House price prediction is a popular topic, and research teams are increasingly performing related studies by using deep learning or machine learning models. However, because some studies have not considered comprehensive information that affects house prices, prediction results are not always sufficiently precise. Therefore, we propose an end to end joint self-attention model for house prediction. In this model, we import data on public facilities such as parks, schools, and mass rapid transit stations to represent the availability of amenities, and we use satellite maps to analyze the environment surrounding houses. We adopt attention mechanisms, which are widely used in image, speech, and translation tasks, to identify crucial features that are considered by prospective house buyers. The model can automatically assign weights when given transaction data. Our proposed model differs from self-attention models because it considers the interaction between two different features to learn the complicated relationship between features in order to increase prediction precision. We conduct experiments to demonstrate the performance of the model. Experimental data include actual selling prices in real estate transaction data for the period from 2017 to 2018, public facility data acquired from the Taipei and New Taipei governments, and satellite maps crawled using the Google Maps application programming interface. We utilize these datasets to train our proposed and compare its performance with that of other machine learning-based models such as Extreme Gradient Boosting and Light Gradient Boosted Machine, deep learning, and several attention models. The experimental results indicate that the proposed model achieves a low prediction error and outperforms the other models. To the best of our knowledge, we are the first research to incorporate attention mechanism and STN network to conduct house price prediction.
Many researchers have incorporated deep neural networks (DNNs) with reinforcement learning (RL) in automatic trading systems. However, such methods result in complicated algorithmic trading models with several defects, especially when a DNN model is vulnerable to malicious adversarial samples. Researches have rarely focused on planning for long-term attacks against RL-based trading systems. To neutralize these attacks, researchers must consider generating imperceptible perturbations while simultaneously reducing the number of modified steps. In this research, an adversary is used to attack an RLbased trading agent. First, we propose an extension of the ensemble of the identical independent evaluators (EIIE) method, called enhanced EIIE, in which information on the best bids and asks is incorporated. Enhanced EIIE was demonstrated to produce an authoritative trading agent that yields better portfolio performance relative to that of an EIIE agent. Enhanced EIIE was then applied to the adversarial agent for the agent to learn when and how much to attack (in the form of introducing perturbations).In our experiments, our proposed adversarial attack mechanisms were > 30% more effective at reducing accumulated portfolio value relative to the conventional attack mechanisms of the fast gradient sign method (FSGM) and iterative FSGM, which are currently more commonly researched and adapted to compare and improve.
Because deep learning models have been used successfully in various fields during recent years, many recommendation systems have been developed using deep learning techniques. However, although deep learning-based recommendation systems have achieved high recommendation performance, their lack of interpretability may reduce users' trust and satisfaction. In this study, we aimed to predict and recommend the purchase of funds by customers in the next month while simultaneously providing relevant explanations. To achieve this goal, we employed a knowledge graph structure and deep learning techniques to embed features of customers and funds into a unified latent space. With the proposed structure, we learned some information that could not be learned using traditional deep learning models and obtained personalized recommendations and explanations simultaneously. Moreover, we obtained complex explanations by changing the training procedure of the model and developed a measure for rating the customized explanations according to their strength and uniqueness. Finally, we obtained some possible special recommendations based on the knowledge graph structure. By evaluating the data set of mutual fund transaction records, we verified the effectiveness of the developed model for providing precise recommendations. We also conducted some case studies of explanations to demonstrate the effectiveness of the developed model for providing usual explanations, complex explanations, and other special recommendations.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.