Search

Parts-of-speech (POS) tagging plays an important role in the field of natural language processing (NLP), such as—retrieval of information, machine translation, spelling check, language processing, sentiment analysis, and so on. Many works have been done for Bangla part-of-speech (POS) tagging using machine learning but the result does not enough. It is a matter of fact that not even a single effective research work has been conducted for Bangla POS tagging using deep learning due to a lack of data scarcity. Considering that our context is the Bangla POS tagging employing both machine learning and deep learning approach. In our research, we have compared some well-known supervised POS tagging approaches (Brill, HMM, unigram, bigram, trigram, and recurrent neural network) for Bangla languages. The supervised POS tagging technique requires a large number of data set to tag accurately. That is why we have used a large number of data set for POS tagging of Bangla languages, which will accept a raw Bangla text to produce a Bangla POS tagged output that can be directly used for other NLP applications. After the comparison, we have found the best tagging approach in terms of performance. Bangla is an inflectional language. That is why it is a very much tough job for grammatical categories of Bangla language. But our proposed model works well for Bangla languages.

The prediction of next word, letter or phrase for the user, while she is typing, is a really valuable tool for improving user experience. The users are communicating, writing reviews and expressing their opinion on such platforms frequently and many times while moving. It has become necessary to provide the user with an application that can reduce typing effort and spelling errors when they have limited time. The text data is getting larger in size due to the extensive use of all kinds of social media platforms and so implementation of text prediction application is difficult considering the size of text data to be processed for language modeling. This research paper’s primary objective is processing large text corpus and implementing a probabilistic model like N-grams to predict the next word when the user provides input. In this exploratory research, n-gram models are discussed and evaluated using Good Turing Estimation, perplexity measure and type-to-token ratio.

To provide smart access to the results of a sentiment analysis system, a graphical user interface has been developed. A text comment along with an image from Twitter is taken in the existing work by assembling thousands of posts spread on different classes. Twitter found more effective where one can express their feelings in a few words and post it as a text or along with pictures also. It founds more interesting platform where people post their views and others comment on it which makes them related and interactive to each other. It has been carried out by collecting Gurmukhi data from Twitter using Twitter API. Collected content has been filtered, and four categories of data have been generated named as happy, love, sad and angry moods of people. For text feature extraction, TF-IDF and N-gram features are used which are reduced further using mutual information (MI). For image feature extraction, hybrid approach using two different LBP variants is used. The resultant values were recorded to obtain the recall, precision, f-measure, and accuracy to check the efficiency of the proposed system in which three types of classifiers, i.e., Naive Bayes, k-nearest neighbor and decision tree, have been used. The proposed modal shows decision tree classifier well performed for classifying the emotions of image and text data.

According to the problem that few existing Android malware detection studies focus on the malicious code in native libraries, an Android malware detection method based on native libraries is proposed in this paper. Firstly, ARM assembly instructions and grayscale images are extracted from native libraries. Secondly, N-gram features, GLCM, and fusion features are extracted from ARM assembly instructions and grayscale images. Finally, different types of machine learning algorithms are trained with these features to establish the best malware detection classifier. The experimental results show that the proposed method achieves accuracy of 86.3% and can detect malicious code in native libraries effectively.

In this paper, we propose four-level process to recommend the best book to the users. The levels are named as grouping of similar sentences by the semantic network, sentiment analysis (SA), clustering of reviewers and recommendation system. In the first level, grouping of similar sentences by the semantic network is done taking pre-processed data using parts of speech (POS) tagger from the datasets of reviewers and books. In the second level, SA is done in two phases which are training phase and testing phase by using deep learning methodology like convolutional neural networks (CNN) with n-gram method. The outcome of this level is given as input to the third level (clustering) which clusters the reviewers based on their age, locality and gender using K-nearest neighbor (KNN) algorithm. In the last level, a recommendation of books is done based on top-n interesting books using collaborative filtering (CF) algorithm. The system of book recommendation is to be done to get the best accuracy within less elapsing time.

Language identification is used to categorize the language of a given document. Language identification categorizes the contents and can have a better search results for a multilingual document. In this work, we classify each line of text to a particular language and focused on short phrases of length 2–6 words for 15 Indian languages. It detects that a given document is in multilingual and identifies the appropriate Indian languages. The approach used is the combination of n-gram technique and a list of short distinctive words. The n-gram model applied is language independent whereas short word method uses less computation. The results show the effectiveness of our approach over the synthetic data.

Web navigation prediction plays a vital role in web, due to its broad research applications. It can be used for personalization, improvise website design, and business intelligence. Main aim of these applications is to enhance user’s satisfaction levels who are visiting the website. Web navigation prediction model tries to predict the future set of the webpage from their historical navigations. The past navigations are collected in the web server log file. Navigations form the sessions of varied length which are used for building the navigation model. Selecting very long sessions or very small sessions degrades the model performance. Thus, selecting an optimal session length is mandate as it would impact the model performance positively. This paper presents pre-investigation measures like page loss, branching factor and session length. We investigate the performance of prediction model based on two different ranges of session length. First range that has been considered is three to seven (3 to 7) and second range is two to ten (2 to 10). The Model has been evaluated on three real datasets. The experimental results show that selecting session of length ranging from 2 to 10 gives better learning hence intensifies accuracy of navigation prediction model. The model accuracy of Set B showed improvement from 0.27 to 8.73% in MSWEB, 0.62 to 2.8% in BMS and 10.81 to 14.23% in Wikispeedia dataset.

Fake news is news articles providing wrong information to readers. Normally these articles are originated from social networking sites which immensely influence young generation as well as various sectors of life. Among the few biggest sources of rumors, fake news and fake information are social networking sites like Facebook and Twitter. Detection of fake news on Twitter is challenging as the amount of resources is limited; less datasets are available, and also a large amount of data is generated by these social networking sites. It is very difficult to predict if the data is fake or not even after having knowledge in the same area. In this report, we explained a method for detection of fake news on Twitter which involves collecting livestream data of Twitter using flumes and performing N-gram analysis on it for feature extraction and then applying decision tree machine learning technique to classify documents as fake or real.

Offensive language, hate speech, and bullying behavior is prevalent during textual communication happening online. Users usually misuse the anonymity available online social media, use this as an advantage, and engage in behavior that is not acceptable socially in actual world. Social media platforms, analytics companies, and online communities had shown much interest and involvement in this field to cope up with this problem by stopping its propagation in social media and its usage. In this paper, we will propose the work done by researchers to form effective strategies for tackling this problem of identifying offense, aggression, and hate speech in user’s textual posts, comments, microblogs, etc.

Sentiment analysis is becoming increasingly popular due to its wide variety of applications. The use of machine learning is paving a way to sophisticated algorithms so as to have better performance. In this paper, the sentiment analysis is performed on a data set containing movie reviews. Firstly, the n-gram technique is used to analyse more than one word in a text document. Second, we introduce the Gini index method for feature selection. Then we use the ensemble learning technique for classification. Finally, the results are procured from the confusion matrix.

Hate speech is a type of written or a spoken statement that is used to demean or humiliate a person or a community. In this era of new age socialism, this type of speech is prevalent on social media platforms, where certain groups of people display offensive behaviour towards some people that may be distributed over gender, religion, nationality, etc. These kinds of activities must be avoided or suspended on social media platforms. Therefore, it is necessary to automate the detection of hateful content that gets circulated on the social media. The research work provides an enhanced technique as compared to the existing techniques with improved performance. The proposed model of IN-Gram compares the performance of detection of hateful content on social media with the traditional TF-IDF, N-Gram and PMI techniques. The proposed approach improves the hate speech detection rate by 10–12% for larger datasets as compared to existing approaches.

With the growth of web technology, there are huge amount of information in the web for the internet users. Users not only use that information but also provide opinions for decision making process. Sentiment analysis or opinion mining is one of text categorization techniques that extract opinion expressed in a piece of text. This system created sentiment annotated corpus. Feature extraction and selection are needed in sentiment analysis to get high performance. N-grams are used for feature selection and TF-IDF is used for feature extraction. Machine learning is creating a computer programs that improve performance with experience. Machine learning is combination of the techniques and basis from both statistics and computer science. There are generally three types of machine learning algorithm such as supervised machine learning, unsupervised machine learning and hybrid learning. Supervised machine learning also known as classification algorithm that models the relationships between the feature set and the label set. In this system, Myanmar sentiment analysis system is implemented using supervised machine learning method. This system shows the comparison results of support vector machine (SVM) and Naive Bayes algorithms.

There have been many achievements in using Artificial Intelligence (AI) to automatically compose melody. But there still need statistical methods to evaluate the degree of similarity between AI-generated melody and human composed melody. With N-gram models, this research compared the distribution of transitions of a combination of rhythm and pitch of melodies between melodies generated by AI (Magenta of Google) and the original British and American folk songs pieces used for training. The result shows that the AI system is good at modeling the common patterns of note transitions which appear more frequently in the data set of human-composed melodies than humans, and such patterns appear more frequently in the generated melodies compared with human-composed melody. The experiment shows that using the N-gram model to analyze the distribution characteristics of pitch and rhythm features of melody to distinguish between the AI-generated melody dataset and human composed melody dataset is a direction worthy of further research.

A language model is a technique that shows which words are more or less likely to be generated during some conversation in any natural language. N-gram language modeling is the pioneer technology used to construct language models. N-gram technique considers preceding words only to predict the upcoming word. Factored language modeling is a formalism that provides a facility to undertake other linguistic knowledge of the words like gender, number, part of speech, stem of word along with word itself to predict next word in a sentence. This paper discusses the effect of various combinations of linguistic features of word on predictability of next word in Hindi-language sentence. The paper also discusses how use of linguistic features decreases the perplexity by 31.71% as compared to perplexity of baseline N-gram language model.

The bag-of-visual-words (BoVW) model has emerged as an effective approach to represent features for focal liver lesions (FLLs). However, most of the previous methods have the limitation of insufficient consideration of the spatiotemporal co-occurrence information, which provokes the low descriptive power of classic visual words. In contrast to previous work, we propose a novel model for multiphase medical image feature generation named the Bi-gram bag-of-spatiotemporal words (Bi-gram BoSTW) to capture the temporal information, as well as, the spatial co-occurrence relationship of the lesion. First, temporal co-occurrence images from multiphase images are constructed. Second, BoVW is employed to extract temporal features from the temporal co-occurrence images and generates the visual words. Finally, we introduce the N-gram schema to add spatial relation to local descriptors. To the best of our knowledge, this is the first work that introduces visual N-grams scheme to contrast-enhanced CT images, which integrates temporal information with spatial co-occurrence relationship and improves the classification performance. The effectiveness of the proposed model is verified on 132 FLLs with confirmed pathology type. The experimental results indicate that (1) the N-gram enriches the semantics and provides more complete representation; (2) the proposed model achieves the best accuracy (83%) with highest training speed (1.5 min) among several well-known methods based on BoVW model.

Sentiment Detection plays a vital role worldwide to measure the acceptance level of any products, movies or facts in the market. Text vectorization (converting text from human readable to machine readable format) and machine learning algorithms are widely used to detect the sentiment of users. This paper presents and evaluates a multi-level architecture based approach using stacked generalization technique named NStackSenti. The presented approach enables the combination of machine learning algorithms to improve the accuracy of detection. Here, Extremely Randomized Tree (ET), Random Forest (RF), Gradient Boost (GB), ADA Boost (ADA), Decision Tree (DT) are used as base classifiers and XGBoost classifier is used as meta estimator. The NStackSenti is applied on two separate datasets to demonstrate the effectiveness in terms of accuracy. NStackSenti provides better accuracy with trigram than unigram and bigram. It provides 83.7% and 86.24% accuracy on 2000 and 50000 data respectively.

Plagiarism is increasing day by day. Plagiarism detection is one of the most complex, but a must requirement. This paper deals with word level plagiarism detection for Marathi text by using N-gram language model and a Marathi corpus. This is most simple in form still provides good depth for understanding and emphasing copy-paste and paraphrased plagiarism detection. It forms basis for sentence as well as paragraph level processing

Various protein sequence classification approaches are developed to classify unknown sequences in to its classes or familes with an certain accuracy. Features extraction from protein sequence is a key technique to implement all approaches. N-gram encoding method is a popular feature extraction procedure. But to maintain the low computational time and high accuracy level of classification, it requires to fix up the upper limit of ‘N’ of N-gram encoding method. On the other hand, the standard deviation value of protein sequence is one of the important feature value which is extracted by N-gram encoding method. This feature can be extracted by two different ways like standard deviation calculation using standard mean value and using floating mean value. It is also important to find proper method to calculate the value of standard deviation. In this paper, an investigational proof has done to find upper limit of N-gram encoding method as well as find the proper technique to calculate the standard deviation value as a feature which are extracted from unknown protein sequence.