Recommender Systems Research Research Articles

Fusing Knowledge and Aspect Sentiment for Explainable Recommendation

The current research of recommender system mainly focuses on improving the accuracy of the recommendation, but pays less attention to explainability. Explainability is essential to enhance users’ trust and satisfaction, which can even increase the likelihood of buying items. In the existing explainable recommender systems, the mining of explicit and implicit features is not comprehensive, and the interaction among these features is not considered plentifully. In addition, the generated recommended reason text is not personalized and rich enough. It is necessary to improve the quality of recommended reason text because it is difficult to meet the needs of different users by using low-quality text. In this paper, we propose a new method that fuses external knowledge and aspect sentiment to predict rating and generate personalized, content-rich recommended reasons, which applies fine-tuning BERT to solve aspect-based sentiment analysis and extends Transformer to generate recommended reason by using bi-directional attention. The experiment results on real-world datasets demonstrated that our method was effective, and our model was superior to the baseline models on various metrics. For rating prediction task, our model can achieve an improvement of 0.6% on average in terms of RMSE. For recommended reason generation task, our model can achieve an improvement of 9.2% to 11.3% over state-of-the-arts in terms of BLEU.

Machine learning-based book recommender system: a survey and new perspectives

The exponential growth of recommender systems research has drawn the attention of the scientific community recently. These systems are very useful in reducing information overload and providing users with the items of their need. The major areas where recommender systems have contributed significantly include e-commerce, online auction, and books and conference recommendation for academia and industrialists. Book recommender systems suggest books of interest to users according to their preferences and requirements. In this article, we have surveyed machine learning techniques which have been used in book recommender systems. Moreover, evaluation metrics applied to evaluate recommendation techniques is also studied. Six categories for book recommendation techniques have been identified and discussed which would enable the scientific community to lay a foundation of research in the concerned field. We have also proposed future perspectives to improve recommender system. We hope that researchers exploring recommendation technology in general and book recommendation in particular will be finding this work highly beneficial.

Machine learning-based book recommender system: a survey and new perspectives

The exponential growth of recommender systems research has drawn the attention of the scientific community recently. These systems are very useful in reducing information overload and providing users with the items of their need. The major areas where recommender systems have contributed significantly include e-commerce, online auction, and books and conference recommendation for academia and industrialists. Book recommender systems suggest books of interest to users according to their preferences and requirements. In this article, we have surveyed machine learning techniques which have been used in book recommender systems. Moreover, evaluation metrics applied to evaluate recommendation techniques is also studied. Six categories for book recommendation techniques have been identified and discussed which would enable the scientific community to lay a foundation of research in the concerned field. We have also proposed future perspectives to improve recommender system. We hope that researchers exploring recommendation technology in general and book recommendation in particular will be finding this work highly beneficial.

An Enhanced Multi-Modal Recommendation Based on Alternate Training With Knowledge Graph Representation

Deep network recommendation is a cutting-edge topic in current recommendation system research, which as a combination of recommendation systems and deep learning theory can effectively improve recommendation accuracy. In a real recommendation scenario, all the effective information in a data set should be extracted, both explicit and implicit, because the comprehensive degree of information is proportional to the recommendation performance. This article proposes an enhanced multi-modal recommendation based on alternate training with knowledge graph representation (SI-MKR) based on the MKR deep learning recommendation model. Our framework is an enhanced recommendation system based on knowledge graph representation, using valuable external knowledge as multi-modal information. The SI-MKR model solves the problem of ignoring the diversity of data types in the multi-modal knowledge-based recommendation system, which adds user and item attribute information from a knowledge graph as an enhancement recommendation multi-tasking training. By analysing the content of the item and user attributes, the SI-MKR model classifies the attributes of the items and users, processes the text type attributes and multi-value type attributes separately for feature extraction, and other types of attributes are used as inputs to the knowledge graph embedding unit. In addition, the knowledge graph data form a triplet unit, thus continuing the knowledge graph data training process. The feature extraction unit of the knowledge graph and the recommended unit are connected through the cross-compression unit for alternate training. During the deep learning framework training process, the recommendation system’s item has a potential correlation with the head entity in the knowledge graph which embodies the idea of multi-tasking. Through extensive experiments on real-world datasets, we demonstrate that SI-MKR achieves substantial gains in movie recommendation over advanced model baselines. Even user-item interactions are sparse, SI-MKR maintains better performance than the MKR model.

Rule-Based Effective Collaborative Recommendation Using Unfavorable Preference

The biggest challenge in collaborative filtering recommendation system research is data sparsity ; it mainly occurs as user rates very few items from widely available options. Data Imputation techniques address the data sparsity problem by filling the missing values and then predicting the likeliness of the user. Most of the existing imputation systems assign high ratings to the items or incorporate additional information to enhance the performance of collaborative filtering recommendations. This paper proposes an association rule-based imputation method (RUBLE) to improve the top- $N$ prediction performance of the collaborative filtering recommendation. The proposed method identifies the unfavorable items of each user using the association rule mining technique and imputes them with low values. The proposed method not only addresses the sparsity problem but also provides a better quality of recommendation by eliminating the unfavorable items in top- $N$ predictions. Existing collaborative methods can quickly adapt to the proposed method as it is method agnostic. The experimental results show that the proposed method enhances the accuracy of the traditional recommender methods by two times on average and significantly outperforms existing imputation based approaches.

Integrating selection-based aspect sentiment and preference knowledge for social recommender systems

PurposeRecommender system approaches such as collaborative and content-based filtering rely on user ratings and product descriptions to recommend products. More recently, recommender system research has focussed on exploiting knowledge from user-generated content such as product reviews to enhance recommendation performance. The purpose of this paper is to show that the performance of a recommender system can be enhanced by integrating explicit knowledge extracted from product reviews with implicit knowledge extracted from analysis of consumer’s purchase behaviour.Design/methodology/approachThe authors introduce a sentiment and preference-guided strategy for product recommendation by integrating not only explicit, user-generated and sentiment-rich content but also implicit knowledge gleaned from users’ product purchase preferences. Integration of both of these knowledge sources helps to model sentiment over a set of product aspects. The authors show how established dimensionality reduction and feature weighting approaches from text classification can be adopted to weight and select an optimal subset of aspects for recommendation tasks. The authors compare the proposed approach against several baseline methods as well as the state-of-the-art better method, which recommends products that are superior to a query product.FindingsEvaluation results from seven different product categories show that aspect weighting and selection significantly improves state-of-the-art recommendation approaches.Research limitations/implicationsThe proposed approach recommends products by analysing user sentiment on product aspects. Therefore, the proposed approach can be used to develop recommender systems that can explain to users why a product is recommended. This is achieved by presenting an analysis of sentiment distribution over individual aspects that describe a given product.Originality/valueThis paper describes a novel approach to integrate consumer purchase behaviour analysis and aspect-level sentiment analysis to enhance recommendation. In particular, the authors introduce the idea of aspect weighting and selection to help users identify better products. Furthermore, the authors demonstrate the practical benefits of this approach on a variety of product categories and compare the approach with the current state-of-the-art approaches.

Tensorized Neural Network for Multi-Aspect Rating-Based Recommendation

Generating personalized recommendations is one of the most crucial aspects in Recommender System research area. Most of the researches only focus on the accuracy of recommendation using collaborative filtering that relies on a single overall rating that represents the overall preferences. However, the user may have a different emphasis on different specific aspects that affect the users’ final rating decisions. Therefore, we present a neural network model that utilize multi-aspects ratings using Tensor Factorization to improve the accuracy of personalization, as well as optimizing the dynamic weights of the aspect. To measure the estimated weights for the aspects, we employ the Higher Order Singular Value Decomposition (HOSVD) technique called CANDECOMP/PARAFAC (CP) decomposition that allows for multi-faceted data processing. We then develop the Neural Network with backpropagation error to train the model with different parameter settings and limited computational time. We also use a non-linear activation function in each hidden layer in various settings. The experimental result measured using MAE shows that the proposed model has significantly outperformed the baseline approach in terms of the prediction accuracy. Based on the observation, the performance of rating prediction has been improved by employing the Tensorized Neural Network model and can overcome the problem of local optimum convergence for multi-aspect rating recommendation.

Various Recommender Systems and Their usage in the Domains of Mobile Based and Online Applications

The activities of the users are surrounded by online shopping, online content fetching and online payment of the various bills. The important point is in case of the online shopping with Amazon, Flipkart and many other online shopping sites provides some sort of intelligent assistance to the user. Based on the past history, based on the user profile. Such kind of the applications were named as recommender systems. The most common categories of recommender systems involves, the collaborative filtering, content-based filtering, multi-criteria recommender systems, risk-aware recommender systems, mobile recommender systems and Hybrid recommender systems. The current work dealing with the process followed by the recommender systems along with various key factors embedded in the usage. The suggestions given by the application to the user depends on user profile, and content searched by the user and the collaboration of other products with the current product. The work focus on the implementation algorithms existing in the process of recommender systems, the above listed categories of recommender systems follow certain key mechanisms depending on the user query. The work also deals with the performance aspects of the recommender systems in case of accuracy and reproducibility in recommender system research. Especially the mobile recommender systems there are certain limitations of region and accuracy of the results. Overall the outcome of the work is to describe the importance of the recommender systems and the internal mechanism followed by various recommender systems.

A Scientometric Analysis on Mobile Recommender Systems

With further advancements in technology, a new emerging topic called 'mobile recommender systems' has gained tremendous momentum. The amount of research on mobile recommender systems can reflect the development of a variety of research topics such as mobile computing and context-aware systems. Hence, performing a scientometric mapping on this topic is needed. However, there is a lack of research in this area that aims to perform a detailed scientometric mapping. The objective of this study is to perform a scientometric mapping related to mobile recommender systems. We analysed publications between 2002 and 2017 indexed in Web of Science. The analysis maps the parameters of total output, the growth of output, authorship, major contributors, thematic trends and emerging topics in the field. Our study serves as a resource for future research by shedding light on how trends in mobile recommender systems research have evolved as well as providing insights into the current state of this field.

개인화 서비스 진전에 따른 자동추천 시스템 연구 동향과 방법론적 특성 연구

본 연구의 목적은 사회과학 연구자들을 위한 추천 시스템 연구의 연구 경향의 변화와 방법론적 동향을 탐색하는 것이다. 각각의 연구 경향에서 사용된 연구방법상의 문제점과 한계를 검토하고 추천 시스템 연구 범위의 확장과 더불어 새롭게 사용되는 연구방법들을 제시한다.<BR>전통적인 추천 시스템 연구 경향은 크게 시스템 중심 접근법과 이용자 중심 접근법으로 나눌 수 있다. 그런데 시스템 중심 접근법의 주요 관심사여왔던 추천서비스 알고리즘의 예측 정확도 향상이 반드시 실제 이용자의 사용 만족도로 이어지지 않는다는 비판이 지속적으로 제기되었다. 이러한 이유에서 등장한 이용자 중심 접근법은 주로 전통적인 사회과학 실험 방법(lab experiments)을 사용하는데, 이러한 연구방법 역시 제한된 수의 피실험자들에게 제한된 변수만을 테스트하는 방식으로 도출된 결과의 낮은 신뢰도와 닞은 외적 타당성이 단점으로 지적되어왔다. 이 글에서는 현재 주요 글로벌 미디어 기업들(구글, 아마존, 페이스북 등)이 적극적으로 활용하고 있는 대규모 온라인 통제실험을 이러한 이용자 중심 접근법의 실험 방법론이 가지고 있는 약점을 극복할 수 있는 대안적 방법론으로 제시한다.<BR>한편 추천서비스가 점점 우리 일상의 한 부분이 되어가면서, 기존의 추천 시스템 평가 위주의 연구영역에서 벗어나 추천 시스템의 사회적 영향에 대한 관심으로 연구 영역이 확장되었다. 이러한 연구 영역의 확장과 더불어 새로운 방법론적 시도가 이루어졌는데, 이 글에서는 온라인 뉴스 추천서비스 사용자들의 파편화 문제를 관계망 분석을 통해 다룬 일련의 연구들과 추천 알고리즘의 편향과 차별 문제에 데이터마이닝 기법과 대규모 온라인 통제실험방법을 접목한 연구들과 연구에 사용된 몇 가지 기법을 소개한다.

RecSys Issues Ontology: A Knowledge Classification of Issues for Recommender Systems Researchers

Scholarly research has extensively examined a number of issues and challenges affecting recommender systems (e.g. ‘cold-start’, ‘scrutability’, ‘trust’, ‘context’, etc.). However, a comprehensive knowledge classification of the issues involved with recommender systems research has yet to be developed. A holistic knowledge representation of the issues affecting a domain is critical for research advancement. The aim of this study is to advance scholarly research within the domain of recommender systems through formal knowledge classification of issues and their relationships to one another within recommender systems research literature. In this study, we employ a rigorous ontology engineering process for development of a recommender system issues ontology. This ontology provides a formal specification of the issues affecting recommender systems research and development. The ontology answers such questions as, “What issues are associated with ‘trust’ in recommender systems research?”, “What are issues associated with improving and evaluating the ‘performance’ of a recommender system?” or “What ‘contextual’ factors might a recommender systems developer wish to consider in order to improve the relevancy and usefulness of recommendations?” Additionally, as an intermediate representation step in the ontology acquisition process, a concept map of recommender systems issues has been developed to provide conceptual visualization of the issues so that researchers may discern broad themes as well as relationships between concepts. These knowledge representations may aid future researchers wishing to take an integrated approach to addressing the challenges and limitations associated with current recommender systems research.

Apparel-based deep learning system design for apparel style recommendation

Purpose The big challenge in apparel recommendation system research is not the exploration of machine learning technologies in fashion, but to really understand clothes, fashion and people, and know what to learn. The purpose of this paper is to explore an advanced apparel style learning and recommendation system that can recognise deep design-associated features of clothes and learn the connotative meanings conveyed by these features relating to style and the body so that it can make recommendations as a skilled human expert. Design/methodology/approach This study first proposes a type of new clothes style training data. Second, it designs three intelligent apparel-learning models based on newly proposed training data including ATTRIBUTE, MEANING and the raw image data, and compares the models’ performances in order to identify the best learning model. For deep learning, two models are introduced to train the prediction model, one is a convolutional neural network joint with the baseline classifier support vector machine and the other is with a newly proposed classifier later kernel fusion. Findings The results show that the most accurate model (with average prediction rate of 88.1 per cent) is the third model that is designed with two steps, one is to predict apparel ATTRIBUTEs through the apparel images, and the other is to further predict apparel MEANINGs based on predicted ATTRIBUTEs. The results indicate that adding the proposed ATTRIBUTE data that captures the deep features of clothes design does improve the model performances (e.g. from 73.5 per cent, Model B to 86 per cent, Model C), and the new concept of apparel recommendation based on style meanings is technically applicable. Originality/value The apparel data and the design of three training models are originally introduced in this study. The proposed methodology can evaluate the pros and cons of different clothes feature extraction approaches through either images or design attributes and balance different machine learning technologies between the latest CNN and traditional SVM.

Research and Application of Clothing Recommendation System Combining Explicit Data and Implicit Data

Research and Application of Clothing Recommendation System Combining Explicit Data and Implicit Data

Hybrid Location-based Recommender System for Mobility and Travel Planning

In recent times, the modern developments of internet technologies and social networks have attracted global researchers to explore the recommender systems for generating personalized location-based services. Recommender Systems (RSs) as proven decision support tools have gained immense popularity to solve information overloading problem among various real-time applications of e-commerce, travel and tourism, movies and e-learning. RSs emerge as a popular and reliable information filtering approach that is capable of suggesting relevant items, movies, and locations to the active target user based on dynamic preferences and interests. Beyond the development of many feature-rich recommendation algorithms, the need for a better full-fledged RS to produce precise and highly relevant recommendations based on ratings and preferences provided by the target user is very high. With the specific focus to the travel domain, the global research community has been involved in the development of a complete travel recommender system that is immune to the sparsity and cold start problems. In this paper, we present a new Hybrid Location-based Travel Recommender System (HLTRS) through exploiting ensemble based co-training method with swarm intelligence algorithms to enhance the personalized travel recommendations. The proposed HLTRS is experimentally validated on the real-world large-scale dataset, and we have made an extensive user study to determine the ability of developed RS to produce user satisfiable recommendations in real-time scenarios. The obtained results and analyses demonstrate the improved performance of the proposed Hybrid Location-based Travel Recommender System over existing baselines of recommender systems research.

Order, context and popularity bias in next-song recommendations

The availability of increasingly larger multimedia collections has fostered extensive research in recommender systems. Instead of capturing general user preferences, the task of next-item recommendation focuses on revealing specific session preferences encoded in the most recent user interactions. This study focuses on the music domain, particularly on the task of music playlist continuation, a paradigmatic case of next-item recommendation. While the accuracy achieved in next-song recommendations is important, in this work we shift our focus toward a deeper understanding of fundamental playlist characteristics, namely the song order, the song context and the song popularity, and their relation to the recommendation of playlist continuations. We also propose an approach to assess the quality of the recommendations that mitigates known problems of off-line experiments for music recommender systems. Our results indicate that knowing a longer song context has a positive impact on next-song recommendations. We find that the long-tailed nature of the playlist datasets makes simple and highly expressive playlist models appear to perform comparably, but further analysis reveals the advantage of using highly expressive models. Finally, our experiments suggest that the song order is not crucial to accurately predict next-song recommendations.

Hybrid bio-inspired user clustering for the generation of diversified recommendations

The research and development of recommender systems are traditionally focused on the enhancement and guaranteeing the recommendation accuracy to achieve user satisfaction. On the other hand, the alternative recommendation qualities such as diversity and novelty have received significant attention from researchers in recent times. In this paper, we present a detailed study of the diversity in recommender systems to help researchers in the development of recommendation approaches to generate efficient recommendations. We have also analyzed the existing works for assessment of impact and quality of diversified recommendations. Based on our detailed investigation of the diversity in recommendations, we shift the generic focus from accuracy objectives to explore beyond the accuracy of recommendations. The need for recommender systems producing diversified recommendations without compromising the accuracy is very high to meet the growing demands of users. To address the personalization problem in travel recommender systems, we present the hybrid swarm intelligence clustering ensemble-based recommendation framework to generate diverse and accurate Point of Interest recommendations. Our proposed recommendation approach employs multiple swarm optimization algorithms to frame a clustering ensemble for the generation of efficient user clustering. We have evaluated our proposed recommendation approach over a real-time large-scale dataset of TripAdvisor to estimate the quality of recommendations in terms of diversity and accuracy. The experimental results demonstrate the enhanced efficiency of the proposed recommendation approach over state-of-the-art techniques.

Deep Learning Based Recommender System

With the growing volume of online information, recommender systems have been an effective strategy to overcome information overload. The utility of recommender systems cannot be overstated, given their widespread adoption in many web applications, along with their potential impact to ameliorate many problems related to over-choice. In recent years, deep learning has garnered considerable interest in many research fields such as computer vision and natural language processing, owing not only to stellar performance but also to the attractive property of learning feature representations from scratch. The influence of deep learning is also pervasive, recently demonstrating its effectiveness when applied to information retrieval and recommender systems research. The field of deep learning in recommender system is flourishing. This article aims to provide a comprehensive review of recent research efforts on deep learning-based recommender systems. More concretely, we provide and devise a taxonomy of deep learning-based recommendation models, along with a comprehensive summary of the state of the art. Finally, we expand on current trends and provide new perspectives pertaining to this new and exciting development of the field.

Personalized Micro-Service Recommendation System for Online News

In the era of artificial intelligence and high technology advance our life is dependent on them in every aspect. The dynamic environment forces us to plan our time with conscious and every minute is valuable. To help individuals and corporations see information that is only relevant to them, recommendation systems are in place. Popular platforms that such as Amazon, Ebay, Netflix, YouTube, make use of advanced recommendation systems to better serve the needed of their users. This research paper gives insight of building a microservice recommendation system for online news. Research in recommendation systems is mainly focused on improving user’s experience based mainly on personalization information, such as preferences, and searching history. To determine the initial preferences of a user an initial menu of topics/themes is provided for the user to choose from. In order to reflect as precise as possible the searching interests regarding news of user, all of his interactions are thoroughly recorded and in depth analyzed, based on advanced machine learning techniques, when adjusting the news topics, the user is interested for. Based on the aforementioned approach, a personalized recommendation system for online news has been developed. Existing techniques has been researched and evaluated to aid the decision about picking the best approach for the software to be implemented. Frameworks/technologies used for the development are Java 8, Spring boot, Spring MVC, Maven and MongoDB.

Learning to Style-Aware Bayesian Personalized Ranking for Visual Recommendation

Recently, product images have been gaining the attention of recommender system researchers in the field of visual recommendation. This is because the visual appearance of products has a significant impact on consumers' decisions. Extensive studies have been done to integrate the features extracted by convolutional neural networks directly into recommendations. This improves the performance of recommender systems. Style features, an important type of features, are rarely considered. Style features play a vital role in the visual recommendation as a user's decision depends largely on whether the product fits his/her style. However, the representation of the conventional image features fails in capturing the styles of a product. To bridge this gap, we propose introducing style feature modeling, which is highly relevant with user preference, into the visual recommendation model. Furthermore, we propose incorporating the style features into collaborative learning to create awareness pertaining to the preferences of users. The experiments conducted on two public implicit feedback datasets demonstrate the effectiveness of our approach for the visual recommendation.

Evolutionary learning approach to multi-agent negotiation for group recommender systems

Recommender systems (RSs) have emerged as a solution to the information overload problem by filtering and presenting the users with information, services etc. according to their preferences. RSs research has focused on algorithms for recommending items for individual users. However, in certain domains, it may be desirable to be able to recommend items for a group of persons, e.g., movies, restaurants, etc. for which some remarkable group recommender systems (GRSs) have been developed. GRSs provide recommendations to groups, i.e., they take all individual group members’ preferences into account and satisfy them optimally with a sequence of items. Taking into consideration the fact that each group member has different behaviour with respect to other members in the group, we propose a genetic algorithm (GA) based multi-agent negotiation scheme for GRS (GA-MANS-GRS) where each agent acts on behalf of one group member. The GA-MANS-GRS is modelled as many one-to-one bilateral negotiation schemes with two phases. In the negotiation phase, we have applied GA to obtain the maximum utility offer for each user and generated the most appropriate ranking for each individual in the group. For the recommendation generation phase, again GA is employed to produce the list of ratings with that minimizes the sum of distances among the preferences of the group members. Finally, the results of computational experiments are presented that establish the superiority of our proposed model over baseline GRSs techniques.