Leveraging Review Properties for Effective Recommendation

Cited by: 0|Views14
Weibo:
We proposed the review-based review properties-based recommendation model model, which leverages the importance of different review properties in capturing the usefulness of reviews thereby enhancing the recommendation performance

Abstract:

Many state-of-the-art recommendation systems leverage explicit item reviews posted by users by considering their usefulness in representing the users' preferences and describing the items' attributes. These posted reviews may have various associated properties, such as their length, their age since they were posted, or their item rating...More

Code:

Data:

0
Full Text
Bibtex
Weibo
Introduction
  • There has been an increase in the amount of available information and interaction choices online.
  • The reviews posted by users on items have a corresponding set of properties, such as their length, the number of days since they were posted or their writing style.
  • A number of studies [32, 44] have previously attempted to leverage the review properties when making recommendations.
  • The underlying premise of such studies is that the review properties encapsulate rich information about both the users’ preferences and the items’ attributes
Highlights
  • In recent years, there has been an increase in the amount of available information and interaction choices online
  • Our experiments focus on investigating the performance of review properties-based recommendation model (RPRM) as well as the effectiveness of our proposed loss functions and negative sampling strategies in comparison to the six strong baselines from the literature
  • We observe that NoProp significantly outperforms all baseline approaches, including the state-of-the-art recommendation approaches, according to both the paired t-test and the Tukey Honest Significant Difference (HSD) test regardless of whether they use any review information
  • We proposed the review-based RPRM model, which leverages the importance of different review properties in capturing the usefulness of reviews thereby enhancing the recommendation performance
  • Inspired by the users’ adoption of information framework [39], we proposed two new loss functions and a negative sampling strategy that account for the usefulness of the review properties
  • We have illustrated the recommendation added-value of RPRM by examining the usefulness of several review properties for a sample of users and their interacted items. This analysis has exemplified the promise of RPRM in guiding online review platforms in customising the presentation of reviews and deploying more effective recommendation systems
Methods
  • The authors describe the proposed RPRM model, which leverages the reviews posted by users to enhance the recommendation task.
  • RPRM takes into account the review properties when modelling the user/item information by learning the importance of different review properties for enriching the representations of the users’ preferences and the items’ attributes.
  • For the length property, the score will correspond to the length of the review
  • These scores could be computed for any property, provided that the property values are mapped into scalars in the range of [0..1] using an adequate function.
  • The computed property scores enable the modelling of reviews from different aspects and examine the relationship between the review usefulness and the review properties
Results
  • RESULTS AND ANALYSIS

    The authors present and analyse the results of the experiments to answer the research questions in Section 4.
  • The BPR-MF approach is a strong baseline and was shown recently to outperform various state-of-theart recommendation approaches from the literature [35]
  • Among these three baselines, NARRE significantly outperforms both the DeepCoNN and JRL approaches according to both the paired t-test and the Tukey HSD test on the two datasets with higher evaluation scores.
  • NARRE is significantly outperformed by the No-Prop variant, despite No-Prop having a simpler structure than NARRE
  • The effectiveness of this simple review-based recommendation approach is consistent with the conclusions in [36].
  • The use of the embedding vectors, which model the users’ preferences and items’ attributes, explains the superior performances of both the No-Prop and NARRE models in comparison to other baseline approaches
Conclusion
  • The authors proposed the review-based RPRM model, which leverages the importance of different review properties in capturing the usefulness of reviews thereby enhancing the recommendation performance.
  • These results demonstrated the advantages of leveraging the agreement on the review properties’ importance between users and items.
  • The authors have illustrated the recommendation added-value of RPRM by examining the usefulness of several review properties for a sample of users and their interacted items.
  • This analysis has exemplified the promise of RPRM in guiding online review platforms in customising the presentation of reviews and deploying more effective recommendation systems
Summary
  • Introduction:

    There has been an increase in the amount of available information and interaction choices online.
  • The reviews posted by users on items have a corresponding set of properties, such as their length, the number of days since they were posted or their writing style.
  • A number of studies [32, 44] have previously attempted to leverage the review properties when making recommendations.
  • The underlying premise of such studies is that the review properties encapsulate rich information about both the users’ preferences and the items’ attributes
  • Objectives:

    To address the recommendation task, the authors aim to accurately estimate the users’ preferences on items so that the authors rank the items that a given user might find the most interesting in higher ranks
  • Methods:

    The authors describe the proposed RPRM model, which leverages the reviews posted by users to enhance the recommendation task.
  • RPRM takes into account the review properties when modelling the user/item information by learning the importance of different review properties for enriching the representations of the users’ preferences and the items’ attributes.
  • For the length property, the score will correspond to the length of the review
  • These scores could be computed for any property, provided that the property values are mapped into scalars in the range of [0..1] using an adequate function.
  • The computed property scores enable the modelling of reviews from different aspects and examine the relationship between the review usefulness and the review properties
  • Results:

    RESULTS AND ANALYSIS

    The authors present and analyse the results of the experiments to answer the research questions in Section 4.
  • The BPR-MF approach is a strong baseline and was shown recently to outperform various state-of-theart recommendation approaches from the literature [35]
  • Among these three baselines, NARRE significantly outperforms both the DeepCoNN and JRL approaches according to both the paired t-test and the Tukey HSD test on the two datasets with higher evaluation scores.
  • NARRE is significantly outperformed by the No-Prop variant, despite No-Prop having a simpler structure than NARRE
  • The effectiveness of this simple review-based recommendation approach is consistent with the conclusions in [36].
  • The use of the embedding vectors, which model the users’ preferences and items’ attributes, explains the superior performances of both the No-Prop and NARRE models in comparison to other baseline approaches
  • Conclusion:

    The authors proposed the review-based RPRM model, which leverages the importance of different review properties in capturing the usefulness of reviews thereby enhancing the recommendation performance.
  • These results demonstrated the advantages of leveraging the agreement on the review properties’ importance between users and items.
  • The authors have illustrated the recommendation added-value of RPRM by examining the usefulness of several review properties for a sample of users and their interacted items.
  • This analysis has exemplified the promise of RPRM in guiding online review platforms in customising the presentation of reviews and deploying more effective recommendation systems
Tables
  • Table1: Recommendation performances. Significant differences w.r.t. ‘No-Prop’ are indicated by ‘*’ (according to both the paired t-test and the Tukey HSD test, < 0.05). 1/2/3 denote a significant difference according to both tests w.r.t. to the indicated approach. ↑ indicates that the corresponding approach is significantly outperformed by RPRM on all ranking metrics according to both tests
  • Table2: Impact of the model’s learning schemes on RPRM. Statistically significant differences with respect to ‘RPRM-basic’ are indicated by ‘*’ (according to both the paired t-test and the Tukey HSD test, < 0.05)
Download tables as Excel
Related work
  • We briefly discuss three bodies of related work, namely recommendation approaches based on reviews, recommendation approaches leveraging the use of review properties, and work investigating users’ behaviour while interacting with information.

    2.1 Review-based Recommendations

    The main objective of applying a recommendation model is to observe the users’ behaviours and to learn how to distinguish among items for a given user, thereby estimating the users’ preferences and recommending suitable items that the users might be interested in. User-generated reviews encapsulate rich semantic information such as the possible explanation of the users’ preferences and the description of specific item attributes [5]. Therefore, many recommendation models have aimed to leverage these reviews to construct user/item representations and to address the recommendation task [1, 7, 15, 19, 27]. Many previous review-based recommendation approaches captured the semantic similarity between the review content [4, 53], which allows to encode additional relationships among the users and items, allowing to better suggest items the users might be interested in. Indeed, the posted reviews by users are valuable in modelling the interactions among users and items from a textual semantic perspective. However, the quality and usefulness of the reviews markedly vary with the increasing amount of users and the available reviews they post online. Therefore, Chen et al [4] applied an attention mechanism to estimate the usefulness of different reviews. Unlike previous work [2, 4] , which used an attention mechanism to learn the usefulness of reviews, we argue that the review properties can be directly leveraged to effectively capture the usefulness of reviews. Moreover, there are many existing approaches [24, 32, 44] that extract the review properties and integrate them as side or contextual information to enhance the recommendation performance. However, unlike our work in this paper, such approaches do not make use of the reviews themselves. In the following, we further describe such approaches using the properties as side information.
Funding
  • Our results show that RPRM significantly outperforms a classical and five state-of-the-art baselines
  • (3) We show that RPRM significantly outperforms one classical and five state-of-the-art recommendation approaches on the commonly-used Amazon and Yelp datasets
  • We observe that NoProp significantly outperforms all baseline approaches, including the state-of-the-art recommendation approaches (namely CASER, DeepCoNN and NARRE), according to both the paired t-test and the Tukey HSD test regardless of whether they use any review information
  • The BPR-MF approach is a strong baseline and was shown recently to outperform various state-of-theart recommendation approaches from the literature [35]
  • The observed performances significantly outperform both No-Prop and all the baseline approaches, including the existing state-of-the-art recommendation models (namely NARRE, CASER and DeepCoNN),) according to both the paired t-test and the Tukey HSD test
  • Furthermore, by integrating all six review properties and weighting their importance in the full RPRM model, we observe that RPRM achieves the best performance among all tested approaches on both used datasets
  • From Table 2, we observe that our proposed two loss functions can consistently improve the performance of the basic RPRM with the exception of the -Cos model setup on the Amazon dataset
Study subjects and analysis
randomly selected users: 3
In Section 5, we showed the effectiveness of modelling the agreement between the users and items in terms of the reviews’ properties. Therefore, in this section, we use three randomly selected users to illustrate the users’ preferences on different review properties and the agreement on the importance of review properties between the users and their interacted items. To this end, Figure 2(a)-(c) plots the learned RPRM property importance scores for the review properties of three randomly selected users, say A, B & C, as well as their interacted items

randomly selected users: 3
Therefore, in this section, we use three randomly selected users to illustrate the users’ preferences on different review properties and the agreement on the importance of review properties between the users and their interacted items. To this end, Figure 2(a)-(c) plots the learned RPRM property importance scores for the review properties of three randomly selected users, say A, B & C, as well as their interacted items. The users’ property importance preferences are shown using a blue dashed line with square markers; their interacted items in the test set are also shown (solid lines in Figure 2(a) and 2(b) and dots in different colours in Figure 2(c)) from the Amazon dataset

Reference
  • Amjad Almahairi, Kyle Kastner, Kyunghyun Cho, and Aaron Courville. 2015. Learning distributed representations from reviews for collaborative filtering. In Proc. of RecSys.
    Google ScholarLocate open access versionFindings
  • Konstantin Bauman, Bing Liu, and Alexander Tuzhilin. 2017. Aspect based recommendations: Recommending items with the most valuable aspects based on user reviews. In Proc. of SIGKDD.
    Google ScholarLocate open access versionFindings
  • Yolanda YY Chan and Eric WT Ngai. 2011. Conceptualising electronic word of mouth activity. Marketing Intelligence & Planning (2011).
    Google ScholarLocate open access versionFindings
  • Chong Chen, Min Zhang, Yiqun Liu, and Shaoping Ma. 2018. Neural attentional rating regression with review-level explanations. In Proc. of WWW.
    Google ScholarLocate open access versionFindings
  • Li Chen, Guanliang Chen, and Feng Wang. 201Recommender systems based on user reviews: the state of the art. User Modeling and User-Adapted Interaction 25, 2 (2015), 99–154.
    Google ScholarLocate open access versionFindings
  • Xu Chen, Hanxiong Chen, Hongteng Xu, Yongfeng Zhang, Yixin Cao, Zheng Qin, and Hongyuan Zha. 2019. Personalized fashion recommendation with visual explanations based on multimodal attention network: Towards visually explainable recommendation. In Proc. of SIGIR.
    Google ScholarLocate open access versionFindings
  • Yifan Chen, Yang Wang, Xiang Zhao, Hongzhi Yin, Ilya Markov, and MAARTEN De Rijke. 2020. Local Variational Feature-based Similarity Models for Recommending Top-N New Items. ACM Transactions on Information Systems (TOIS) 38, 2 (2020), 1–33.
    Google ScholarLocate open access versionFindings
  • Zhiyong Cheng, Ying Ding, Xiangnan He, Lei Zhu, Xuemeng Song, and Mohan S Kankanhalli. 201A3NCF: An Adaptive Aspect Attention Model for Rating Prediction. In Proc. of IJCAI.
    Google ScholarLocate open access versionFindings
  • Cindy Man-Yee Cheung, Choon-Ling Sia, and Kevin KY Kuan. 2012. Is this review believable? A study of factors affecting the credibility of online consumer reviews from an ELM perspective. Journal of the Association for Information Systems 13, 8 (2012).
    Google ScholarLocate open access versionFindings
  • Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In Proc. of NAACL-HLT.
    Google ScholarLocate open access versionFindings
  • Raffaele Filieri and Fraser McLeay. 2014. E-WOM and accommodation: An analysis of the factors that influence travelers’ adoption of information from online reviews. Journal of Travel Research 53, 1 (2014), 44–57.
    Google ScholarLocate open access versionFindings
  • Xinyu Guan, Zhiyong Cheng, Xiangnan He, Yongfeng Zhang, Zhibo Zhu, Qinke Peng, and Tat-Seng Chua. 2019. Attentive Aspect Modeling for Review-aware Recommendation. ACM Transactions on Information Systems (TOIS) 37, 3 (2019), 28.
    Google ScholarLocate open access versionFindings
  • Qing Guo, Zhu Sun, Jie Zhang, and Yin-Leng Theng. 2020. An Attentional Recurrent Neural Network for Personalized Next Location Recommendation. In Proc. of AAAI.
    Google ScholarLocate open access versionFindings
  • Ruining He and Julian McAuley. 2016. Ups and downs: Modeling the visual evolution of fashion trends with one-class collaborative filtering. In Proc. of WWW.
    Google ScholarLocate open access versionFindings
  • Xiangnan He, Tao Chen, Min-Yen Kan, and Xiao Chen. 20Trirank: Reviewaware explainable recommendation by modeling aspects. In Proc. of CIKM.
    Google ScholarLocate open access versionFindings
  • Yoon Kim. 2014. Convolutional Neural Networks for Sentence Classification. In Proc. of EMNLP.
    Google ScholarLocate open access versionFindings
  • Diederik P. Kingma and Jimmy Ba. 2015. Adam: A Method for Stochastic Optimization. In Proc. of ICLR.
    Google ScholarLocate open access versionFindings
  • Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, and Radu Soricut. 2020. ALBERT: A Lite BERT for Self-supervised Learning of Language Representations. In Proc. of ICLR.
    Google ScholarLocate open access versionFindings
  • Chenliang Li, Xichuan Niu, Xiangyang Luo, Zhenzhong Chen, and Cong Quan. 20A Review-Driven Neural Model for Sequential Recommendation. In Proc. of IJCAI.
    Google ScholarLocate open access versionFindings
  • Yuqi Li, Weizheng Chen, and Hongfei Yan. 2017. Learning Graph-based Embedding For Time-Aware Product Recommendation. In Proc. of CIKM.
    Google ScholarLocate open access versionFindings
  • Guang Ling, Michael R Lyu, and Irwin King. 2014. Ratings meet reviews, a combined approach to recommend. In Proc. of RecSys.
    Google ScholarLocate open access versionFindings
  • Wei Liu, Zhi-Jie Wang, Bin Yao, and Jian Yin. 2019. Geo-ALM: POI Recommendation by Fusing Geographical Information and Adversarial Learning Mechanism.. In Proc. of IJCAI.
    Google ScholarLocate open access versionFindings
  • Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. RoBERTa: A Robustly Optimized BERT Pretraining Approach. CoRR abs/1907.11692 (2019).
    Findings
  • Jarana Manotumruksa, Craig Macdonald, and Iadh Ounis. 2018. A contextual attention recurrent architecture for context-aware venue recommendation. In Proc. of SIGIR.
    Google ScholarLocate open access versionFindings
  • Julian McAuley and Jure Leskovec. 2013. Hidden factors and hidden topics: understanding rating dimensions with review text. In Proc. of RecSys.
    Google ScholarLocate open access versionFindings
  • Julian McAuley, Rahul Pandey, and Jure Leskovec. 2015. Inferring networks of substitutable and complementary products. In Proc. of SIGKDD. 785–794.
    Google ScholarLocate open access versionFindings
  • Jianmo Ni, Jiacheng Li, and Julian McAuley. 2019. Justifying recommendations using distantly-labeled reviews and fine-grained aspects. In Proc. of EMNLPIJCNLP. 188–197.
    Google ScholarLocate open access versionFindings
  • Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. 2019. Pytorch: An imperative style, high-performance deep learning library. In Proc. of NeurIPS.
    Google ScholarLocate open access versionFindings
  • Richard E Petty and John T Cacioppo. 2012. Communication and persuasion: Central and peripheral routes to attitude change. Springer Science & Business Media.
    Google ScholarFindings
  • Hamed Qahri-Saremi and Ali Reza Montazemi. 2019. Factors Affecting the Adoption of an Electronic Word of Mouth Message: A Meta-Analysis. Journal of Management Information Systems 36, 3 (2019).
    Google ScholarLocate open access versionFindings
  • Tieyun Qian, Bei Liu, Quoc Viet Hung Nguyen, and Hongzhi Yin. 2019. Spatiotemporal representation learning for translation-based POI recommendation. ACM Transactions on Information Systems (TOIS) 37, 2 (2019), 1–24.
    Google ScholarLocate open access versionFindings
  • Sindhu Raghavan, Suriya Gunasekar, and Joydeep Ghosh. 2012. Review quality aware collaborative filtering. In Proc. of RecSys.
    Google ScholarLocate open access versionFindings
  • Zhaochun Ren, Shangsong Liang, Piji Li, Shuaiqiang Wang, and Maarten de Rijke. 2017. Social collaborative viewpoint regression with explainable recommendations. In Proc. of WSDM.
    Google ScholarLocate open access versionFindings
  • Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme. 2009. BPR: Bayesian Personalized Ranking from Implicit Feedback. In Proc. of UAI.
    Google ScholarLocate open access versionFindings
  • Steffen Rendle, Walid Krichene, Li Zhang, and John R. Anderson. 2020. Neural Collaborative Filtering vs. Matrix Factorization Revisited. In Proc. of RecSys.
    Google ScholarLocate open access versionFindings
  • Noveen Sachdeva and Julian McAuley. 2020. How Useful are Reviews for Recommendation? A Critical Review and Potential Improvements. In Proc. of SIGIR.
    Google ScholarLocate open access versionFindings
  • Tetsuya Sakai. 2018. Laboratory Experiments in Information Retrieval - Sample Sizes, Effect Sizes, and Statistical Power. Springer.
    Google ScholarFindings
  • Sungyong Seo, Jing Huang, Hao Yang, and Yan Liu. 2017. Interpretable convolutional neural networks with dual local and global attention for review rating prediction. In Proc. of RecSys.
    Google ScholarLocate open access versionFindings
  • Stephanie Watts Sussman and Wendy Schneier Siegal. 2003. Informational influence in organizations: An integrated approach to knowledge adoption. Information systems research 14, 1 (2003), 47–65.
    Google ScholarLocate open access versionFindings
  • Jiaxi Tang and Ke Wang. 2018. Personalized top-n sequential recommendation via convolutional sequence embedding. In Proc. of WSDM.
    Google ScholarLocate open access versionFindings
  • Liang Rebecca Tang, Soocheong Shawn Jang, and Alastair Morrison. 2012. Dualroute communication of destination websites. Tourism Management 33, 1 (2012), 38–49.
    Google ScholarLocate open access versionFindings
  • Petar Veličković, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. 2018. Graph Attention Networks. Proc. of ICLR.
    Google ScholarFindings
  • Xi Wang, Anjie Fang, Iadh Ounis, and Craig Macdonald. 2019. Evaluating Similarity Metrics for Latent Twitter Topics. In Proc. of ECIR.
    Google ScholarLocate open access versionFindings
  • Xi Wang, Iadh Ounis, and Craig Macdonald. 2019. Comparison of Sentiment Analysis and User Ratings in Venue Recommendation. In Proc. of ECIR.
    Google ScholarLocate open access versionFindings
  • Xi Wang, Iadh Ounis, and Craig Macdonald. 2020. Negative Confidence-Aware Weakly Supervised Binary Classification for Effective Review Helpfulness Classification. In Proc. of CIKM.
    Google ScholarLocate open access versionFindings
  • Chenghuan Wu and David R Shaffer. 1987. Susceptibility to persuasive appeals as a function of source credibility and prior experience with the attitude object. Journal of personality and social psychology 52, 4 (1987), 677.
    Google ScholarLocate open access versionFindings
  • Jibang Wu, Renqin Cai, and Hongning Wang. 2020. Déjà vu: A Contextualized Temporal Attention Mechanism for Sequential Recommendation. In Proc. of WWW.
    Google ScholarLocate open access versionFindings
  • Jheng-Hong Yang, Chih-Ming Chen, Chuan-Ju Wang, and Ming-Feng Tsai. 2018. HOP-rec: high-order proximity for implicit recommendation. In Proc. of RecSys.
    Google ScholarLocate open access versionFindings
  • Feng Yu, Qiang Liu, Shu Wu, Liang Wang, and Tieniu Tan. 2016. A dynamic recurrent model for next basket recommendation. In Proc. of SIGIR.
    Google ScholarLocate open access versionFindings
  • Jia-Dong Zhang and Chi-Yin Chow. 2015. GeoSoCa: Exploiting geographical, social and categorical correlations for point-of-interest recommendations. In Proc. of SIGIR.
    Google ScholarLocate open access versionFindings
  • Yongfeng Zhang, Qingyao Ai, Xu Chen, and W Bruce Croft. 2017. Joint representation learning for top-n recommendation with heterogeneous information sources. In Proc. of CIKM.
    Google ScholarLocate open access versionFindings
  • Pengpeng Zhao, Haifeng Zhu, Yanchi Liu, Jiajie Xu, Zhixu Li, Fuzhen Zhuang, Victor S Sheng, and Xiaofang Zhou. 2019. Where to go next: A spatio-temporal gated network for next poi recommendation. In Proc. of AAAI.
    Google ScholarLocate open access versionFindings
  • Lei Zheng, Vahid Noroozi, and Philip S Yu. 2017. Joint deep modeling of users and items using reviews for recommendation. In Proc. of WSDM.
    Google ScholarLocate open access versionFindings
Your rating :
0

 

Tags
Comments