Ongoing




  • TITLE:Explainable sEXism Identification in Social NeTworks
    ABSTRACT: In this project, we aim to develop an accurate text classification model utilizing a diverse range of machine learning methods, including classical techniques and advanced post-modeling approaches. Our goal is to tackle five distinct classification problems, examining the impact of each token within a sentence (row) on the final class for each year. This will be achieved through the weighted average of the last layer in the neural network or by utilizing machine learning explanation libraries.
    Additionally, we will engage a human observer to assess the model's accuracy in classification and explanation generation. To further refine our model, we will employ an autoencoder approach to learn the underlying patterns of selfish behavior and implement it on the final dataset. The resulting model will be analyzed using various performance evaluation metrics.
    Lastly, we will define a reward function to obtain the optimal class and explain a reinforcement learning model based on the evaluation of a pedestrian agent. The results will be compared to human analysis to ensure the effectiveness of our approach.
    HIGHLIGHTS: [Image 1] | [Image 2]


  • TITLE:Explainable NLP: A Systematic SURVEY
    ABSTRACT: Explainable natural language processing (XNLP) is a growing subfield of natural language processing that aims to make machine learning models more transparent and understandable to humans. In recent years, there has been a growing demand for XNLP due to the increasing need for transparency, interpretability, and accountability in NLP models. The goal of this survey article is to provide a comprehensive overview of XNLP and its applications, as well as to highlight the importance of humans in the loop and human evaluation. We conducted a systematic literature review to identify relevant articles and surveys published in the last decade.
    HIGHLIGHTS: [Image 1]


  • TITLE:Interpretable analysis of financial fraud data
    ABSTRACT: As continued in the previous project, In the first part, both oversampling and undersampling methods will be used in order to solve the challenge of the unbalanced data set. In the second part, an innovative combined unsupervised and supervised method will be used to improve the accuracy of the machine learning model. Then we will use a deep learning model to increase the complexity. Increasing complexity is not always desirable and this will be evaluated. Another serious challenge in the field of fraud detection that prevents the creation of efficient and reliable models in the real world is the issue of uncertainty. This means that fraudsters in the real world are always changing their methods, and therefore the model must be designed in such a way that it has an acceptable performance in the face of completely new observations and is different from the statistical distribution of the training data because the process of updating the deep learning model with the occurrence of new and different observations by experts is very time-consuming.
    In the first part, the oversampling method had better results. In the second part, the use of the innovative method led to the improvement of the accuracy of the model compared to five well-known models (decision tree, random forest, support vector machine, logistic regression, and k-nearest neighbor) in the literature. In the third part, the implementation of the dropout approach in the neural network increased the accuracy of this complex model.
    HIGHLIGHTS: [Image 1]


  • TITLE:Music Recommendation System with Human Participation and Cognitive Factors
    ABSTRACT: As continue of the previous project. In today's world of digital streaming services and vast music libraries, finding the perfect song or artist can be an overwhelming task. Existing recommendation systems often rely on algorithms and machine learning techniques to predict user preferences and make personalized suggestions. However, these systems can sometimes lack the human touch and consideration of cognitive factors that play a crucial role in shaping our music preferences and listening experiences.
    To address these limitations, we introduce a music recommendation system that incorporates human participation and cognitive factors into the recommendation process. This novel approach combines the power of autoencoder-based recommendation systems with the insight and personal touch of human expertise. The system takes into account factors such as mood, context, and personal associations, which are difficult for traditional algorithms to capture.
    By integrating these elements, our music recommendation system aims to provide a more holistic and emotionally resonant user experience. This project seeks to create a more engaging and immersive platform where users can discover new music that not only matches their preferences but also resonates with their unique cognitive and emotional states.
    HIGHLIGHTS: [Image 1]

Industry




  • TITLE:Video Advertising Optimization
    ABSTRACT: In this project, we try to focus on optimizing the issue of video advertising in the film and series section of Cafe Bazaar Company, using the provided data. In this project, different groups of users were tested with different approaches in terms of the number and location of ads in each video, and the resulting information, including people's reactions such as closing, rejecting, or viewing ads, was collected. Therefore, one of the main goals of this project is to choose the best plan to increase the final income, but in addition, user satisfaction and several other plans can be proposed, which we will discuss in the following.
    HIGHLIGHTS: [Image 1] | [Image 2]


  • TITLE:Identify the behavior and categorization of an online retailer customers by unsupervising data mining methods
    ABSTRACT: This project, which was done for one of the prominent retailers in Spain. They were categorized based on customers' purchase data as well as some personal characteristics of customers. Also, a recommended system was developed based on customers' purchasing history.
    HIGHLIGHTS: [Image 1] | [Image 2]


  • TITLE:Find Best place for Warehouse by GIS clustering
    ABSTRACT: In this project we will propose four approaches for the problem:
    • Locating new warehouses in other cities in Iran based on GIS clustering.
    • Proposing a new method based on best practices in warehousing operation in order to increase warehouse space.
    • Robotic systems.
    • Using “ Same-day-delivery” practice.
    HIGHLIGHTS: [Image 1]


  • TITLE:Intelligence in bicycle collection
    ABSTRACT: In this study, which was based on the 4th generation of bicycles, more than twenty factors affecting the collection of factors related to bicycle health, demand factors, and factors related to bicycle travel were analyzed and then by implementing a model based on  Demand forecast and forecast of future trips and bicycle health as well as based on geographical clustering based on the density of high priority collection, a comprehensive model was developed.
    HIGHLIGHTS: [Image 1]


  • TITLE:Find the best place for bicycle depot
    ABSTRACT: In this research, based on the number of completed trips and also the limitation of the logistics team for collecting bicycles, an intelligent model was developed that offers the best choice for the depot in each time period. In this model, logistic cost, number, and distance of bicycles, and its density are based on the analysis of geographical data.
    HIGHLIGHTS: [Image 1]


  • TITLE:Identify and analyze high-risk users for preventive measures
    ABSTRACT: Due to bike-sharing regulations as well as the rules of insurance companies, users are not allowed to use bicycles on highways, tunnels, and bridges, so identifying those who violate these rules is of particular importance. In this study, we first attempt to identify high-risk users using the spatial join method and identify the effective factors to detect this abuse. Then, by classifying different machine learning methods, it is possible to classify high-risk users according to travel and user characteristics so as to deal with repeat offenders. Finally, after trying several methods based on the accuracy of the model, the best method is selected. The results show, the best way to classify high-risk users is logistic regression and LDA, which gives us an accuracy of about 85%. Finally, we analyze the travel characteristics and personality of high-risk users and compare these results with those of ordinary users.
    HIGHLIGHTS: [Image 1] | [Image 2]


  • TITLE:Analysis of consumption and health of bicycle batteries
    ABSTRACT: In this study, by obtaining the factors affecting the discharge of the shared bicycle battery, a model was developed to predict the charge of each bicycle as well as to predict the abnormal discharge process and the resulting battery failure.
    HIGHLIGHTS: [Image 1]


  • TITLE:Bicycle breakdown analysis and modeling of preventive repairs
    ABSTRACT: In this research, the number of failure parts of the ridge is calculated based on the average distance, time and number of trips traveled and a preventive prediction model of failure is developed based on it.
    HIGHLIGHTS: [Image 1]


  • TITLE:Categorize users based on RFM index
    ABSTRACT: In this research, users of shared bicycles are classified based on three indicators:
    • The number of days since the last trip.
    • Average number of trips per user month.
    • The average cost paid for each trip.
    HIGHLIGHTS: [Image 1]

Academic




  • TITLE:A dynamic model for investigating the role of smoking on the incidence of lung cancer
    ABSTRACT: Nowadays, lung cancer is the deadliest and the most common type of cancer regardless of the gender factor according to global statistics. This has led to the reality that the death rate from lung cancer has risen sharply in recent years. Cigarette smoking is one of the most important causes of lung cancer and the association between cigarettes and lung cancer has been proven by large cohort studies. In this paper, the system dynamics approach is used as the main methodology. We have created a system dynamic model to analyse the relationship between smoking and lung cancer and to study factors such as the rate of smoking, death rate because of lung cancer, lung cancer treatment and costs, the impact of investment and budget allocation in different sectors by government, social welfare level of individuals, etc.
    Furthermore based on the developed model, some policies suggested to decrease the costs of this problem in society. Proposed policies are based on the impact of several different scenarios, such as increasing the contribution and budget allocation for prevention and advertising to counteract cigarette smoking as well as rising cigarette prices. Based on the model these policies can yield good results in reducing death rates and other costs of this problem. According to the results, in the long run, with a 44 percent increase in the price of cigarettes as well as a 10 percent increase in the smoking prevention budget, the number of smokers will be reduced by 8 percent.
    HIGHLIGHTS: [Image 1] | [Image 2]


  • TITLE:Discrete Simulation of Hospital Patient Service System
    ABSTRACT: Waiting time and length of stay have been reported as a factor in decreasing hospital satisfaction, especially in developing countries. This paper describes a model of outpatient hospital workflow in a developing country and optimizes patient waiting time as well as total length of stay. Using discrete event simulation, many alternative scenarios, including adding more work time, changing staffing and staffing responsibilities, are tested to better understand hospital settings. The results show that it is possible to achieve a 9.9% reduction throughout the patient's stay and can be done without adding more resources to the hospital.
    HIGHLIGHTS: [Image 1]


  • TITLE:Energy Efficiency Optimization by Queue Theory in Wireless Body Area Networks
    ABSTRACT: In this paper, we have tried to measure the effect of the queue length on the WBAN sensor. In a variety of articles, we deal with delays and queuing times. In this paper we consider a queuing model G / M / 1 and for The information required to enter this system and the queue are three priorities. Given the formulation presented, we will see that level 1 information has a higher priority than the other two, and will arrive sooner.
    HIGHLIGHTS: [Image 1]


  • TITLE:Optimization of system costs and waiting time in the queue for donation Kidney
    ABSTRACT: The use of queuing theory and supply chain systems has a particular place in optimizing the organ transplantation process and the member's inventory literature as a perishable commodity. Nevertheless, a small volume of articles in this area review the policies of supplying members to link to the theory of queue approach, despite the increasing importance of this topic in the health sector.
    In this paper, the principles of the quadratic theory have been used to optimize the control and vitamins of the donated and received members and to increase the productivity of the supply chain of perishable goods (living organs) for organ transplantation. People who donate members are logged in as queue system customers based on the Poisson process. Hospitality rates are assumed to give us a queue for the queue. The members who at this stage play the role of the queuing system inputs with the model and group service, enter the with the appropriate vehicle to make the organ transplant there. In, the queuing model is of the type. The resulting work will ultimately be pursued to minimize system costs by using a proper solution.
    HIGHLIGHTS: [Image 1] | [Image 2]


  • TITLE:Cost Optimization, Fuel Consumption and Pollution Generation in Urban Waste
    ABSTRACT: This study expands the multi-objective math model for urban waste management, which includes integrated three-level automation routing decisions, cost optimization, and maximization of produced biogas. Analyzing these decisions simultaneously may not only lead to an effective structure in the waste management network, it may also reduce the potential risk of waste management. In addition, due to the inherent complexity of the waste management system, uncertainty is inevitable, and the process must be considered as probable to ensure assurance in the decision-making process. In this research, the amount of pollution produced by waste collection machines as well as the amount of energy produced from the process of waste conversion to biogas is considered. Finally, considering the three factors of cost, pollution and production of biogas, we seek to find the best result in a reasonable time.
    HIGHLIGHTS: [Image 1]