Mr Sajjad Mohammadian, Dr Nader Naghshineh, Dr Maryam Nakhoda,
Volume 8, Issue 2 (9-2021)
Abstract
Background and Aim: The meaning of cross-domain recommendation is that instead of dealing with each domain independently, transfer knowledge gained in one domain (source) to another domain (target) and use it. The present article systematically reviews the research in this field in terms of foundations, applications and challenges.
Method: The Prisma guidelines had been used. Search in Persian and English scientific information sources with related keywords were conducted and 98 English language sources were found in the period 2007 to 2021. Applying the initial refinement, inclusion and exclusion criteria by experts, 28 English documents were selected to enter in the systematic review.
Findings: There are four levels of cross-domain recommendations: Attributes, types, items and systems. Machine learning algorithms are used to predict user rating in cross-domain recommendations, and three categories of: Prediction, ranking, and classification criteria are used to evaluate predictions based on confusion matrix. Cross-domain recommendations can be used to increase the accuracy of recommendations, resolve cold start problems, cross-sell, and improve personalization by transferring knowledge between domains. The most challengeable recommendations of cross-domain is the differences between domains. These differences include the mismatch between the properties of the domains and/or unclear relationships between the domains. In addition, differences in domain size and poor performance of basic algorithms in predicting user rating are other challenges in cross-domain recommendations.
Conclusion: While this subject has been shaped in the last decade, but the keen attention of computer science and information researchers shows its importance. Items level are the main category of cross-domain recommendations. Due to the formation of e-business groups, in the future, cross-domain recommendations at the system level will be given more consideration. Cross-domain recommendations could be used to improve the performance of recommender systems, user modeling in human-computer interaction, and e-commerce.
Ms. Maryam Abolghasemi, Dr. Fatemeh Fahimnia,
Volume 8, Issue 4 (2-2022)
Abstract
Background and Aim: In processing large data, scientists have to perform the tedious task of analyzing hefty bulk of data. Machine learning techniques are a potential solution to this problem. In citizen science, human and artificial intelligence may be unified to facilitate this effort. Considering the ambiguities in machine performance and management of user-generated data, this paper aims to explain how machine learning can be combined with the active citizenship concept. In addition, it discusses the necessary conditions for advancing the citizen science and beyond.
Method: The review method and comprehensive systematic study was applied to assess the concept of machine learning, citizen science and human-computer interaction.
Results: Many research problems seem to be computationally insolvable and may demand human cognitive skills. Therefore, due to classification activities which are performed in the majority of large-scale citizenship science projects, in addition to participants who may learn lessons about the science, machines also learn lessons about human and imitate him and slowly its learning capacity enhances over time. Artificial intelligence, particularly machine learning is a debatable topic with related ambiguities and biases which should strongly take into consideration.
Conclusion: The application of machine learning techniques carries many advantages including classification time cut and masterful evaluations in the process of making decisions on big data sets. However, algorithms usually act as a black box where data biases are not observable at first glance. Taking this problem into consideration may mitigate serious risks arising from the application of such techniques.
