Abstract In this study, a novel record linkage system for E-commerce products is presented. Our system aims to cluster the same products that are crawled from different E-commerce websites into the same cluster. The proposed system achieves a very high success rate by combining both semi-supervised and unsupervised approaches. Unlike the previously proposed systems in the literature, neither a training set nor structured corpora are necessary. The core of the system is based on Hierarchical Agglomerative Clustering (HAC); however, the HAC algorithm is modified to be dynamic such that it can efficiently cluster a stream of incoming new data. Since the proposed system does not depend on any prior data, it can cluster new products. The system uses bag-of-words representation of the product titles, employs a single distance metric, exploits multiple domain-based attributes and does not depend on the characteristics of the natural language used in the product records. To our knowledge, there is no commonly used tool or technique to measure the quality of a clustering task. Therefore in this study, we use ELKI (Environment for Developing KDD-Applications Supported by Index-Structures), an open-source data mining software, for performance measurement of the clustering methods; and show how to use ELKI for this purpose. To evaluate our system, we collect our own dataset and make it publicly available to researchers who study E-commerce product clustering. Our proposed system achieves 96.25% F-Measure according to our experimental analysis. The other state-of-the-art clustering systems obtain the best 89.12% F-Measure.

Abstract In this paper, we introduced two novel collaborative filtering techniques for recommendation systems in cases of various cold-start situations and incomplete datasets. The first model establishes an asymmetric weight matrix between items without using item meta-data and eradicates the disadvantages of neighborhood approaches by automatic determination of threshold values. Our first model, z-scoREC, is also regarded as a pure deep-learning model because it performs like a vanilla auto-encoder in transforming column vectors with z-score normalization similar to batch normalization. With the second model, ImposeSVD, we aimed to enhance the shortcomings of the PureSVD in cases of cold-start and incomplete data by preserving its straightforward implementation and non-parametric form. The ImposeSVD model relies on the z-scoREC, produces synthetic new predictions for the users by decomposing the latent factors from the imposed matrix. We evaluated our method on the well-known datasets and found out that our method was outperforming similar approaches in the specific scenarios including recommendations for cold-start users, strength in cold-start systems, and diversification of long-tail item recommendations in lists. Our z-scoREC model also outperformed familiar neighbor-based approaches when operated as a recommender system and gave a closer appearance to the decomposition methods despite its simple and rigid cost framework.

Cyberbullying refers to bullying and harassment of defenseless or vulnerable people such as children, teenagers, and women through any means of communication (e.g., e-mail, text messages, wall posts, tweets) over any online medium (e.g., social media, blogs, online games, virtual reality environments). The effect of cyberbullying may be severe and irreversible and it has become one of the major problems of cyber-societies in today’s electronic world. Prevention of cyberbullying activities as well as the development of timely response mechanisms require automated and accurate detection of cyberbullying acts. This study focuses on the problem of cyberbullying detection over Facebook activity content written in Turkish. Through extensive experiments with the various machine and deep learning algorithms, the best estimator for the task is chosen and then employed for both cross-domain evaluation and profiling of cyber-aggressive users. The results obtained with fivefold cross-validation are evaluated with an average-macro F1 score. These results show that BERT is the best estimator with an average macro F1 of 0.928, and employing it on various datasets collected from different OSN domains produces highly satisfying results. This article also reports detailed profiling of cyber-aggressive users by providing even more information than what is visible to the naked eye.

Abstract Sentiment analysis (SA) is an essential task for many domains where it is crucial to know users’ public opinion about events, products, brands, politicians and so on. Existing works on SA have concentrated on English texts including Twitter feeds and user reviews on hotels, movies and products. On the other hand, Facebook, as an online social network (OSN), has attracted quite limited attention from the research community. Among these, SA work on Turkish text obtained from OSNs are extremely scarce. In this paper, our aim is to perform SA on public Facebook data collected from Turkish user accounts. Our study differs from existing studies in terms of the data set scale, the natural language of the texts in the data set and the extent of experimental analyses that include both machine learning and deep learning techniques. We extensively report not only the results of different learning models involving SA but also statistical distribution of metadata of user activities across various user attributes (e.g. gender and age). Our experimental results indicate that recurrent neural networks achieve the best accuracy (i.e. 0.916) with word embeddings. To the best of our knowledge, this is the best result for SA on Facebook data in the context of the Turkish language.

Online Social Networks (OSNs) are very popular platforms for social interaction. Data posted publicly over OSNs pose various threats against the individual privacy of OSN users. Adversaries can try to predict private attribute values, such as gender, as well as links/connections. Quantifying an adversary’s capacity in inferring the gender of an OSN user is an important first step towards privacy protection. Numerous studies have been made on the problem of predicting the gender of an author/user, especially in the context of the English language. Conversely, studies in this field are quite limited for the Turkish language and specifically in the domain of OSNs. Previous studies for gender prediction of Turkish OSN users have mostly been performed by using the content of tweets and Facebook comments. In this article, we propose using various features, not just user comments, for the gender prediction problem over the Facebook OSN. Unlike existing studies, we exploited features extracted from profile, wall content, and network structure, as well as wall interactions of the user. Therefore, our study differs from the existing work in the broadness of the features considered, machine learning and deep learning methods applied, and the size of the OSN dataset used in the experimental evaluation. Our results indicate that basic profile information provides better results; moreover, using this information together with wall interactions improves prediction quality. We measured the best accuracy value as 0.982, which was obtained by combining profile data and wall interactions of Turkish OSN users. In the wall interactions model, we introduced 34 different features that provide better results than the existing content-based studies for Turkish.

Sentiment analysis identifies people's opinions, sentiments about a product, a service, an organization, or an event. Because of huge review documents, researchers explore different feature selection methods that aim to eliminate non valuable features. However, not much work has been done on feature selection methods for sentiment analysis of Turkish reviews. In this study, we propose a new feature selection method called Query Expansion Ranking that is based on query expansion term weighting methods, which are used in Information Retrieval domain to determine the most valuable terms for query expansion. We compare Query Expansion Ranking with Chi Square method, which is a well-known and successful feature selector, and Document Frequency Difference which is a feature selection method proposed for sentiment analysis of English reviews. Experiments are conducted on four Turkish product review datasets that are book, DVDs, electronics, and kitchen appliances reviews by using a supervised machine learning classification method, namely Naïve Bayes Multinomial classifier. We show that our new proposed method improves sentiment analysis performance in terms of classification accuracy and time. In the experimental evaluation, we also show that our new feature selector improves classification accuracy better than Chi Square, and Document Frequency Difference methods.

Abstract This paper deals with the problem of modeling Web information resources using expert knowledge and personalized user information for improved Web searching capabilities. We propose a “Web information space” model, which is composed of Web‐based information resources (HTML/XML [Hypertext Markup Language/Extensible Markup Language] documents on the Web), expert advice repositories (domain‐expert‐specified metadata for information resources), and personalized information about users (captured as user profiles that indicate users' preferences about experts as well as users' knowledge about topics). Expert advice, the heart of the Web information space model, is specified using topics and relationships among topics (called metalinks), along the lines of the recently proposed topic maps. Topics and metalinks constitute metadata that describe the contents of the underlying HTML/XML Web resources. The metadata specification process is semiautomated, and it exploits XML DTDs (Document Type Definition) to allow domain‐expert guided mapping of DTD elements to topics and metalinks. The expert advice is stored in an object‐relational database management system (DBMS). To demonstrate the practicality and usability of the proposed Web information space model, we created a prototype expert advice repository of more than one million topics/metalinks for DBLP (Database and Logic Programming) Bibliography data set. We also present a query interface that provides sophisticated querying facilities for DBLP Bibliography resources using the expert advice repository.

In this article, we discuss the issues involved in adding a native score management system to object-relational databases, to be used in querying Web metadata (that describes the semantic content of Web resources). The Web metadata model is based on topics (representing entities), relationships among topics (called metalinks ), and importance scores (sideway values) of topics and metalinks. We extend database relations with scoring functions and importance scores. We add to SQL score-management clauses with well-defined semantics, and propose the sideway-value algebra (SVA), to evaluate the extended SQL queries. SQL extensions and the SVA algebra are illustrated through two Web resources, namely, the DBLP Bibliography and the SIGMOD Anthology.SQL extensions include clauses for propagating input tuple importance scores to output tuples during query processing, clauses that specify query stopping conditions, threshold predicates (a type of approximate similarity predicates for text comparisons), and user-defined-function-based predicates. The propagated importance scores are then used to rank and return a small number of output tuples. The query stopping conditions are propagated to SVA operators during query processing. We show that our SQL extensions are well-defined, meaning that, given a database and a query Q, under any query processing scheme, the output tuples of Q and their importance scores stay the same.To process the SQL extensions, we discuss two sideway value algebra operators, namely, sideway value algebra join and topic closure, give their implementation algorithms, and report their experimental evaluations.