Sentiment Analysis of Twitter for Recommender System
1 Sajida Fayyaz, 2Amatul Musawir, 3Hafiz Ali Hamza Gondal*, 4 Syed Muhammad Mahdi
Abstract: The web is flooded with evaluative text that proves to be worthwhile resource of opinions available on different products, markets, occasions or individuals and so on. People are more interested to shop online and prefer to go through user reviews before making a purchase. Opinions within reviews prove to be functional for manufacturers as well that it may assist enhancing the design plus quality. Numerous research works have been carried out, on diverse sort of reviews available online for various products, named as text mining and opinion mining. Few earlier approaches to calculate polarity are keyword spotting (identifying the keywords from certain text), the lexical affinity (probabilistic affinity used to allocate arbitrary words for a specific category to demonstrate either result is negative or else positive), the statistical method (functions on distinct patterns as well as word or event co-occurrences) plus concept level techniques (make use of semantic networks to infer information from concepts of natural language). Recommender systems so far have not mechanized for hybrid cars yet. The system of recommendation is to propose the latest trending hybrid cars. Important parameters are to be discovered from the information, experience and feedback of twitter users about hybrid cars. Investigating every single review even for a single hybrid car is not that much easy. Hence this study concentrates on sentiment classification techniques among which lexicon-based dictionary is implemented to carry out Natural Language processing (NLP) for optimizing the outcomes. System recommends on the basis of polarity mapping and mining fractional parts of data to designate in terms of positive or negative review. The obtained percentage of positivity of each car model after comparison determines which hybrid brand is prominent and trending.
Smart Anti-Theft Vehicular Security and Accident Detection System using GPS and GSM Technology
*Adnan Shamim, Eurusha Pious, Muhammad Adil
Abstract: The electronic road safety system is presented in this paper, which can prevent automobile theft by using GPS and GSM technology. This system provides security by using ignition cut off mechanism. GSM technology is used to assist the owner in critical situations like theft by sending an alert message to the owner. If the ignition is switched on by an unauthorized individual then owner can respond by switching off the ignition by replaying to the GSM module. The proposed framework can call paramedics in the case of an accident
Sound Classification using Multilayer Neural Network
Manzar Iqbal, Muhammad Zakir Khan, Mumtaz Ali, NaimatUllah
Abstract: Human environment consist of a mix-up of different sounds having different frequencies and temporal structure. These sounds must be classified for better understanding. The multi-layer neural network is able to classify sounds by extracting different features of sound. A multi-layer neural network consists of 3 layers (input layer, hidden layer, and output layer). Spectrogram and time-frequency graphs are responsible to find the frequency of each sound which helps in the classifying of the sound base of their frequency and amplitude. The accuracy of the model is evaluated on a data set called UrbanSound8k which is consists of 8,733 libelled sounds divided into 10 different classes. The goal of this paper is to classify environmental sounds into many meaningful sounds. This approach can further be used in the different sound applications of daily life.
Enhancing Content Based Image Retrieval Systems by using CNN as Feature Extractor
Amjad Shah1 , Yasir ali1 ,
Abstract: The Contents Based Image Retrieval (CBIR) has become a daunting task due to significant evolution in multimedia contents and its related visual complexity. CBIR encompasses various phases from query to the retrieval of images but the most significant phase is feature extraction. In a bid to enhance the accuracy of CBIR, numerous studies have been performed but the output of Convolution Neural Networks (CNN) is amazing in the domain of computer vision. CNN has the potential of extracting features and it can be applied as classifier as well. In our proposed work, the CNN has been applied as feature extractor in place of traditional feature extractors. Similarity measurement was calculated with the help of Euclidian distance formula, precision and recall were used for measuring the performance of the system. In this paper it has been discussed that how our proposed work produces better results as compared to the previous works
