Research Transcripts in Computer, Electrical and Electronics Engineering

Enhanced Text Clustering Approach using Hierarchical Agglomerative Clustering with Principal Components Analysis to Design Document Recommendation System

Gauri Chaudhary — 2021-03-30

Considering the increased usage and our increasing dependency in today’s world on electronic data, substantial part of which is in textual form, it becomes necessary to devise scientific methods to infer and extract knowledge from such abundant electronic documents for strategic decision making in any target domain under consideration. The purpose of this study is to develop a common platform where all the similar text from multiple source documents from internet can be fetched and grouped using text mining and document clustering techniques. This chapter elaborates the method of hierarchical agglomerative text clustering approach to identify similar groups within documents. The method of Principal Components Analysis on text data is also further elaborated. Further combination of the two methods is proposed to find suitable clusters in text data and the results obtained show better quality clusters. For the purpose of experiments, plot summaries of movies from Wikipedia are used as the source document corpus. Various document pre-processing techniques are also explained and applied to the documents. The proposed method to get suitable clusters of similar movies can be used for recommendation to users. R programming is used for implementation of algorithms and visualization of the results.

Elucidation of Bangla Language Classification Using Neural Network Approach

Pritom Sarker — 2021-03-30

Bangladesh has two principal languages called Sadu and Cholit. In the early times, Sadhu was operational and was composed of Sanskrit components but the current era has shifted to Cholit language, which is now being used most commonly. Sadhu was mostly used for formal documentation purposes and it is the need of the hour to translate them to Cholit language because it is more speaker friendly and can be easily understandable. Therefore, in this chapter efforts were done to familiarize the current era with the Sadhu language by creating software. Few sentences were selected and the final dataset was obtained by Principal Component Analysis (PCA). Python is used for different machine learning algorithms. Maximum work was done on Scikit Learn which is Term Frequency-Inverse Data Frequency (TF-DF) Vectorizer’s class. The best performance was given by Neural Network with high precision. Speed was also anticipated and values were determined through graphs. The results showed that it translated all words from Sadhu to Cholit efficiently and in a well-oriented way. Therefore, Sadhu’s complexity has been removed in this era.

Facial Expression Recognition Using Neural Network

Md. Forhad Ali — 2021-03-30

Human emotions are states of mental health that resolve spontaneously rather than through conscious exertion, and are accompanied by physiological changes in the facial muscles that signify expressions. Nonverbal communication methods such as expressions, eye movements, and gestures are used in many applications of human-computer interaction. Identifying emotions is not an easy task because there is no difference between the emotions of a face, and there is also a lot of complexity and variability. The machine learning algorithm uses some open features to model the face. In this work, convolutional neural networks (CNNs) were developed to identify the expression of facial emotions. Facial expressions play an important role in the nonverbal communication that takes place in a person’s inner emotions that are reflected on his or her face.This work has been used the Viola-Jones algorithm to detect the eye and lips region from a face and then with the help of the neural network. Also, Machine Learning techniques, Deep Learning models, and Neural Network algorithms are used for emotion recognition. This work will be proposed as an effective way to detect anger, contempt, disgust, fear, happiness, sadness, and surprise.

Development Software for Preprocessing Voice Signals

Niyozmatova N — 2021-03-30

At present, the most important task of modern science is the creation for a person of natural means of communication with a computer, where speech input of information is carried out in the most convenient way for the user. Speech recognition is one of the challenges. As practice shows, the quality of recognition depends on the properties of the preprocessing system. To improve the quality of recognition, it is necessary to develop effective and high-speed methods and algorithms for signal preprocessing. This article proposes a new approach and algorithm for extracting features of speech signals. Based on the proposed algorithm, the identification problem is solved. In addition, the chapter presents a description of the software module for each stage of the preliminary processing of speech signals. This software is a voice-based identity tool.

Application of Classifiers for Assortment of Online Reviews

Biplob Kumar — 2021-03-30

In Bangladesh, Ecommerce is flourishing day by day especially in the time of crisis the world is facing. There are many platforms available on these sites among which Daraz is the most successful marketplace. This online platform allowed people the ease to do shopping but a large number of reviews and comments made it difficult to opt for the best option. In this paper, the focus is on cataloguing the positive and negative reviews. For this purpose various classifiers were used by using Python. Data cleaning was done and after application of Term Frequency -Inverse Data Frequency with Principal Component analysis it was found that Ridge classifier performed best with more training time then other classifiers and depicted high accuracy. This classifier could help different businesses on different platforms to identify the positive and negative reviews and can provide customers with details about the quality of products.

The Method of Depth Map Calculating Based on Soft Operators in Multi-Agent Robotic Stereo Vision Systems

M. V. Bobyr — 2021-03-30

The fuzzy method of depth map calculating using stereo images obtained on the path of mobile robots (agents) is considered. The method is based on SAD (sum of absolute difference) algorithm composition and fuzzy inference. Its special feature is soft arithmetic operators with fuzzy implication usage. The accuracy of the constructing depth map method is estimated by RMSE (root mean square error). The bestest soft operator has minimum RMSE. The method of depth map calculating which has seven steps is presented. The proposed method has showed that the accuracy of the SAD algorithm increases by 20% when soft operators is used. This conclusion is confirmed by the simulation results presented in the chapter.

Bandwidth Control Sectoring Technique Protocol for Data Dissemination in Wireless Sensor Networks

Deepika U. Shevatkar — 2021-03-30

The present chapter relates to a system for data collection in wireless sensor networks. The object is to provide an improved Quality of Services using bandwidth control sectoring technique for WSNs. To reduce the power consumption through system. The nodes are randomly deployed in the network. There are sectors formed according to the equal number of sector heads. There is one sink node, which collects data from sector heads. Common nodes are deployed randomly to transmit data packets to the respective level 1 node within a sector. Thus, bandwidth will be controlled, and congestion in the network is reduced. Bandwidth control sectoring technique (BCST) achieves the various quality parameters of Network. QoS like period of time (delay), power resource efficiency, delivery ratio, loss ration including throughput. This chapter shows the best protocol for data dissemination in wireless sensor networks. This is designed for achieving maximum QoS of wireless sensor networks.