[ Link ]

Due to the extensive growth of information available online, recommender systems play a more significant role in serving people’s interests. Traditional recommender systems mostly use an accuracy-focused approach to produce recommendations. Today’s research suggests that this single-dimension approach can lead the system to be biased against a series of items with certain attributes. Biased recommendations across groups of items can endanger the interests of item providers along with causing user dissatisfaction with the system. This study aims to manage a new type of intersectional bias regarding the geographical origin and popularity of items in the output of state-of-the-art collaborative filtering recommender algorithms. We introduce an algorithm called MFAIR, a multi-facet post-processing bias mitigation algorithm to alleviate these biases. Extensive experiments on two real-world datasets of movies and books, enriched with the items’ continents of production, show that the proposed algorithm strikes a reasonable balance between accuracy and both types of the mentioned biases. According to the results, our proposed approach outperforms a well-known competitor with no or only a slight loss of efficiency.

[ Link ]

Financial statements are analytical reports published periodically by financial institutions explaining their performance from different perspectives. As these reports are the fundamental source for decision-making by many stakeholders, creditors, investors, and even auditors, some institutions may manipulate them to mislead people and commit fraud. Fraud detection in financial statements aims to discover anomalies caused by these distortions and discriminate fraud-prone reports from non-fraudulent ones. Although binary classification is one of the most popular data mining approaches in this area, it requires a standard labeled dataset, which is often unavailable in the real world due to the rarity of fraudulent samples. This paper proposes a novel approach based on the generative adversarial networks (GAN) and ensemble models that is able to not only resolve the lack of non-fraudulent samples but also handle the high-dimensionality of feature space. A new dataset is also constructed by collecting the annual financial statements of ten Iranian banks and then extracting three types of features suggested in this study. Experimental results on this dataset demonstrate that the proposed method performs well in generating synthetic fraud-prone samples. Moreover, it attains comparative performance with supervised models and better performance than unsupervised ones in accurately distinguishing fraud-prone samples.

[ Link ]

In this paper, we investigate the application of end-to-end and multi-module frameworks for G2P conversion for the Persian language. The results demonstrate that our proposed multi-module G2P system outperforms our end-to-end systems in terms of accuracy and speed. The system consists of a pronunciation dictionary as our look-up table, along with separate models to handle homographs, OOVs and ezafe in Persian created using GRU and Transformer architectures. The system is sequence-level rather than word-level, which allows it to effectively capture the unwritten relations between words (cross-word information) necessary for homograph disambiguation and ezafe recognition without the need for any pre-processing. After evaluation, our system achieved a 94.48% word-level accuracy, outperforming the previous G2P systems for Persian.

[ Link ]

Ranking items is a vital component in almost every application dealing with selecting the most suitable items among a pool of candidates. Yet, specific individuals or groups may be systematically disadvantaged in getting the opportunity of appearing on the ranking list. The fair ranking problem aims at mitigating the bias imposed on protected groups (i.e., disadvantaged groups) while preserving the total quality of the ranking list as high as possible. FA*IR is one of the existing algorithms, which finds the exact solutions for only one protected group, considering a given minimum number of protected items at every prefix of a ranking list. However, when an item belongs to more than two protected groups achieving optimal solutions gets more difficult. This paper proposes an algorithm called FARGO, a fair ranking algorithm based on the genetic algorithm (GA) enhanced by the simulated annealing (SA) that is able to handle any number of protected groups. A new objective function is also proposed by incorporating the main goals of the problem, which is utilized as FARGO’s fitness function. Furthermore, a novel evaluation metric named Expected Gain Ratio (EGR) is introduced to assess a fair ranking algorithm’s output. Experimental results on real-world datasets demonstrate that FARGO attains comparative performance with FA*IR for one protected group and finds near-optimal solutions for more than one protected group in terms of NDCG and EGR. Note that involving other concepts such as exposure is not a matter of this paper and can be an interesting subject for further studies.

[ Link ]

One of the most important aspects of any person’s life is personality, which affects one’s speech, decision, well-being, feeling and mental health. Personality detection is usually based on data collected by a questionnaire that comprises some critical problems such as the lack of direct access to the individuals and explicit personal information. However nowadays, one of the valuable resources for such studies is social networks. The footprint and tracking of users on social networks have provided valuable information for personality recognition. Specifically, this research introduces an intelligence personality recognition system based on modeling user behavior using sophisticated features, i.e., Statistical, Emotional, and Linguistic. Furthermore, a dataset called KNTU_Personality based on the MBTI personality model with the profile information and tweets has been collected. The experimental study follows two scenarios with complementing objectives. First, the sensitivity analysis is performed respecting to setting parameters, introduced features and different learning algorithms. Next, the proposed system has been compared with well-known personality detection systems. The results demonstrate the superiorities of the proposed system regarding its counterparts in terms of F-Score, Precision, Recall and Accuracy.

[ Link ]

Background and Objectives: Simplicity and flexibility constitute the two basic features for graph models which has made them functional models for real-life problems. The attributive graphs are too popular among researchers because of their efficiency and functionality. An attributive graph is a graph of the nodes and edges of which can be attributive. Nodes and edges as structural dimension and their attributes as contextual dimension made graphs more flexible in modeling real problems. Methods: In this study, a new clustering algorithm is proposed based on K-Medoid which focuses on graph’s structure dimension, through heat diffusion algorithm and contextual dimension through weighted Jaccard coefficient in a simultaneous matter. The calculated clusters through the proposed algorithm are of denser and nodes with more similar attributes. Results: DBLP and PBLOG real data sets are applied to evaluate and compare this algorithm with new and well-known cluster algorithms. Conclusion: Results indicate the outperformers of this algorithm in relation to its counterparts as to structure quality, cluster contextual and time complexity criteria.

[ Link ]

With the rapid growth of the Internet, search engines play vital roles in meeting the users’ information needs. However, formulating information needs to simple queries for canonical users is a problem yet. Therefore, query auto-completion, which is one of the most important characteristics of the search engines, is leveraged to provide a ranked list of queries matching the user’s entered prefix. Although query auto-completion utilises useful information provided by search engine logs, time-, semantic- and context-aware features are still important resources of extra knowledge. Specifically, in this study, a hybrid query auto-completion system called TIPS (Time-aware Personalised Semantic-based query auto-completion) is introduced to combine the well-known systems performing based on popularity and neural language model. Furthermore, this system is supplemented by time-aware features that blend both context and semantic information in a collaborative manner. Experimental studies on the standard AOL dataset are conducted to compare our proposed system with state-of-the-art methods, that is, FactorCell, ConcatCell and Unadapted. The results illustrate the significant superiorities of TIPS in terms of mean reciprocal rank (MRR), especially for short-length prefixes.

[ Link ]

Annually, web search engine providers spend a lot of money on re-ranking documents in search engine result pages (SERP). Click models provide advantageous information for re-ranking documents in SERPs through modeling interactions among users and search engines. Here, three modules are employed to predict users’ clicks on SERPs simultaneously, the first module tries to predict users’ click behaviors using Probabilistic Graphical Models, the second module is a Time-series Deep Neural Click Model which predicts users’ clicks on documents and finally, the third module is a similarity-based measure which creates a graph of document-query relations and uses SimRank Algorithm to predict the similarity. After running these three simultaneous processes, three click probability values are fed to an MLP classifier as inputs. The MLP classifier learns to decide on top of the three preceding modules, then it predicts a probability value which shows how probable a document is to be clicked by a user. The proposed system is evaluated on the Yandex dataset as a standard click log dataset. The results demonstrate the superiority of our model over the well-known click models in terms of perplexity.

[ Link ]

One of the most interesting topics in the scope of social network analysis is dynamic community detection, keeping track of communities’ evolutions in a dynamic network. This article introduces a new Louvain-based dynamic community detection algorithm relied on the derived knowledge of the previous steps of the network evolution. The algorithm builds a compressed graph, where its supernodes represent the detected communities of the previous step and its superedges show the edges among the supernodes. The algorithm not only constructs the compressed graph with low computational complexity but also detects the communities through the integration of the Louvain algorithm into the graph. The efficiency of the proposed algorithms is widely investigated in this article. By doing so, several evaluations have been performed over three standard real-world data sets, namely Enron Email, Cit-HepTh, and Facebook data sets. The obtained results indicate the superiority of the proposed algorithm with respect to the execution time as an efficiency metric. Likewise, the results show the modularity of the proposed algorithm as another effectiveness metric compared with the other well-known related algorithms.

[ Link ]

Social network analysis has become an important topic for researchers in sociology and computer science. Similarities among individuals form communities as the basic constitutions of social networks. Regarding the importance of communities, community detection is a fundamental step in the study of social networks typically modeled as large-scale graphs. Detecting communities in such large-scale graphs which generally suffers from the curse of dimensionality is the main objective followed in this study. An efficient modularity-based community detection algorithm called MDPCluster is introduced in order to detect communities in large-scale graphs in a timely manner. To address the high dimensionality problem, first, a Louvain-based algorithm is utilized by MDPCluster to distinguish initial communities as super-nodes and then a Modified Discrete Particle Swarm Optimization algorithm, called MDPSO is leveraged to detect communities through maximizing modularity measure. MDPSO discretizes Particle Swarm Optimization using the idea of transmission tendency and also escapes from premature convergence thereby a mutation operator inspired by Genetic Algorithm. To evaluate the proposed method, six standard datasets, i.e., American College Football, Books about US Politics, Amazon Product Co-purchasing, DBLP, GR-QC and HEP-TH have been employed. The first two are known as synthetic datasets whereas the rest are real-world datasets. In comparison to eight state-of-the-art algorithms, i.e., Stationary Genetic Algorithm, Generational Genetic Algorithm, Simulated Annealing-Stationary Genetic Algorithm, Simulated Annealing-Generational Genetic Algorithm, Grivan–Newman, Danon and Label Propagation Algorithm, the results indicate the superiorities of MDCluster in terms of modularity, Normalized Mutual Information and execution time as well.

[ Link ]

Nowadays, with the widespread use of the internet, users meet their information needs with the help of search engines. Users tend to retrieve the most relevant results by entering short phrases in the search engines. Customizing the retrieved results helps attain this goal. In this study, research works in the fields of query suggestion, particularly query auto-completion have been studied with special attention to customization. First, the sophisticated customizing features were classified into four dimensions: time, location, context, and demographic features. Then, related works were investigated regarding algorithm, dataset and evaluation measures. Regarding the literature, we found that the research works employing context or time as sophisticated features for customization are more than those using location or demographic features. While the location dimension has been recently taken into consideration, using other dimensions has a long background. Moreover, in the related works, the AOL dataset and Mean Reciprocal Rank (MRR) are known as the most frequent dataset and evaluation measure, respectively.

[ Link ]

Given the financial crises in the world, one of the most important issues of banking industry is the assessment of customers’ credit to distinguish bad credit customers from good credit customers. The problem of customer credit risk assessment is a binary classification problem, which suffers from the lack of data and sophisticated features as main challenges. In this paper, an adaptive neuro-fuzzy inference system is exploited to tackle the customer credit risk assessment problem regarding the mentioned challenges. First of all, a SOMTE-based algorithm is introduced to overcome the data imbalancing problem. Then, several efficient features are identified using a MEMETIC metaheuristic algorithm, and finally an adaptive neuro-fuzzy system is exploited for distinguishing bad credit customers from good ones. To evaluate and compare the performance of the proposed system, the standard German credit data dataset and the well-known classification algorithms are utilized. The results indicate the superiority of the proposed system compared to some well-known algorithms in terms of precision, accuracy, and Type II errors.

[ Link ]

Being aware of people’s attitudes and emotions about a specific person or an event can have a high impact on the decisions of individuals and organizations. With the rise of social networks, specifically Instagram, many people are sharing their attitudes on this social network. Analyzing the emotions of users of this social network can help managers make organizational decisions and predict essential events such as elections. In this research, the EAS system designed and implemented to extract emotions and visualize them. As a practical example, the Instagram users’ feelings about the two main candidates for the 12th Iranian presidential election also examined. The data were Instagram Persian comments collected using a developed crawler. The result shows a more positive feeling about Rouhani in comparison with Raeisi. Also, the lexicon-based analysis of Rouhani revealed a high level of trust emotion, along with anger and disgust. The crawled and preprocessed dataset is publicly available at https://github.com/sfdk74/EAS.

[ Link ]

Machine translation is an important field of research and development. Word reordering is one of the main problems in machine translation. It is an important factor of quality and efficiency of machine translations and becomes more difficult when it deals with structurally divergent language pairs. To overcome this problem, we introduce a neural reordering model, using phrasal dependency trees which depict dependency relations among contiguous non-syntactic phrases. The model makes the use of reordering rules, which are automatically learned by a probabilistic neural network classifier from a reordered phrasal dependency tree bank. The proposed model combines the power of the lexical reordering and syntactic pre-ordering models by performing long-distance reorderings. The proposed reordering model is integrated into a standard phrase-based statistical machine translation system to translate input sentences. Our method is evaluated on syntactically divergent language-pairs, English → Persian and English → German using WMT07 benchmark. The results illustrate the superiority of the proposed method in terms of BLEU, TER and LRscore on both translation tasks. On average the proposed method retrieves a significant impact on precision and recall values respect to the hierarchical, lexicalized and distortion reordering models.

[ Link ]

Articulateness and plasticity are two essential attributes that make a graph as an efficient model to real life problems. Nowadays, the attributed graph is received lots of attentions because of usability and effectiveness. In this study, a novel k-Medoid based clustering algorithm, which focuses simultaneously on both structural and contextual aspects using Signal and the weighted Jaccard similarities, are introduced. Two real life data-sets, Political Blogs and DBLP bibliography, are employed in order to evaluate and compare the proposed algorithm with state-of-the-art clustering algorithms.

[ Link ]

Intelligent machine translation (MT) is becoming an important field of research and development as the need for translations grows. Currently, the word reordering problem is one of the most important issues of MT systems. To tackle this problem, we present a source-side reordering method using phrasal dependency trees, which depict dependency relations between contiguous non-syntactic phrases. Reordering elements are automatically learned from a reordered phrasal dependency tree bank and are utilized to produce a source reordering lattice. The lattice finally is decoded by a monotone phrase-based SMT to translate a source sentence. The approach is evaluated on syntactically divergent language pairs, ie English to Persian and English to German, using the workshop of machine translation 2007 (WMT07) benchmark.

[ Link ]

Nowadays, text summarization is one of the most important active research fields in information retrieval. The most of the supervised extractive summarization systems utilize learning-to-rank methods to score sentences according to their importance. They need a high-quality comprehensive summarization corpus, which is labeled manually by human experts. Unfortunately, this sort of corpus is not available for most low-resource languages such as Persian. In this study, first of all, a comprehensive human-labeled summarization corpus (called Bistoon) collected by the crowdsourcing approach is introduced, and then a Persian summarizer based on a novel semi-supervised summarization approach, which is a combination of co-training and self-training, is presented to overcome the absence of sufficient data.

[ Link ]

Recently, social networks have provided an important platform to detect trends of real-world events. The trends of real-world events are detected by analysing flow of massive bulks of data in continuous time steps over various social media platforms. Today, many researchers have been interested in detecting social network trends, in order to analyse the gathered information for enabling users and organisations to satisfy their information need. This article is aimed at complete surveying the recent text-based trend detection approaches, which have been studied from three perspectives (algorithms, dimension and diversity of events). The advantages and disadvantages of the considered approaches have also been paraphrased separately to illustrate a comprehensive view of the previous works and open problems.

[ Link ]