the tags while k-Nearest Neighbors performed slightly better giving an accuracy of 75%.

The success of pretrained transformer language models (LMs) in natural language processing has led to a wide range of pretraining setups. In particular, these models employ a variety of subword tokenization methods, most notably byte-pair encoding (BPE) (Sennrich et al., 2016; Gage, 1994), the WordPiece method (Schuster and Nakajima, 2012), and unigram language modeling (Kudo, 2018), to segment text. However, to the best of our knowledge, the literature does not contain a direct evaluation of the impact of tokenization on language model pretraining. We analyze differences between BPE and unigram LM tokenization, finding that the latter method recovers subword units that align more closely with morphology and avoids problems stemming from BPE's greedy construction procedure. We then compare the fine-tuned task performance of identical transformer masked language models pretrained with these tokenizations. Across downstream tasks and two languages (English and Japanese), we find that the unigram LM tokenization method matches or outperforms BPE. We hope that developers of future pretrained LMs will consider adopting the unigram LM method over the more prevalent BPE., 5 pages, 3 figures. To be published in Findings of EMNLP 2020

Prior work on Sign Language Translation has shown that having a mid-level sign gloss representation (effectively recognizing the individual signs) improves the translation performance drastically. In fact, the current state-of-the-art in translation requires gloss level tokenization in order to work. We introduce a novel transformer based architecture that jointly learns Continuous Sign Language Recognition and Translation while being trainable in an end-to-end manner. This is achieved by using a Connectionist Temporal Classification (CTC) loss to bind the recognition and translation problems into a single unified architecture. This joint approach does not require any ground-truth timing information, simultaneously solving two co-dependant sequence-to-sequence learning problems and leads to significant performance gains. We evaluate the recognition and translation performances of our approaches on the challenging RWTH-PHOENIX-Weather-2014T (PHOENIX14T) dataset. We report state-of-the-art sign language recognition and translation results achieved by our Sign Language Transformers. Our translation networks outperform both sign video to spoken language and gloss to spoken language translation models, in some cases more than doubling the performance (9.58 vs. 21.80 BLEU-4 Score). We also share new baseline translation results using transformer networks for several other text-to-text sign language translation tasks.

Blockchain has the capacity to transform the industries disruptively as it presents new features like smart contracts, tokenization of content, eliminating counterfeit products, supply chain improvement, digital twins, and end-to-end security. This letter presents a Diamond Accountability Model (DAM) where a public or private authority is included in the blockchain transactions providing non-repudiation of digital transactions, holistic security and effective governance. The proposed technique aims to present a decentralized solution for multi-agent applications where many partnering organizations have to collaborate without suffering from the security, accountability, maintenance, scalability, and integrity problems over distributed cyber-physical systems (CPS). The proposed scheme positions authorities also in the chain that enables additional accountability and trust. The scheme proposes a verification mechanism where the authorized organizations are also included in the verification of transactions over the blockchain. In order to elucidate, a conceptual use case is presented where at least two partnering organizations collaborate with each other within a decentralized but also authorized blockchain-enabled scheme.

Information technology has explored the growth of text documents data in many organizations and the structural arrangement of voluminous data is a complex task. Handling the text document data is a challenging process involving not only the training of models but also numerous additional procedures, e.g., data pre-processing, transformation, and dimensionality reduction. In this paper, we describe the system’s architecture, the technical challenges, and the novel solution we have built. We propose a Recurrent Convolutional Neural network (RCNN), based text information retrieval system which efficiently retrieves the text documents and information for the user query. Pre-processing using tokenization and stemming, retrieval using TF-IDF (Term Frequency-Inverse Document Frequency), and RCNN classifier which captures the contextual information is implemented. A real-time advanced search system is developed on a huge set of MAHE University dataset. The performance of the proposed text document retrieval system is compared with other existing algorithms and the efficacy of the method is discussed. The proposed RCNN-based text document information retrieval model performs better in terms of precision, recall, and F-measure. A high-quality and high-performance text docu�ment retrieval search system is presented