Multinomial Naive Bayes Classifier Research Articles

Linguistic features evaluation for hadith authenticity through automatic machine learning

Abstract There has not been any research that provides an evaluation of the linguistic features extracted from the matn (text) of a Hadith. Moreover, none of the fairly large corpora are publicly available as a benchmark corpus for Hadith authenticity, and there is a need to build a ‘gold standard’ corpus for good practices in Hadith authentication. We write a scraper in Python programming language and collect a corpus of 3,651 authentic prophetic traditions and 3,593 fake ones. We process the corpora with morphological segmentation and perform extensive experimental studies using a variety of machine learning algorithms, mainly through automatic machine learning, to distinguish between these two categories. With a feature set including words, morphological segments, characters, top N words, top N segments, function words, and several vocabulary richness features, we analyze the results in terms of both prediction and interpretability to explain which features are more characteristic of each class. Many experiments have produced good results and the highest accuracy (i.e. 78.28%) is achieved using word n-grams as features using the Multinomial Naive Bayes classifier. Our extensive experimental studies conclude that, at least for Digital Humanities, feature engineering may still be desirable due to the high interpretability of the features. The corpus and software (scripts) will be made publicly available to other researchers in an effort to promote progress and replicability.

A novel feature extraction method based on highly expressed SNPs for tissue-specific gene prediction

BackgroundGene expression provides a means for an organism to produce gene products necessary for the organism to live. Variation in the significant gene expression levels can distinguish the gene and the tissue in which the gene is expressed. Tissue-specific gene expression, often determined by single nucleotide polymorphisms (SNPs), provides potential molecular markers or therapeutic targets for disease progression. Therefore, SNPs are good candidates for identifying disease progression. The current bioinformatics literature uses gene network modeling to summarize complex interactions between transcription factors, genes, and gene products. Here, our focus is on the SNPs’ impact on tissue-specific gene expression levels. To the best of our knowledge, we are not aware of any studies that distinguish tissue-specific genes using SNP expression levels.MethodWe propose a novel feature extraction method based on highly expressed SNPs using k-mers as features. We also propose optimal k-mer and feature sizes used in our approach. Determining the optimal sizes is still an open research question as it depends on the dataset and purpose of the analysis. Therefore, we evaluate our algorithm’s performance on a range of k-mer and feature sizes using a multinomial naive Bayes (MNB) classifier on genes in the 49 human tissues from the Genotype-Tissue Expression (GTEx) portal.ConclusionsOur approach achieves practical performance results with k-mers of size 3. Based on the purpose of the analysis and the number of tissue-specific genes under study, feature sizes [7, 8, 9] and [8, 9, 10] are typically optimal for the machine learning model.

Using natural language processing to classify social work interventions.

Health care organizations are increasingly employing social workers to address patients' social needs. However, social work (SW) activities in health care settings are largely captured as text data within electronic health records (EHRs), making measurement and analysis difficult. This study aims to extract and classify, from EHR notes, interventions intended to address patients' social needs using natural language processing (NLP) and machine learning (ML) algorithms. Secondary data analysis of a longitudinal cohort. We extracted 815 SW encounter notes from the EHR system of a federally qualified health center. We reviewed the literature to derive a 10-category classification scheme for SW interventions. We applied NLP and ML algorithms to categorize the documented SW interventions in EHR notes according to the 10-category classification scheme. Most of the SW notes (n = 598; 73.4%) contained at least 1 SW intervention. The most frequent interventions offered by social workers included care coordination (21.5%), education (21.0%), financial planning (18.5%), referral to community services and organizations (17.1%), and supportive counseling (15.3%). High-performing classification algorithms included the kernelized support vector machine (SVM) (accuracy, 0.97), logistic regression (accuracy, 0.96), linear SVM (accuracy, 0.95), and multinomial naive Bayes classifier (accuracy, 0.92). NLP and ML can be utilized for automated identification and classification of SW interventions documented in EHRs. Health care administrators can leverage this automated approach to gain better insight into the most needed social interventions in the patient population served by their organizations. Such information can be applied in managerial decisions related to SW staffing, resource allocation, and patients' social needs.

Quantifying Perception of Security Through Social Media and Its Relationship With Crime

The perception of security refers to the subjective evaluation of risks related to security events and the magnitude of their consequences. Negative perceptions have severe implications in society. These feelings are commonly quantified by citizen’s surveys, which are time-consuming and do not adapt well to the changing dynamics linked to security. Recently, Twitter social network emerged as an alternative to quantify some of these feelings dynamically. However, most of these approaches focused on counting the amount of content related to crime, which is only one of the relevant factors determining the perception of security. This work introduces a novel model for quantifying the Perception of Security on Twitter based on sentiment analysis and studies its relationship with actual crime. The model relies on an automatic strategy for filtering content related to security based on a support vector machine classifier and quantifying the sentiment in the posts using a multinomial naive Bayes classifier. This model studied more than 1.700.000 tweets in Bogota, collected during more than one year (March 18 of 2019 - April 28 of 2020). Results suggest that machine learning-based approaches may outperform previous security content filtering strategies by 21% in F1-score and provide quantifications on the sentiments of about 40% on accuracy. Trained models were used to provide daily estimates of perception of security, which are different from those captured by counting the number of posts related to security. Furthermore, the associations between perceptions and actual crimes exhibited monthly variations and showed that in some months, perceptions were more related to robbery, a crime with high incidence in the city. These results may help decision-makers devise strategies to reduce the impact of the negative perception on citizens.

Implementation of Machine Learning in Quantum Key Distributions

In the future massive applications of quantum communication network, it is crucial to realize real-time selection of the optimal quantum key distributions (QKD) protocol for the improvement of system security and optimizing resources configuration. In principle, this can be done by utilizing algorithms such as exhaustive traversal or local search algorithm, whose time cost, however, is unbearable. Here we for the first time propose to employ machine learning methods into the selecting of optimal QKD protocol and apply random forest (RF) as an example for illustration. With the help of the easy-to-train classifier of RF, we can achieve a highly efficient optimal QKD protocols selector. Besides, we also do comparisons among RF and other machine learning methods, e.g., supporting vector machine, K-nearest neighbors algorithm, multinomial naive bayes classifier, and convolutional neural networks. Results demonstrate that, besides its advantage in efficiency, the RF classifier also excels both in preciseness and robustness with an accuracy over 98% for the testing set and an enjoyable receiver operating characteristic.

Enriching Domain Concepts with Qualitative Attributes (A Text Mining based Approach)

Attributes, whether qualitative or non-qualitative are the formal description of any real-world entity and are crucial in modern knowledge representation models like ontology. Though ample evidence for the amount of research done for mining non-qualitative attributes (like part-of relation) extraction from text as well as the Web is available in the wealth of literature, on the other side limited research can be found relating to qualitative attribute (i.e., size, color, taste etc.,) mining. Herein this research article an analytical framework has been proposed to retrieve qualitative attribute values from unstructured domain text. The research objective covers two aspects of information retrieval (1) acquiring quality values from unstructured text and (2) then assigning attribute to them by comparing the Google derived meaning or context of attributes as well as quality value (adjectives). The goal has been accomplished by using a framework which integrates Vector Space Modelling (VSM) with a probabilistic Multinomial Naive Bayes (MNB) classifier. Performance Evaluation has been carried out on two data sets (1) HeiPLAS Development Data set (106 adjective-noun exemplary phrases) and (2) a text data set in Medicinal Plant Domain (MPD). System is found to perform better with probabilistic approach compared to the existing pattern-based framework in the state of art

Investigating differential linguistic patterns exhibited by Major Depressive Disorder (MDD) Patients and building a Long Short Term Memory Network + Convolutional Neural Network Model, Logistic Regression model, and a Multinomial Naive Bayes Classifier Algorithm to develop Spero, a hybrid app based Early-MDD diagnosis system

Major Depressive Disorder (MDD), otherwise known as Depression, is the leading psychiatric disorder globally in terms of the number of individuals it affects. Despite this there is no effective and reliable early diagnostics system for MDD. Hence, through this study, we aimed to fill this void by not only investigating linguistic differences in posts made on social media by people exhibiting and people not exhibiting symptoms of MDD but also by developing various machine learning architectures to build an accessible, sensitive, and accurate MDD early diagnostics system. Through the differential linguistic analysis we conducted on the dataset we manually scraped and filtered, we clearly demonstrated that there indeed were certain linguistic and topical features that were different amongst depressed and healthy patients. Furthermore, we also successfully built three different ML Algorithms in which our Long Short Term Memory Network (LSTM) + Convolutional Neural Network (CNN) Model attained an accuracy of 95.00%, our Multinomial Naive Bayes Classifier Algorithm attained an accuracy of 92%, and our Logistic Regression Model achieved an accuracy of 87.627%. Ultimately, given the LSTM + CNN Model’s high accuracy, weighted precision (0.95), recall (0.95), and f-1 score (0.95), we decided to integrate it into an app built on Swift UI to develop Spero, a first of its kind early diagnostics system for MDD.

Stratification of Spam and Ham Short Message Service using Machine Learning Hash Vectorization

With the increase in the usage of mobile technology, the rate of information is duplicated as a huge volume. Due to the volume duplication of message, the identification of spam messages leads to challenging task. The growth of mobile usage leads to instant communication only through messages. This drastically leads to hackers and unauthorized users to the spread and misuse of sending spam messages. The identification of spam messages is a research oriented problem for the mobile service providers in order to raise the number of customers and to retain them. With this overview, this paper focuses on identifying and prediction of spam and ham messages. The SMS Spam Message Detection dataset from KAGGLE machine learning Repository is used for prediction analysis. The identification of spam and ham messages is done in the following ways. Firstly, the levels of spread of target variable namely spam or ham is identified and they are depicted as a graph. Secondly, the essential tokens that are responsible for the spam and ham messages are identified and they are found by using the hashing Vectorizer and it is portrayed in the form of spam and Ham messages word cloud. Thirdly, the hash vectorized SMS Spam Message detection dataset is fitted to various classifiers like Ada Boost Classifier, Extra Tree classifier, KNN classifier, Random Forest classifier, Linear SVM classifier, Kernel SVM classifier, Logistic Regression classifier, Gaussian Naive Bayes classifier, Decision Tree classifier, Gradient Boosting classifier and Multinomial Naive Bayes classifier. The evaluation of the classifier models are done by analyzing the Performance analysis metrics like Accuracy, Recall, FScore, Precision and Recall. The implementation is done by python in Anaconda Spyder Navigator. Experimental Results shows that the Linear Support Vector Machine classifier have achieved the effective performance indicators with the precision of 0.98, recall of 0.98, FScore of 0.98 , and Accuracy of 98.71%.

Count Vectorized Spam and Ham Discernment of Short Message Service using Machine Learning Classification

With the growing volume and the amount of spam message, the demand for identifying the effective method for spam detection is in claim. The growth of mobile phone and Smartphone has led to the drastic increase in the SMS spam messages. The advancement and the clean process of mobile message servicing channel have attracted the hackers to perform their hacking through SMS messages. This leads to the fraud usage of other accounts and transaction that result in the loss of service and profit to the owners. With this background, this paper focuses on predicting the Spam SMS messages. The SMS Spam Message Detection dataset from KAGGLE machine learning Repository is used for prediction analysis. The analysis of Spam message detection is achieved in four ways. Firstly, the distribution of the target variable Spam Type the dataset is identified and represented by the graphical notations. Secondly, the top word features for the Spam and Ham messages in the SMS messages is extracted using Count Vectorizer and it is displayed using spam and Ham word cloud. Thirdly, the extracted Counter vectorized feature importance SMS Spam Message detection dataset is fitted to various classifiers like KNN classifier, Random Forest classifier, Linear SVM classifier, Ada Boost classifier, Kernel SVM classifier, Logistic Regression classifier, Gaussian Naive Bayes classifier, Decision Tree classifier, Extra Tree classifier, Gradient Boosting classifier and Multinomial Naive Bayes classifier. Performance analysis is done by analyzing the performance metrics like Accuracy, FScore, Precision and Recall. The implementation is done by python in Anaconda Spyder Navigator. Experimental Results shows that the Multinomial Naive Bayes classifier have achieved the effective prediction with the precision of 0.98, recall of 0.98, FScore of 0.98 , and Accuracy of 98.20%..

Use of NLP Based Combined Features for Sentiment Classification

Sentiment analysis is the technique of automatic detection of the belief or the mood of an author towards a certain subject in textual form. To extract the opinion present in text, the machine needs expertise in the area of natural language processing. In this paper, machine learning based document-level sentiment classification is performed on Amazon product reviews to classify them as positive and negative. Two NLP based feature extraction techniques (Word Relation and POS based) are used in this study to determine the features that are sentiment bearing. The features are extracted as basic features (unigrams, bigrams and trigrams) and their combinations (unigrams+bigrams, unigrams+trigrams, unigrams+bigrams+trigrams). In order to identify the features that are most informative and to bring down the computational time of the classification algorithms, feature selection techniques are used. Performance of independent and combined feature sets is assessed using accuracy, precision, recall and F-measure. From the experiments conducted, it is observed that combined features outperformed independent features using Boolean Multinomial Naive Bayes (BMNB) classifier.

Identifying System Location Specifics based on Classification of Worldwide Tweets

As social networking sites are gaining populism across the globe, people are more enthusiastic about sharing their thoughts on Various networking Platforms. Facebook and Twitter have become a leading destination for sharing various kinds of information. In the existing literature the focus is to access the information published in the networking platforms in the real-time, and they do not focus on obtaining the geo-location of the user. Here we propose a monitoring system that classifies the tweets using some reliable techniques which can be used across the globe without any security concerns. As there is a lot of fake news available in the digital form, there is a definite need to access the user information and his geo-location metrics. In this paper, we have introduced Naive Bayes Multinomial classifier and a few other models which performs a spatiotemporal analysis. This study also identifies a comprehensive set of performance metrics which can access the tweet’s country of origin by using eight tweet-inherent features. The outcome of this analysis can be used by various cyber-crime departments to deal with the numerous cybercrime cases on networking platforms, and real-time decisive actions can be taken.

Solving the twitter sentiment analysis problem based on a machine learning-based approach

Twitter Sentiment Analysis (TSA) as part of a text classification task has been widely attended by researchers in recent years. This paper presents a machine learning approach to solving the TSA problem in three phases. In the second phase, a suitable value for representing each feature in the Vector Space Model is determined through the weighted combination of the values obtained from four methods (i.e., Term Frequency and Inverse Document Frequency, semantic similarity, sentiment scoring using SentiWordNet, and sentiment scoring based on the class of tweets). In this manner, finding the percentage of contributions or weights of each method is defined as an optimization problem and solved using a genetic algorithm. Also, the weighted values obtained from four methods are combined based on the Einstein sum as an important T-conorm method. Finally, the performance of the proposed method is tested based on the accuracy of support vector machine and multinomial naive Bayes classification algorithms on four famous Twitter datasets, namely the Stanford testing dataset, STS-Gold dataset, Obama-McCain Debate dataset, and Strict Obama-McCain Debate dataset. The obtained results show the high superiority of the proposed method in comparison with the other methods.

Smart healthcare framework for ambient assisted living using IoMT and big data analytics techniques

Abstract In the era of pervasive computing, human living has become smarter by the latest advancements in IoMT (Internet of Medical Things), wearable sensors and telecommunication technologies in order to deliver smart healthcare services. IoMT has the potential to revolutionize the healthcare industry. IoMT interconnects wearable sensors, patients, healthcare providers and caregivers via software and ICT (Information and Communication Technology). AAL (Ambient Assisted Living) enables integration of new technologies to be part of our daily life activities. In this paper, we have provided a novel smart healthcare framework for AAL to monitor the physical activities of elderly people using IoMT and intelligent machine learning algorithms for faster analysis, decision making and better treatment recommendations. Data is collected from multiple wearable sensors placed on subject’s left ankle, right arm, and chest, is transmitted through IoMT devices to the integrated cloud and data analytics layer. To process huge amounts of data in parallel, Hadoop MapReduce techniques are used. Multinomial Naive Bayes classifier, which fits into the MapReduce paradigm, is utilized to recognize the motion experienced by different body parts and provides higher scalability and better performance with parallel processing when compared to serial processor. Our proposed framework predicts 12 physical activities with an overall accuracy of 97.1%. This can be considered as an optimal solution for recognizing physical activities to remotely monitor health conditions of elderly people.

Performance analysis of various machine learning-based approaches for detection and classification of lung cancer in humans

Lung cancer is one of the most common causes of death among all cancer-related diseases (Cancer Research UK in Cancer mortality for common cancers. http://www.cancerresearchuk.org/health-professional/cancer-statistics/mortality/common-cancers-compared , 2017). It is primarily diagnosed by performing a scan analysis of the patient’s lung. This scan analysis could be of X-ray, CT scan, or MRI. Automated classification of lung cancer is one of the difficult tasks, attributing to the varying mechanisms used for imaging patient’s lungs. Image processing and machine learning approaches have shown a great potential for detection and classification of lung cancer. In this paper, we have demonstrated effective approach for detection and classification of lung cancer-related CT scan images into benign and malignant category. Proposed approach firstly processes these images using image processing techniques, and then further supervised learning algorithms are used for their classification. Here, we have extracted texture features along with statistical features and supplied various extracted features to classifiers. We have used seven different classifiers known as k-nearest neighbors classifier, support vector machine classifier, decision tree classifier, multinomial naive Bayes classifier, stochastic gradient descent classifier, random forest classifier, and multi-layer perceptron (MLP) classifier. We have used dataset of 15750 clinical images consisting of both 6910 benign and 8840 malignant lung cancer related images to train and test these classifiers. In the obtained results, it is found that accuracy of MLP classifier is higher with value of 88.55% in comparison with the other classifiers.

Mixture of latent multinomial naive Bayes classifier

Naive Bayes classifier has been extensively applied in various domains in the past few decades due to its simple structure and remarkable predictive performance. However, it is based on a strong assumption which confines its usage for many real-world applications; conditional independence of attributes given class information. In this paper, we propose mixture of latent multinomial naive Bayes (MLMNB) classifier as an extension of naive Bayes to relax the independence assumption. MLMNB incorporates a latent variable in a predefined Bayesian network structure to model the dependencies among attributes, yet avoids burden complexities of structural learning approaches. We theoretically prove that MLMNB automatically shrinks to naive Bayes classifier whenever conditional independence assumption holds. Expectation-maximization (EM) algorithm is modified for the parameter estimation. The experimental results on 36 datasets from the University of California, Irvine (UCI) machine learning repository show that MLMNB achieves a substantial predictive performance as compared with the state-of-the-art modifications of naive Bayes classifier, in terms of classification accuracy (ACC), conditional log-likelihood (CLL), and area under the ROC curve (AUC).

Analysis and prediction of presynaptic and postsynaptic neurotoxins by Chou's general pseudo amino acid composition and motif features

Presynaptic neurotoxins and postsynaptic neurotoxins are two important neurotoxins isolated from venoms of venomous animals and have been proven to be potential effective in neurosciences and pharmacology. With the number of toxin sequences appeared in the public databases, there was a need for developing a computational method for fast and accurate identification and classification of the novel presynaptic neurotoxins and postsynaptic neurotoxins in the large databases. In this study, the Multinomial Naive Bayes Classifier (MNBC) had been developed to discriminate the presynaptic neurotoxins and postsynaptic neurotoxins based on the different kinds of features. The Minimum Redundancy Maximum Relevance (MRMR) feature selection method was used for ranking 400 pseudo amino acid (PseAA) compositions and 50 top ranked PseAA compositions were selected for improving the prediction results. The motif features, 400 PseAA compositions and 50 PseAA compositions were combined together, and selected as the input parameters of MNBC. The best correlation coefficient (CC) value of 0.8213 was obtained when the prediction quality was evaluated by the jackknife test. It was anticipated that the algorithm presented in this study may become a useful tool for identification of presynaptic neurotoxin and postsynaptic neurotoxin sequences and may provide some useful help for in-depth investigation into the biological mechanism of presynaptic neurotoxins and postsynaptic neurotoxins.

Automatic approval prediction for software enhancement requests

Software applications often receive a large number of enhancement requests that suggest developers to fulfill additional functions. Such requests are usually checked manually by the developers, which is time consuming and tedious. Consequently, an approach that can automatically predict whether a new enhancement report will be approved is beneficial for both the developers and enhancement suggesters. With the approach, according to their available time, the developers can rank the reports and thus limit the number of reports to evaluate from large collection of low quality enhancement requests that are unlikely to be approved. The approach can help developers respond to the useful requests more quickly. To this end, we propose a multinomial naive Bayes based approach to automatically predict whether a new enhancement report is likely to be approved or rejected. We acquire the enhancement reports of open-source software applications from Bugzilla for evaluation. Each report is preprocessed and modeled as a vector. Using these vectors with their corresponding approval status, we train a Bayes based classifier. The trained classifier predicts approval or rejection of the new enhancement reports. We apply different machine learning and neural network algorithms, and it turns out that the multinomial naive Bayes classifier yields the highest accuracy with the given dataset. The proposed approach is evaluated with 40,000 enhancement reports from 35 open source applications. The results of tenfold cross validation suggest that the average accuracy is up to 89.25%.

Prediction of presynaptic and postsynaptic neurotoxins by combining various Chou\u2019s pseudo components

Presynaptic and postsynaptic neurotoxins are two groups of neurotoxins. Identification of presynaptic and postsynaptic neurotoxins is an important work for numerous newly found toxins. It is both costly and time consuming to determine these two neurotoxins by experimental methods. As a complement, using computational methods for predicting presynaptic and postsynaptic neurotoxins could provide some useful information in a timely manner. In this study, we described four algorithms for predicting presynaptic and postsynaptic neurotoxins from sequence driven features by using Increment of Diversity (ID), Multinomial Naive Bayes Classifier (MNBC), Random Forest (RF), and K-nearest Neighbours Classifier (IBK). Each protein sequence was encoded by pseudo amino acid (PseAA) compositions and three biological motif features, including MEME, Prosite and InterPro motif features. The Maximum Relevance Minimum Redundancy (MRMR) feature selection method was used to rank the PseAA compositions and the 50 top ranked features were selected to improve the prediction accuracy. The PseAA compositions and three kinds of biological motif features were combined and 12 different parameters that defined as P1-P12 were selected as the input parameters of ID, MNBC, RF, and IBK. The prediction results obtained in this study were significantly better than those of previously developed methods.

Disambiguating context-dependent polarity of words: An information retrieval approach

The paper introduces PolaritySim – a novel approach to disambiguating context-dependent sentiment polarity of words. The task of resolving the polarity of a given word instance as positive or negative is addressed as an information retrieval problem. At the pre-processing stage, a vector of context features is built for each word w based on all its occurrences in the positive polarity corpus (consumer reviews with high ratings) and another vector – on its contexts in the negative polarity corpus (reviews with low ratings). Lexico-syntactic context features are automatically generated from dependency parse graphs of the sentences containing the word. These two vectors are treated as “documents”, one with positive and one with negative polarity. To resolve the contextual polarity of a specific instance of the word w in a given sentence, its context feature vector is built in the same way, and is treated as the “query”. An information retrieval (IR) model is then applied to calculate the similarity of the “query” to each of the two “documents”, with the polarity of the best matching “document” attributed to the “query”. The method uses no prior polarity sentiment lexicons or purposefully annotated training datasets. The only external resource used is a readily available corpus of user-rated reviews. Evaluation on different domains shows more effective performance compared to state-of-the-art baselines, Support Vector Machines (SVM) and Multinomial Naive Bayes (MNB) classifiers, on three out of four datasets. PolaritySim, SVM and MNB were also evaluated with an out-of-domain training corpus. The results indicate that PolaritySim is more effective and robust when used with an out-of-domain corpus compared to SVM and MNB. We conclude that an IR based approach can be an effective and robust alternative to machine learning approaches for disambiguating word-level polarity using either within-domain, or out-of-domain training corpora.

A model for sentiment and emotion analysis of unstructured social media text

Sentiment analysis has applications in diverse contexts such as in the gathering and analysis of opinions of individuals about various products, issues, social, and political events. Understanding public opinion can help improve decision making. Opinion mining is a way of retrieving information via search engines, blogs, microblogs and social networks. Individual opinions are unique to each person, and Twitter tweets are an invaluable source of this type of data. However, the huge volume and unstructured nature of text/opinion data pose a challenge to analyzing the data efficiently. Accordingly, proficient algorithms/computational strategies are required for mining and condensing tweets as well as finding sentiment bearing words. Most existing computational methods/models/algorithms in the literature for identifying sentiments from such unstructured data rely on machine learning techniques with the bag-of-word approach as their basis. In this work, we use both unsupervised and supervised approaches on various datasets. Unsupervised approach is being used for the automatic identification of sentiment for tweets acquired from Twitter public domain. Different machine learning algorithms such as Multinomial Naive Bayes (MNB), Maximum Entropy and Support Vector Machines are applied for sentiment identification of tweets as well as to examine the effectiveness of various feature combinations. In our experiment on tweets, we achieve an accuracy of 80.68% using the proposed unsupervised approach, in comparison to the lexicon based approach (the latter gives an accuracy of 75.20%). In our experiments, the supervised approach where we combine unigram, bigram and Part-of-Speech as feature is efficient in finding emotion and sentiment of unstructured data. For short message services, using the unigram feature with MNB classifier allows us to achieve an accuracy of 67%.