Search

The revolution of web technologies and e-commerce platforms such as Amazon and eBay have promoted the businesses and have become essential in our daily life. Despite that these tools have helped the ease of purchases, there is a lot of scams in these kinds of technologies. Unfortunately, several companies use fake opinions to influence the customers about buying a product or to demote the competitors’ one. The detection of deceptive opinion spam is a hard task because of the way it is written. The majority of existing deceptive opinion detection models focus on machine learning with hand-engineered feature extraction. Unfortunately, these architectures do not provide the semantic information of the reviews, which is the key to the detection of deceptive opinions. In this paper, we address the comparison between the different neural network architectures and their effectiveness in the detection of deceptive opinion spam. The results show that Convolutional Neural Networks perform better compared to other models.

Ensuring an adaptable and interactive tools to analyze data objects is an advisable objective of machine learning algorithms. Many methods exist, and new methods, or improvements in existing ones are proposed regularly to deal with a variety of problems in different areas. We develop a variant of the well-known Possibilistic Fuzzy c-Means Clustering algorithm PFCM that takes into account the observer-biased framework, Possibilistic fuzzy c-means with focal point PFCMFP. the accuracy of the proposed method is verified by cluster validity measures. The experimental results have shown that the accuracy of the new method increases significantly, compared to the initial PFCM algorithm. To elaborate this study, we have used a dataset of individual household electric power consumption, that is accessed publicly at the UCI Machine Learning Repository.

Given the current energy challenge, renewable energy appear as a real and strategic solution for electricity generation, but its intermittent nature requires us to combine two or more power sources to ensure continuity of electricity supply. In this paper we present a multi-agent system based on wind and photovoltaic power prediction using neural network trained by Extreme Learning Machine algorithm, to estimate the amount of photovoltaic and wind energy produced by photovoltaic panels and wind turbines respectively. This process aims to implement a hybrid renewable energy system with generation units (wind turbines and photovoltaic panels) and storage units (batteries). To address uncertainties in the system, we use a fuzzy logic technique to decide which auxiliary energy source will fill the need energetic. For the simulation, we use Java Agent Development Framework to implement the approach and analyze the results.

International audience

According to Microsoft, by 2025, 100% of new cars will be connected and by 2030, 15% of new cars will be autonomous, and will take care of sending, receiving and analyzing “large amounts of data”. Automobiles are becoming data centers on wheels. All these pieces of information can be used by many stakeholders (road authorities, leasing companies, municipalities, car manufacturers, insurers, workshops, emergency services, etc.) in the mobility space to improve processes. Taxi’s industry is a concrete example that is evolving rapidly. New competitors and technologies are changing the way traditional taxi services do business. In spite of this growth has created new effectualness, it has also created new challenges. Exploiting and understanding of the taxi supply and demand could increase the efficiency of the city’s taxi fleet management system. This paper describes a recommender system that recommends the most relevant pick-up points for a driver. It is based on spatiotemporal features, through data preprocessing and real-time recommendations. These recommendations could inform the drivers on where to position their taxi’s. We implemented a recommender system based on KMeans+Regression. Putting in the test with various features sets, we were able to achieve positive results. The effectiveness of our approach is demonstrated by the results, achieving approximately 98% accuracy in most cases.