جستجوی مقالات مرتبط با کلیدواژه « Convolutional neural networks » در نشریات گروه « علوم انسانی »

Today, people face different decision-making criteria when purchasing products and services. One of these criteria is using the reviews of the previous purchasers of products and services. A large volume of reviews is seen as a challenge for these people. The present study aimed to create a model to analyze users’ sentiments and to classify their reviews to solve the mentioned challenge.

The present study investigated the buyers’ reviews of mobile phones purchased on the Digikala Website from 2015 to 2016. To analyze the sentiments, and to classify the reviews, deep learning-based algorithms, and convolutional networks, subtypes of deep networks, were suggested. Prior to preprocessing and homogenizing the data, the study used a pre-trained Fastext model to convert the words into integer vectors and deliver them as inputs to the proposed deep network.

To train the selected model, the training algorithm was carried out on it 90 times. To validate the performance of the selected model, confusion matrix, accuracy, recall, F1-score, and precision rate criteria were used.

The present study used the deep networks approach, convolutional networks, and bidirectional long short-term memory to classify the buyers’ reviews of the mobile phone from the website above at 93% accuracy, and after 90 training periods.

Convolutional neural networks, as one of the most important models of deep learning, have gained high accuracy on these issues. In this study, discussion and text analysis at the sentence level and improving the performance of neural networks to detect fake news has been convolution. The network of words for bags of words in the data model so that each word according to the two-dimensional vector space to become matrices. One of the limitations of convolutional networks is that it works at the word level and cannot consider the relationship and distance between sentences. And sentence-level analysis is a major problem in this research. Sentence level analysis is a major problem in this research.In this research, a basic model based on convolutional networks is proposed in which documents are given to the network in the form of 3D tensors to solve the mentioned problem. Considering 3D tensors allows the model to learn the position of words in a sentence and achieve more accurate results in detecting fake news.

This study is applied research in which about 42,000 Persian news from different cities of Iran were collected from Twitter and using preprocessing, additional and useless data is deleted and after tagging the deleted texts, the news text is used for the proposed approach using Python software and related libraries.

During testing, some machine learning algorithms had more power in classification problems, but with the changes in the structure of the convolutional network algorithm, better results were obtained than machine learning algorithms and other similar algorithms.

Considering 3D tensors allows the model to learn the position of words in a sentence, and this proposed model has gained considerable accuracy compared to the proposed approaches in the literature. The proposed model without adding additional overhead in terms of the number of features and network depth, by changing the input has been able to achieve better and more acceptable results than other approaches in the literature and achieve an accuracy of more than 94%.

The development of automatic road and building detection systems in aerial imagery are always faced with challenges such as the appearance of buildings, illumination changes, imaging angles, and the density of roads and buildings in urban areas, to name a few. In recent years, employing multi-layered approach in artificial neural networks, known as deep neural networks, has attracted many researchers in this field (and the other fields alike), achieving stunning results. However, the use of fully connected layers in this approach, significantly increases the average processing time and results in an overfitted model. In addition, in most of these methods, a single-class approach has been considered. That is, detecting the roads and the buildings from natural scenes is not possible at the same time, and therefore, it is necessary to build separate binary models for each of them. The main goal of this research is to design a new architecture by which the produced model can be able to simultaneously detect roads and buildings from natural scenes, and thus minimizing the complexity of the classification process. In addition, in the proposed architecture, excluding all fully connected layers from the traditional multi-layered architectures is considered in order to reduce the average processing time. The results of the experiments performed on the Massachusetts dataset, show that the proposed architecture performs 38% faster than the other deep neural network-based methods, and also increases the accuracy by an average of 2%.

Convolutional neural networks are being applied to identification problems in a variety of fields, and in some areas are showing higher discrimination accuracies than conventional methods. Hence, in this research, an attempt is made to apply a convolutional neural network to the prediction of corporate bankruptcy. The financial statements ratios has been choice 66 companies that have been delisted from the Iran Stock Market due to de facto bankruptcy as well as the financial statements of 66 listed companies over 2000 to 2019 financial periods. In this method, a set of financial ratios are derived from the financial statements and represented as a grayscale image. The image generated by this process is utilized for training and testing a convolutional neural network. The images for the bankrupt and continuing enterprises classes are used for training the convolutional neural network based on GoogLeNet. The findings shows, in prediction of going concern of firms, Convolutional neural network has predicted with 50 percent of precision. This means that 50 percent of continues firms and 50 percent of bankrupt firms has been predicted precisely.

In recent years, many articles and researches have been published on the use of machine learning methods and algorithmic trading in financial markets in order to earn returns. The aim of this study is to create an automated trading system using image processing by convolutional neural network. For this purpose, initially, after receiving the data required for the selected stocks, 28 technical analysis indicators were selected and the values of each were calculated separately for each stock. Then the time series of these indicators were converted to 2D images, and as a result, for each data on the time series of the stock price, a 2D image with dimensions of 28 x 28 was created. After labeling each image with one of the buy, sell, or hold labels, these images entered the convolutional neural network. Also, to evaluate the return and risk of the proposed system, a method for buying and selling based on the results of the model in the past has been introduced. The results show that in 80% of cases, this method is more effective than the buy and hold strategy. It also always performs better in terms of standard deviation risk and maximum drawdown. Also, the results show the high impact of trading commission on the Tehran Stock Exchange on the return of the model. In such a way that the model loses many times the profit earned for the payment of the commission.