Data, information and knowledge management in the field of smart business
Fatemeh Rezaimehr; Chitra Dadkhah
Abstract
AbstractRecently, the Internet has played a significant and substantial role in people's lives. However, the content available in the global web environment should align with users' daily needs, providing them with useful and up-to-date information tailored to their tastes. In this context, recommender ...
Read More
AbstractRecently, the Internet has played a significant and substantial role in people's lives. However, the content available in the global web environment should align with users' daily needs, providing them with useful and up-to-date information tailored to their tastes. In this context, recommender systems assist users by suggesting items that closely match their preferences in less time. Today, with the exponential growth of data, the utilization of recommender systems has surged. Conversely, these systems encounter challenges such as evolving user preferences over time, cold start problem, sparsity within the user-item matrix, the infiltration of fake users in the systems, and their adverse impact on the recommendation lists. The objective of this paper is to propose a recommender system grounded in time and trust factors to enhance the efficiency and precision of system recommendations. Initially, the proposed system addresses the data sparsity dilemma by incorporating reliable implicit ratings into the user-item matrix. Subsequently, it constructs a weighted user-user network based on user rating timestamps and trust relationships among users, thereby mitigating the cold start problem and accounting for changing user preferences over time. The proposed recommender system employs a novel community detection algorithm introduced in this paper to identify the nearest neighbors of active users and recommends the top @k items based on the collaborative filtering approach. Evaluation results of the proposed system, tested on a film recommender system using the Epinions dataset, demonstrate its superior efficiency compared to basic systems.IntroductionToday, with the increasing tendency of users to use websites for obtaining information, online shopping, and using social networks for expressing personal opinions, the ways of obtaining information and establishing connections among users have undergone significant changes. Consequently, users are confronted with the big of data. Managing this data and selecting the appropriate options from this vast collection and presenting it to users is one of the main reasons for the development of information retrieval systems and search engines. In this regard, Recommendation Systems (RSs) help users choose the best options and recommend items that are closer to their preferences in the shortest possible time. Different models of RS such as collaborative filtering, content-based, knowledge-based, and newly developed context-aware RS, have been presented by researchers (Casillo et al., 2022). Each has its own advantages and disadvantages, which can be combined to create a hybrid RS. It should be noted that RS face challenges, including changes in user preferences over time, cold start for new users or items, sparsity of the user-item matrix, attack by fake users, and their negative impact on the recommendation list. In this paper, a time- and trust-based recommendation system is presented to enhance the performance and accuracy of recommendations. Our proposed system initially solves the data sparsity problem by adding reliable implicit ratings to the user-item rating matrix. It then generates a weighted user-user network based on the time of user feedback on items and trust relationships among users. This approach addresses the cold start problem and the change in user preferences over time. Our system is based on a novel community detection algorithm presented in this article, which identifies the nearest neighboring users with similar tastes to the active user and recommends the top-k items using the collaborative filtering method. The evaluation of the proposed system is performed on an Epinions dataset for a movie recommendation system. The evaluation uses metrics such as accuracy, recall, F1 score, mean absolute error, and root mean square error. The experimental results indicate the superior performance of the proposed system compared to similar systems.Literature ReviewIn the recent years, the researchers attempt to improve the accuracy of their recommendation for retaining the users and increasing the profit. Some of the papers has worked on optimizing the performance of their proposed RS using evolutionary algorithms (Tohidi & Dadkhah, 2020) and the others used the additional information such as time, location, etc. Trust-based RSs have been recently introduced to the community of computer science. Recent studies have shown that incorporating social factors or trust statements in RSs leads to the improvement of recommendation quality (P. Moradi & Ahmadian, 2015; S. Ahmadian, M. Meghdadi, & Afsharchi, 2018b). So far, several trust-based CF approaches have been proposed to overcome data sparsity and cold-start problems as well as to increase recommendable items (Ghavipour & Meybodi, 2016; Moradi, Ahmadian, & Akhlaghian, 2015; P. Massa & Avesani, 2007; Ranjbar Kermany & Alizadeh, 2017). Trust statements can be explicitly collected from users or can be implicitly inferred from users behaviors (S. Ahmadian, M. Meghdadi, & Afsharchi, 2018a; S. Ahmadian, P. Moradi, & Akhlaghian, 2014). Liu and Lee proposed a specific approach which does not directly use the trust information; instead they take into account the number of exchanged messages among the users of the system to construct the trust network (Liu & Lee, 2010). Alahmadi and Zeng presented a framework to apply short texts posted by users friends in microblogs as an additional data source to build the trust network (Alahmadi & Zeng, 2015). Since explicit trust statements are directly specified by the users, they are more accurate and reliable than implicit ones in determining social relationships among users (Cho, Kwon, & Park, 2009; Ingoo, Kyong, & Tae, 2003; Lathia, Hailes, & Capra, 2008; Manolopoulus, Nanopoulus, Papadopoulus, & Symeonidis, 2008).The research In (Abdul-Rahman & Hailes, 2000) has been shown that a user constructs his/her social connections with someone who has similar tastes. Massa and Avesani showed that adding social network data to traditional collaborative filtering improves the recommendation results (P. Massa & Avesani, 2007). Gharibshah and Jalili studied the relation between RSs and connectedness of users-items bipartite interaction network (Gharibshah & Jalili, 2014). Guo et al. proposed a method which merged the ratings of users trusted neighbors with the other information sources to identify their preferences (G. Guo, J. Zhang, & Thalmann, 2014). Yang et al. proposed a Bayesian inference based recommendation method for online social networks (X. Yang, Y. Guo, & Liu, 2013). In this method, the similarity value between each pair of users is measured using a set of conditional probabilities derived from their mutual ratings. Jiang et al. introduced a framework to incorporate interpersonal influences of users in social network with their individual preferences to improve the accuracy of social recommendation (Jiang, Cui, Wang, Zhu, & Yang, 2014).Purchase/rating time is one of the most important contextual information that can be used to design RSs with high precision (Xiong, Chen, Huang, Schneider, & Carbonell, 2010). The main motivation for time-aware RS is that in realistic scenarios users tastes might change over time.MethodologyWe propose a time and trust-aware RS using a graph-based community detection method consists of four steps: 1: developing a user-item rating matrix, 2: constructing a time weighted user-user network, 3: performing graph- based community detection, 4: recommending Top-N items. In the first step, the user-item rating matrix is developed by adding some implicit ratings and the quality of the implicit ratings is evaluated using a reliability measurement. In the second step, a time-weighted user-user network is constructed based on the combination of trust relationships and similarity between users. Moreover, the timestamps of user-item ratings are considered to calculate the similarity between users. In the third step, a graph-based community detection method classifies similar users into appropriate communities. Finally, in the fourth step, it predicts the rating for each unobserved item and top-N recommendations is generated for the target user.We proposed a new community detection method that consists of three phases. First, the initial centers of communities are obtained using a sparsest subgraph of weighted user-user network. It should be noted that the initial centers must have the maximum dissimilarities with each other based on the general concept of clustering and community detection algorithms. Then users can be assigned to their nearest communities. For each user proposed system calculated the fitness function. User has associated to community which has high value of fitness function. Then the centers of communities were updated in order to maximize a fitness function. This process is iteratively repeated until members of communities do not change and steady state is achieved. A set of communities are identified where the users are assigned to their corresponding communities. Some of the communities may have overlap and they can be merged. The final communities were used as the nearest neighbors set of the active user in the same community for the recommendation.ConclusionOur proposed algorithm solves the sparsity of rating matrix by adding the implicit rating and solved cold-start problem for new users by considering the trust between the users. We applied the proposed algorithm on extended Epinions dataset and compared its performance with similar algorithms. The experimental results showed that our proposed algorithm outperforms the other algorithms according to the accuracy and recommends the top@N items with high precision.
Data science, intelligence and future analysis
Monireh Hosseini; Elnaz Galavi
Abstract
Community detection is an important topic for social network analysis and is also essential to understanding complex networks structure. In community detection, the goal is to determine the groups in which the group nodes are densely connected to each other. In this research, deep learning techniques ...
Read More
Community detection is an important topic for social network analysis and is also essential to understanding complex networks structure. In community detection, the goal is to determine the groups in which the group nodes are densely connected to each other. In this research, deep learning techniques have been used to control graph data with high dimensions, while presenting a comprehensive and integrated architecture of community recognition methods with deep learning. Community detection classic approaches are suitable for networks with low dimensions. Therefore, the reduction of complex network dimensions is counted as a significant topic in community detection. In this paper, in order to reveal the direct and indirect connections among nodes, first a new similarity matrix of network topology is built. Then, a stacked auto-encoder is designed to decrease dimensions based on unsupervised learning. In order to detect communities, various clustering algorithms are then tested and utilized. Evaluation of the proposed research model is performed by surveying various experiments on standard criteria and six real data sets of Karate, Dolphins, Football, Polbooks, Cora and Citeseer. The proposed method evaluation outcomes show a higher accuracy in the identification of communities in the football data set compared to the twelve proposed algorithms used in past researches, and show a significant improvement in other data sets compared to the thirteen algorithms.
Introduction
Today, due to the increasing use of the Internet, social networks have found an important role in the real life of people. In social networks, some nodes are more connected than the entire network nodes, which are called communities(Sperli, 2019). Community Detection is an important topic for social network analysis and is also essential to understanding complex network structure In community detection, the goal is to determine the groups in which the group nodes are densely connected.
There are many methods for community detection, but deep learning has shown excellent performance in a wide range of research fields, such as social networks, graph embedding, etc.
In this research, deep learning techniques have been used to control graph data with high dimensions, while presenting a comprehensive and integrated architecture of community detection methods with deep learning.
Research Questions
Is it possible to create a new similarity matrix from the graph of complex networks that fully reveals the similarity relationships between network nodes?
What is the appropriate method of deep learning to represent the features of complex networks in low dimensions?
Is it possible to provide a suitable framework with model flexibility for networks of different sizes for community detection using the deep learning method?
Can more accurate clustering results be achieved for community detection?
Literature Review
2.1.Community detection classic approaches are suitable for networks with low dimensions. Therefore, the reduction of complex network dimensions is counted as a significant topic in community detection. The disadvantage of the high-dimensional network is the huge computational costs incurred by community detection methods. Therefore, a method is needed to transform high-dimensional graphs into a lower-dimensional space, where important information about network structure and node properties is still preserved. According to past research, autoencoders are the dominant method for mapping data points in lower-dimensional spaces (Souravlas et al, 2021).
2.2.To display the network, using the proximity matrix as the network similarity matrix can describe the similarity relationship between the nodes in the network. But the relationship between nodes in a social network is complex. On the other hand, in addition to the similarity between nodes that are directly connected, there are different degrees of similarity between nodes that are not directly connected (Su et al., 2020).
2.3. Wu et al. (2020) and Geng et al. (2020) reconstructed the adjacency matrix to represent the network. Dhilber and Bhavani (2020) used a cubic matrix for the input of the stack autoencoders, as did the work of Yang et al. (2016). Xie et al. (2018) first proposed a new representation of network similarity and then fed it with a sparse filtering model to extract meaningful features of network nodes. But in addition to the problem of lack of neighbor information in the proximity matrix based on Su et al.'s (2020) research, using only one function to check the similarity between nodes cannot fully reveal the topological information of the network. Therefore, a similarity matrix should be presented that can solve the proposed gaps.
Methodology
In this paper, to reveal the direct and indirect connections among nodes, first, a new similarity matrix of network topology is built. To construct the new similarity matrix, two matrices are used, i.e. proximity matrix and S∅rensen–Dice's (S∅) similarity matrix in Xie et al. (2018) 's research. In the next step to extract low-dimensional graph features, the new similarity matrix is given as input to the stack autoencoder networks, which have several hidden layers for unsupervised training. Then, using the newly learned features that are in the low-dimensional matrix with the help of K-means, DBSCAN, and SNNDPC clustering algorithms, communities are detected.
Conclusion
Evaluation of the proposed research model is performed by surveying various experiments on standard criteria and six real data sets of Karate, Dolphins, Football, Polbooks, Cora, and Citeseer. The proposed method evaluation outcomes show a higher accuracy in the detection of communities in the football data set compared to the twelve proposed algorithms used in past research and show a significant improvement in other data sets compared to the thirteen algorithms. In addition to these cases, the superiority of the similarity matrix used in this research was proved as a key prerequisite for community detection.
Keywords: Community Detection, Deep Learning, Autoencoder, Complex Networks.
Zahra Shirani; Amir Jalaly Bidgoly
Abstract
In recent years, the number of users of social networks has grown significantly. The big challenge for these networks’ audience is How to communicate with the people present on these networks. Friend recommender systems try to fix this challenge by offering suggestions. In this study, data from ...
Read More
In recent years, the number of users of social networks has grown significantly. The big challenge for these networks’ audience is How to communicate with the people present on these networks. Friend recommender systems try to fix this challenge by offering suggestions. In this study, data from the social and scientific network of Kousarent were used. In this research, using 10 types of relationships between users without considering the friendship relationships, network graph created, and then by using 3 algorithms Louvain, Kmeans and Hierarchical graph clustering was performed to identify communities. Clusters obtained from Louvain's clustering algorithm had higher percentages of matching with friendships. Then, weights were calculated by genetic algorithm for each of 10 relationships and by applying Louvain clustering algorithm on the network graph, the highest percentage of matching with the optimal weight of each of the 10 relationships was obtained. In this case, the resulting clusters are optimal clusters containing the most similar users. So other users in the same cluster can be suggested as friends. The weight of the edges between the individuals in the graph was also used to prioritize the bids. At the end, the friend proposed method was evaluated and the percentage of suggested friends matched with the individual's true friends was calculated.
Amir Hossein Hosseinian; Babak Teimourpour; Bagher Jamali Hondori
Abstract
Detecting existing communities in social networks is a significant process in analyzing these networks. In recent years, the community detection problem has become popular for detecting structures of social networks. Due to high importance of this problem, various algorithms have been developed ...
Read More
Detecting existing communities in social networks is a significant process in analyzing these networks. In recent years, the community detection problem has become popular for detecting structures of social networks. Due to high importance of this problem, various algorithms have been developed in the literature to find communities of complex networks. In this research, a hybrid meta-heuristic consisting of the genetic algorithm (GA) and the invasive weed optimization (IWO) method have been proposed which aims to find appropriate and high quality solutions for the community detection problem. In this hybrid method, the initial solutions are generated via the IWO algorithm, and thereafter the optimization process is continued by means of the genetic algorithm. The proposed algorithm is known as the GAIWO. Fitness of solutions is determined in terms of the modularity density criterion. Modularity density has a maximization essence and determines the quality of detected communities. To evaluate the efficiency of the GAIWO, four other methods have been employed and their results have been compared. Comparisons have been made on several networks with different sizes. Input parameters of all algorithms have been tuned by a design of experiments approach. The outputs indicate appropriate efficiency of the proposed algorithm. Validation of the results have been investigated by means of the Normalized Mutual Information (NMI) metric.
