Browsing by Author "Marouf, Ahmed Al"

Now showing 1 - 20 of 27

A Computer Vision System for Bangladeshi Local Mango Breed Detection using Convolutional Neural Network (CNN) Models
(Scopus, 2020) Haque, A.S. M. Farhan Al; Rahman, Md. Riazur; Marouf, Ahmed Al; Khan, Md. Abbas Ali
Magnifera Indica, traditionally known as mango, is a drupe found around the world in over 500 species. India has produced 19.5 million metric tons of mango in 2017. In Bangladesh, mango has been referred as the national tree and government has included endemic species of mango as geographical index (GI) of Bangladesh. Recognizing specific breeds has become a significant computer vision task. In this paper, we have proposed the convolutional neural network (CNN) based approach for detecting five mango species namely, Chosha, Fazli, Harivanga, Lengra and Rupali from 15000 different images. For better experimentation, we have applied three different models of CNN and analyzed the recognition rates with various criteria. For performance evaluation, we have utilized the classic metrics such as precision, recall, F1-score, ROC and accuracy. Among the experimented three models, the third model, outperformed in terms of accuracy with 92.80%.
An Overall Workflow of Deep Learning in Modern Technology
(International Journal of Innovative Technology and Exploring Engineering, 2019-06) Khan, Thaharim; Rabbani, Masud; Siddiquee, Shah Md. Tanvir; Marouf, Ahmed Al
Modern technology blessed us with many amenities. Those invention of modern science has lessened our workload with many others flexibility. Modern technology has been growing up significantly after the implementation of artificial intelligence (AI) in various sector. Machine Learning (ML), Natural language processing (NLP), Expert system (ES), Computer Vision (CV), Planning & Optimization (P&O), Robotics (RT), Deep Learning (DL), Image Recognition (IR) all are intertwined with AI. Deep learning, one of the most interesting affiliate of AI which terminate the provision of usual invention. One can think of various type of innovation about which only come to their imagination. Some devising work which appears in front of us seems like as a human being is doing that output. The algorithm that deep learning use is more efficient and acceptable for those modern amenities which are called the blessings of modern science. All that devising works in modern age is used by us but could not understand the way how that devising element perform or what the mechanism is. This project gives one a clear view about the deep learning. This paper focus on those algorithms and working procedure which is needed for developing innovative things and which make deep learning more acceptable to us.
Bangla Abusive Language Detection Using Machine Learning on Radio Message Gateway
(2021 6th International Conference on Communication and Electronics Systems (ICCES), IEEE, 2021-08-02) Ritu, Sumaiya Salim; Mondal, Joysurya; Mia, Md. Moinu; Marouf, Ahmed Al
In the era of modern technology, machine learning and natural language processing has been adopted to be applied in several application areas. Natural language processing consists of diversified techniques such as text classification, text summarization, named entity recognition, sentiment analysis. Text classification is considered to be the area of research where the text gets segmented into different category sentences or paragraphs from a single text genre. This paper presents a mechanism for detecting Bangla abusive language from a real-time radio message gateway. Online radio stations nowadays accept communications and voices of their target audience from web-based applications or social media platforms, such as Facebook or Twitter pages. This paper has created a dataset with more than 45000 Bangla sentences, which are labeled as abusive and non-abusive. Sample online radio message gateway has been introduced and machine learning algorithms such as multinomial naive bias (MNB), logistic regression (LR), and random forest (RF) classifiers are utilized to predict the abusive languages. One of the significant prospects of this work would be applied during live radio programs where listeners try to communicate by sending live messages. Our proposed mechanism can check and map the live messages with the dataset and segregate the positive comments or messages only, by filtering the abusive comments. Among the applied classifiers, it has been found that the random forest classifier has performed better than the other two classifiers by leveraging approximately 76% accuracy.
BanglaMusicStylo: A Stylometric Dataset of Bangla Music Lyrics
(IEEE, 2018-12-03) Hossain, Rafayet; Marouf, Ahmed Al
With the rapid growth of Bangla music industry huge volume of Bangla songs are produced every day. Immense number of producers, lyricists, singers and artists are involved in production of songs from different genres. Among many genres of Bangla music; classical, folk, baul, modern music, Rabindra Sangeet, Nazrul Geeti, film music, rock music and fusion music has gained the highest popularity. Lyricists try to express their feelings and views towards any situation or subject through their writings. Therefore, each lyricist have their own dictionary of thoughts to put on music lyrics. In this paper, we have presented “BanglaMusicStylo”, the very first stylometric dataset of Bangla music lyrics. We have collected 2824 Bangla song lyrics of 211 lyricists in a digital form. All the lyrics are stored in text format for further use. This dataset could be used for stylometric analysis such as authorship attribution, linguistic forensics, gender identification from textual data, Bangla music genre classification, vandalism detection, emotion classification etc. Identifying the significant research opportunities in this area, we have formalized this dataset which could be used for stylometric analysis.
Comparative Analysis of Feature Selection Algorithms for Computational Personality Prediction from Social Media
(IEEE Transactions on Computational Social Systems, IEEE, 2020-02-19) Marouf, Ahmed Al; Mahmud, Hasan; Hasan, Md. Kamrul
With the rapid growth of social media, users are getting involved in virtual socialism, generating a huge volume of textual and image contents. Considering the contents such as status updates/tweets and shared posts/retweets, liking other posts is reflecting the online behavior of the users. Predicting personality of a user from these digital footprints has become a computationally challenging problem. In a profile-based approach, utilizing the user-generated textual contents could be useful to reflect the personality in social media. Using huge number of features of different categories, such as traditional linguistic features (character-level, word-level, structural, and so on), psycholinguistic features (emotional affects, perceptions, social relationships, and so on) or social network features (network size, betweenness, and so on) could be useful to predict personality traits from social media. According to a widely popular personality model, namely, big-five-factor model (BFFM), the five factors are openness-to-experience, conscientiousness, extraversion, agreeableness, and neuroticism. Predicting personality is redefined as predicting each of these traits separately from the extracted features. Traditionally, it takes huge number of features to get better accuracy on any prediction task although applying feature selection algorithms may improve the performance of the model. In this article, we have compared the performance of five feature selection algorithms, namely the Pearson correlation coefficient (PCC), correlation-based feature subset (CFS), information gain (IG), symmetric uncertainly (SU) evaluator, and chi-squared (CHI) method. The performance is evaluated using the classic metrics, namely, precision, recall, f-measure, and accuracy as evaluation matrices.
Fingertip Detection and Finger Identification for Real-time Hand Gesture Recognition using Kinect
(Department of Computer Science and Engineering (CSE), Islamic University of Technology (IUT), Board Bazar, Gazipur-1704, Bangladesh, 2014-11-15) Marouf, Ahmed Al; Shondipon, Shaumic
Fingertip Detection and finger identification is one of the main challenges of gesture spotting and recognition. The topic is mostly related to the Human computer Interaction(HCI) area. We have proposed a system for detection of fingertip and identification of finger by several steps like pre-processing, processing, hand segmentation, palm point identification, fingertip detection, finger identification. Pre-processing includes the thresholding ,RGB to Grayscale conversion and color & depth calibration. Determining the minimum depth value from the Kinect camera, determining the segmentation threshold, cropping the region of interest and edge detection are the steps of processing. Depth and color segmentation and calibration is done for hand segmentation. For fingertip detection we have merged two existing idea from state of the art in a manner that it minimizes the limitations of both approach. Ellipse fitting technique is applied for palm point determination and point-to-point scanning is done for fingertip locating. For finger identification, we have proposed a new model, 4Y - model based on iterative Hill climbing Algorithm for finger identification. The main contribution in this area would be the new approaches of fingertip detection and finger identification.
Identifying Neuroticism from User Generated Content of Social Media Based on Psycholinguistic Cues
(2nd International Conference on Electrical, Computer and Communication Engineering, IEEE, 2019-02-09) Marouf, Ahmed Al; Hasan, Md. Kamrul; Mahmud, Hasan
Social media has become a huge repository of textual data and images as each of the users' are creating posts, sharing views or news, capturing the moments via photos etc. Sharing or posting statuses/tweets could be considered as a common feature among the popular social networking sites like Facebook, Twitter, and Google+ etc. User generated textual data such as statuses or tweets could be considered as the essential language to communicate in social media with others. This paper investigates the possibilities of identifying negative personality trait based on the psycholinguistic cues extracted from the language used in social media. Predicting personality traits based on widely accepted framework of Big Five Factor Model (BFFM) is a challenging task. According to the model, there are four positive traits namely openness to experience, conscientiousness, agreeableness and extraversion, while there is only one negative trait neuroticism. The tendency of experiencing negative emotions such as anger, sad, anxiety, depression, instability are referred as neuroticism. We have used psycholinguistic cues extracted using linguistic enquiry and word count (LIWC) for predicting neuroticism. We have applied five different classifiers to evaluate the prediction model.
Leveraging Sensor Fusion and Sensor-Body Position for Activity Recognition for Wearable Mobile Technologies
(Scopus, 2021) Alam, Ashraful; Das, Anik; Tasjid, Shahriar; Marouf, Ahmed Al
—Smart devices like smartphones and smartwatches have made this world smarter. These wearable devices are created through complex research methodologies to make them more usable and interactive with its user. Various interactive mobile applications such as augmented reality (AR), virtual reality (VR) or mixed reality (MR) applications solely depend on the in-built sensors of the smart devices. A lot of facilities can be taken from these devices with sensors such as accelerometer and gyroscope. Different physical activities such as walking, jogging, sitting, etc., can be important for analysis like health state prediction and duration of exercise by using those sensors based on artificial intelligence. In this paper, we have implemented machine learning and deep learning algorithms to detect and recognize eight activities namely, walking, jogging, standing, walking upstairs, walking downstairs, sitting, sitting-in-a-car and cycling; with a maximum of 99.3% accuracy. A few activities are almost similar in action, such as sitting and sitting-in-a-car, but difficult to distinguish; which makes it more challenging to predict tasks. In this paper, we have hypothesized that with more sensors (sensor fusion) and data collection points (sensor-body positions) a wide range of activities can be recognized and the recognition accuracies can be increased. Finally, we showed that the combination of all the sensors data of both pocket/waist and wrist can be used to recognize a wide range of activities accurately. The possibility of using the proposed methodologies for futuristic mobile technologies is quite significant. The adaptation of most recent deep learning algorithms such as convolutional neural network (CNN) and bi-directional Long Short Time Memory (Bi-LSTM) demonstrated high credibility of the methods presented as experimentation.
Leveraging Smartphone Sensors for Detecting Abnormal Gait for Smart Wearable Mobile Technologies
(Scopus, 2021) Tasjid, Md Shahriar; Marouf, Ahmed Al
Walking is one of the most common modes of terrestrial locomotion for humans. Walking is essential for humans to perform most kinds of daily activities. When a person walks, there is a pattern in it, and it is known as gait. Gait analysis is used in sports and healthcare. We can analyze this gait in different ways, like using video captured by the surveillance cameras or depth image cameras in the lab environment. It also can be recognized by wearable sensors. e.g., accelerometer, force sensors, gyroscope, flexible goniometer, magneto resistive sensors, electromagnetic tracking system, force sensors, and electromyography (EMG). Analysis through these sensors required a lab condition, or users must wear these sensors. For detecting abnormality in gait action of a human, we need to incorporate the sensors separately. We can know about one's health condition by abnormal human gait after detecting it. Understanding a regular gait vs. abnormal gait may give insights to the health condition of the subject using the smart wearable technologies. Therefore, in this paper, we proposed a way to analyze abnormal human gait through smartphone sensors. Though smart devices like smartphones and smartwatches are used by most of the person nowadays. So, we can track down their gait using sensors of these intelligent wearable devices. In this study, we used twenty-three (N=23) people to record their walking activities. Among them fourteen people have normal gait actions, and nine people were facing difficulties with their walking due to their illness. To do the stratification of the gait of the subjects, we have adopted five machine learning algorithms with addition a deep learning algorithm. The advantages of the traditional classification are analyzed and compared among themselves. After rigorous performance analysis we found support vector machine (SVM) showing 96% accuracy, highest among the tradition classifiers. 70%, 84%, and 95% accuracy is obtained by the logistic regression, Naïve Bayes, and k-Nearest Neighbor (kNN) classifiers, respectively. As per the state-of-the art, deep learning classifiers has been proven to outperform the traditional classifiers in similar binary classification problems. We have considered the scenario and applied the 2D convolutional neural network (2D-CNN) classification algorithm, which outperformed the other algorithms showing accuracy of 98%. The model can be optimized and can be integrated with the other sensors to be utilized in the mobile wearable devices.
Looking Behind the Mask
(2nd International Conference on Electrical, Computer and Communication Engineering, ECCE 2019, IEEE, 2019-04-04) Marouf, Ahmed Al; Ajwad, Rasif; Ashrafi, Adnan Ferdous
With the facilities of social media platforms like Facebook, Twitter, Google+, YouTube etc. people are capable of expressing their views & news, sharing moments via photos, liking, commenting and sharing others posts. The online social networks (OSNs) are not only giving positive supports to its users, but also creating opportunities to assassin personals by the trolls. Trolls are usually the OSN users who try to hide themselves while doing bad comments, false accusations, starting controversies, spreading fake news or rumors which could be considered as character assassination of individuals. The online behavior of an OSN user could be tracked via his/her digital footprints. Though tracking huge number of users who are generating billions of textual and image data every day, could be considered as a challenging task. In this paper, we have proposed a novel detection system for identifying character assassination from social media platforms. The proposed method first predicts the personality traits using users' textual data. Therefore, LIWC, SlangNet, SentiWordNet, SentiStrength, Colloquial WordNet has been utilized as a psycholinguistic tool. LIWC-based feature engineering has been performed on the comments of the trolls as well as the victim user. SlangNet and Colloquial WordNet are used for detecting English slang words in the comments as it is evident that slangs are the basic communicative way to defame someone.
Lyricist Identification Using Stylometric Features Utilizing BanglaMusicStylo Dataset
(2019 International Conference on Bangla Speech and Language Processing, ICBSLP 2019, IEEE, 2020-05-13) Marouf, Ahmed Al; Hossian, Rafayet
This paper presents a profile-based approach utilizing supervised learning methods to identify the lyricist of Bangla songs written by two legendary poets & novelist Kazi Nazrul Islam and Rabindranath Tagore. The problem statement for this paper could be considered as authorship attribution using stylometric features on Bangla lyrics. We have utilized the BanglaMusicStylo dataset, which consists of 856 and 620 songs of Rabindranath Tagore and Kazi Nazrul Islam, respectively. The traditional authorship attribution works found in the literature are based on the novels written by the authors, not Bangla song lyrics. Using the Bangla song lyrics made it a challenging task, as the word choices made by the authors in songs depends on the rhythms, completeness, situation and many more. In this paper, we have tried to fusion different types of stylometric features, such as lexical, structural, stylistic etc. For experimentation, we have designed the prediction model based on supervised learning exploiting Naïve Bayes (NB), Simple Logistic Regression (SLR), Decision Tree (DT), Support Vector Machine (SVM), and Multilayer Perceptron (MLP). The experimental model consists of several steps including data pre-processing, feature extraction, data processing, and classification model. After performance evaluation, we have got approximately 86.29% accuracy from SLR, which is quite satisfactory.
Mango Species Detection from Raw Leaves Using Image Processing System
(Springer, 2021) Hena, Most. Hasna; Sheikh, Md. Helal; Reza, Md. Shamim; Marouf, Ahmed Al
Mango is the national tree of Bangladesh which is one of the most popular fruits here during the hot summer enriching the highest quality of nutrition. Various species of mango cover the fruit market making the summer festivities. In recent times, different species of mango are also being exported to different countries of the world. So more and more people are entering into the commercial mango cultivation nowadays as new farmers. It is necessary for them to know which mango species they are cultivating and what is the market demand of that species. It is hard for the new farmers to find out the species just by asking and trusting the sapling seller. So, we plan to establish a system that can accurately ensure the species of the mango sapling. This research used convolutional neural network (CNN) and deep learning for training the dataset. This method can showcase the species of the mango sapling only by observing the image of a leaf holding an accuracy of 78.65%.
MCNN-LSTM: Combining CNN and LSTM to Classify Multi-Class Text in Imbalanced News Data
(IEEE, 2023-08-29) Hasib, Khan Md.; Azam, Sami; Karim, Asif; Marouf, Ahmed Al; Shamrat, F M Javed Mehedi; Montaha, Sidratul; Yeo, Kheng Cher; Jonkman, Mirjam
"Searching, retrieving, and arranging text in ever-larger document collections necessitate more efficient information processing algorithms. Document categorization is a crucial component of various information processing systems for supervised learning. As the quantity of documents grows, the performance of classic supervised classifiers has deteriorated because of the number of document categories. Assigning documents to a predetermined set of classes is called text classification. It is utilized extensively in a wide range of data-intensive applications. However, the fact that real-world implementations of these models are plagued with shortcomings begs for more investigation. Imbalanced datasets hinder the most prevalent high-performance algorithms. In this paper, we propose an approach name multi-class Convolutional Neural Network (MCNN)-Long Short-Time Memory (LSTM), which combines two deep learning techniques, Convolutional Neural Network (CNN) and Long Short-Time Memory, for text classification in news data. CNN’s are used as feature extractors for the LSTMs on text input data and have the spatial structure of words in a sentence, paragraph, or document. The dataset is also imbalanced, and we use the Tomek-Link algorithm to balance the dataset and then apply our model, which shows better performance in terms of F1- score (98%) and Accuracy (99.71%) than the existing works. The combination of deep learning techniques used in our approach is ideal for the classification of imbalanced datasets with underrepresented categories. Hence, our method outperformed other machine learning algorithms in text classification by a large margin. We also compare our results with traditional machine learning algorithms in terms of imbalanced and balanced datasets."
Performance Evaluation of Traditional Classifiers on Prediction of Credit Recovery
(Scopus, 2020) Pradhan, Mohammad Rajib; Akter, Sima; Marouf, Ahmed Al
In the era of Big Data, machine learning is an emerging technique to analyze the large volume of data and is used to make critical business decisions. It is broadly used in different area such as medical, telecom, social media, banking data analysis and so on to learn the data and perform predictive analysis as well as building recommendation system. With the progression of technology, data availability and computing power, most of the banks and financial institutions are adapting their business model with technological development. Credit risk analysis is a cardinal field for banking and financial institutions, and there are numerous credit risk technique exists to predict the creditworthiness of the customer and loan default probability. In this study, we explore credit defaulter dataset of Bangladeshi bank and conduct several traditional machine learning classifier to predict the delinquent clients who possessed the highest probability of short-term credit recovery. Furthermore, we perform feature engineering to identify the important features for credit recovery prediction. We then apply our final features on different machine learning classifier and compare the predictive accuracy with the other classifier. We observe that random forest classifier gives 90% accuracy in credit recovery prediction. Finally, we propose a noble strategy to identify the potential customer for recovering the credit amount by using supervised machine learning techniques.
Predicting and Staging Chronic Kidney Disease of Diabetes (Type-2) Patient Using Machine Learning Algorithms
(International Journal of Innovative Technology and Exploring Engineering, Blue Eyes Intelligence Engineering & Sciences Publication, 2019-10-02) Basak, Setu; Alam, Md. Mahbub; Rakshit, Aniruddha; Marouf, Ahmed Al; Majumder, Anup
Mortality because of unending kidney disease increments essentially in recent years. Nowadays, about 422 million patients are suffering from diabetes among them around 30 percent of patients with Type 1 (adolescent beginning) diabetes and around 10 to 40 percent of those with Type 2 (grown-up beginning) diabetes in the end will experience the negative impacts of kidney damage. It is evident, that early detection of Chronic Kidney Disease (CKD) can mitigate the level of damage in the adulthood. In this paper, we have presented a comparative analysis based on the performance of five different algorithms-Naive Bayes (NB), In-stance Based Learning (IBK), Random Forest (RF), Decision Stump (DS) and Decision Tree (J48) for predicting CKD of diabetes patients only by urine test. Among all the algorithms the IBK gives the best result. Our comparison of different algorithms will help people with diabetes to find out if they are having CKD or not.
Predicting Users’ Personality from Social Media using Linguistic and Social Network Features
(Department of Computer Science and Engineering, Islamic University of Technology, Gazipur, Bangladesh, 2019-11-15) Marouf, Ahmed Al
Social media such as Facebook, Twitter, Google+ etc. has become a huge repository of textual data and images as each of the users’ are creating posts, sharing views or news, capturing the moments via photos etc. User generated textual data such as statuses can be considered as the essential language to communicate in social media with others. Predicting personality traits from these social media data is a sophisticated task performed in computational social science. Among several personality prediction models, the Big Five Factor Model is one of the widely used personality traits hypothesis used by computational psychologists. The five traits that are centered for identifying ones personality are Openness-to-experience (O), Conscientiousness (C), Extraversion (E), Agreeableness (A), and Neuroticism (N). The first four traits are considered as positive traits and the only negative personality trait is neuroticism. In this thesis, we have focused on predicting these personality traits utilizing linguistic & social network features and identifying the prominent features using feature selection algorithms for each of the traits separately. We have evaluated the efficiency of machine learning techniques using the extracted features. To determine the most prominent features for individual personality traits and features that are commonly found in every personality traits, manual and automated feature selection has been applied. It is anticipated that the analysis reported in this study can be applied to develop personalized recommendation systems in social media, predicting personality disorder and identifying the trust issues in social media.
Real-time Bangladeshi Currency Detection System for Visually Impaired Person
(2019 International Conference on Bangla Speech and Language Processing, ICBSLP 2019, IEEE, 2020-05-13) Sarker, Md. Ferdousur Rahman; Raju, Md. Israfil Mahmud; Marouf, Ahmed Al; Hafiz, Rubaiya; Hossain, Syed Akhter; Protik, Munim Hossain Khandker
This paper presents a real-time Bangladeshi currency detection system for visually impaired persons. The proposed system exploits the image processing algorithms to facilitate the visually impaired people to prosperously recognize banknotes. The recent banknotes of Bangladesh have blind embossing or blind dots, which could be effective to recognize the value of the bill by touching. As the embossing fades away in the long-term used notes, detecting right value of the banknote using image processing algorithms could be considered as a challenging task. Particularly in Bangladesh, each banknote seems similar using the direct exertion of simplified image processing algorithms. In this paper, a recognition system was implemented that can detect Bangladeshi banknote in different viewpoints and scales. The detection system is also able to detect currency those are rumpled, decrepit or even worn. The detection system includes image preprocessing, image analysis and image recognition. To enhance the determination of currency recognition, the descriptor of an individual input scene is matched with various training images of the same category. After that, by analyzing their matching result it recognizes the currency with higher confidence. For real-time recognition, we have deployed the system into a mobile application.
Recognizing Hand-based Actions based on Hip-Joint centered Features using KINECT
(IEEE, 2018-07-19) Marouf, Ahmed Al; Sarker, Md. Ferdousur Rahman; Siddiquee, Shah Md. Tanvir
Microsoft Kinect provides skeletal joints to extract different features which can be applied to identify different actions performed by subjects. As human moves, skeletal joints contribute to the movements and human actions are nothing but different types of movements in specific orders. Hand wave, hand shaking, push, pull, clapping, throw, catch these are some hand based actions which are difficult to recognize properly in an automated system. Hip-joint plays a vital role to determine joint-based features from human skeleton, as it is approximately the middle joint of the whole skeleton. The joint relative distances (JRD) and joint relative angles (JRA) are used as principle features in recent action recognition methodologies. In this paper, we have proposed a new methodology based on hip-joint centered features which are based on basic physiological movements that contributes to the decent accuracy in identifying hand based actions.
Recognizing Language and Emotional Tone from Music Lyrics Using IBM Watson Tone Analyzer
(Proceedings of 2019 3rd IEEE International Conference on Electrical, Computer and Communication Technologies, ICECCT 2023, IEEE, 2019-10-17) Marouf, Ahmed Al; Hossain, Rafayet; Sarker, Md. Rahmatul Kabir Rasel; Pandey, Bishwajeet; Siddiquee, Shah Md. Tanvir
Music has a soothing impact on listener's mood and emotional states. Apart from the rhythm, sequence, instrumental effects on a song, lyrics could be considered as the most vital element. Lyricists' mood and affection towards a song while writing could be understand from the lyrics. Lyrics does have the elements of fictions such as language tone, language style, diction and voice are well maintained in music lyrics. Understanding the tone of a song both language and emotional tones are essential to develop different interactive applications. Music players, video repositories, video sharing sites could use the understandings to recommend next song to play according to the music interest or mood of the listeners. In this paper, we have investigated the possibilities to use IBM Watson Tone Analyzer, an API service to analyze language and emotional tones from song lyrics. We have extracted the features from a 300 English song dataset using the supported API service and formulated a machine learning methodology to classify the language tone (analytical, confident and tentative) and emotional tone (anger, fear, joy and sadness). For classification, we have applied different classifiers including Naïve Bayes, decision tree, random forest, sequential minimal optimization and simple logistic regression.
Recommendation Approach of English Songs Title Based on Latent Dirichlet Allocation Applied on Lyrics
(Proceedings of 2019 3rd IEEE International Conference on Electrical, Computer and Communication Technologies, ICECCT 2019, IEEE, 2019-10-17) Hossain, Rafayet; Sarker, Md. Rahmatul Kabir Rasel; Mimo, Mehejabin; Marouf, Ahmed Al; Pandey, Bishwajeet
The significance of music has evolved due to the vast diversity of entertainment industry. Songs are the widely used entertainment segment that can influence directly to the heart of the listeners. Choosing a suitable title for a song is considered as a common problem faced by the music directors. As the title gives the first impression of the song and only by the title listeners usually decide whether they will listen to this song or not, thus makes it a challenging task to determine. Lyrics are the most influential part of a particular song apart from the tune, rhythm, fusion, singer, genre etc. In this paper, we propose an approach to estimate and recommend the title of the song based on its lyrics. We have applied Latent Dirichlet Allocation (LDA) to find the hidden or implied topic of the song. The output of the LDA algorithm provides scoring on the significant words, which are passed to an estimation process to generate a song title. The proposed approach was experimented on over 200 English songs database having vast diversity in genre. The approach could be evaluated by the existing song title and the evaluation process is same as any recommendation system.