M.A. Theses
Permanent URI for this collection
Browse
Browsing M.A. Theses by Subject "Artificial intelligence."
Now showing 1 - 2 of 2
Results Per Page
Sort Options
Item A hybrid article recommendation system based on deep learning and co-publication network analytics(Thesis (M.A.) - Bogazici University. Institute for Graduate Studies in the Social Sciences, 2019., 2019.) Atlanel, Büşra.; Durahim, Ahmet Onur.In recent years, with the rapid development of world wide web, researchers are spending more effort and time to reach the most relevant academic work for their studies because of the information overload. Preventing users from being distracted by a tremendous amount of publications and simplification of the research process makes recommendation systems more valuable. Traditional recommendation systems generally suffer from limited coverage, data sparsity, and cold start problem. In order to tackle these problems and achieve better performance, many recommender systems started to use neural network models. Being an effective neural network model, deep learning technology can transform article titles and abstract information into text embeddings and capture non-linear relationships between these text embeddings. In addition to deep learning on text embeddings, the relationship between authors has a huge effect on their future preferences. The research of copublication relationship with social network analysis improves the performance of the recommendation systems. In this study, the aim is to propose a hybrid article recommendation system that incorporates deep learning for article text similarity using Deep Siamese BiLSTM and social network analysis through node embeddings using co-publication and citation networks to exploit the network structure to provide benefit for recommender systems. Experiments conducted in this research show that the proposed model achieved a prediction rate of 7% on average when the number of articles to be recommended is taken as 100.Item Prediction of stock price direction by artificial neural network approach(Thesis (M.A.)-Bogazici University. Institute for Graduate Studies in the Social Sciences, 2008., 2008.) Şenol, Doğaç.; Özturan, Meltem.The stock market has always been an attractive area for researchers since no method has been found yet to predict the stock price behavior precisely. It carries a higher risk than any other investment area, due to its high rate of uncertainty and volatility, thus making the stock price behavior difficult to forecast. For years, conventional methods have been developed but they have succeeded partially or have completely failed to deal with the nonlinear and complex behavior of stock prices. Artificial neural networks approach is a relatively new, active and promising field on the prediction of stock price behavior. Artificial neural networks (ANNs) are mathematical models simulating the learning and decision making processes of the human brain. Because of their nature of easy adaptation to noisy data, and solving complex and nonlinear problems, they fit into the area of stock price behavior prediction. The Istanbul Stock Exchange (ISE) is the only stock market in Turkey, which has an emerging economy. The market situations and economic fluctuations in Turkey create more uncertainty and volatility in the stock market when compared to emerged markets. This study tries to reduce the effect of this uncertainty and volatility by modeling the change in stock price direction of stocks, identifying the theory and steps involved in applying ANN in financial markets and developing a software package to be used for predicting directional daily stock price behavior. It also discusses the appropriate ways to use this process in developing trading systems, further discussing the superiority of ANN over traditional methodologies.