TopicMF: Simultaneously Exploiting Ratings and Reviews for Recommendation

By Yang Bao, Hui Fang, Jie Zhang. 2014

302 citations recorded by the source corpus.

Although users' preference is semantically reflected in the free-form review texts, this wealth of information was not fully exploited for learning recommender models. Specifically, almost all existing recommendation algorithms only exploit rating scores in order to find users' preference, but ignore the review texts accompanied with rating information. In this paper, we propose a novel matrix factorization model (called TopicMF) which simultaneously considers the ratings and accompanied review texts. Experimental results on 22 real-world datasets show the superiority of our model over the state-of-the-art models, demonstrating its effectiveness for recommendation tasks.

Subfields: Artificial Intelligence, Information Systems

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Yongfeng Zhang
2. Xu Chen
3. Philip S. Yu
4. Qiang Yang
5. Xiangnan He
6. Wayne Xin Zhao
7. Ji-rong Wen
8. Yu Zhang

In Information Systems

1. Yongfeng Zhang
2. Xu Chen
3. Xiangnan He
4. Qingyao Ai
5. Philip S. Yu
6. Xing Xie
7. Ji-rong Wen
8. Wayne Xin Zhao

Citations in the corpus

• VBPR: Visual Bayesian Personalized Ranking from Implicit Feedback (2015, methodology): “Others have developed content-based and hybrid models that make use of a variety of information sources, including text (and context), taxonomies, and user demographics (Bao, Fang, and Zhang, 2014; Kanagal et al., 2012; Lu et al., 2015; Qiao et al., 2014).”
• Learning Distributed Representations from Reviews for Collaborative Filtering (2015, methodology): “[13], [12] and [1]), do not clearly identify how the data was split into train and test sets.”
• Joint Deep Modeling of Users and Items Using Reviews for Recommendation (2017, methodology): “All competing techniques which are based on topic modeling [38, 3, 8] use the traditional bag of words techniques.”
• Improving Conversational Recommender Systems via Knowledge Graph based Semantic Fusion (2020, methodology): “er, user-item interaction data is usually sparse. To tackle the data sparsity problem, many techniques have been developed by utilizing the side information of items, such as review and taxonomy data [1, 9, 30]. As a comparison, CRS mainly focuses on the recommendation setting through conversation i”
• Interpretable Convolutional Neural Networks with Dual Local and Global Attention for Review Rating Prediction (2017, methodology): “As pointed out in [3], the authors hope to model the user’s preferences directly to their rating behavior in their future work, which is what our model is designed to do.”
• Rating-Boosted Latent Topics: Understanding Users and Items with Ratings and Reviews (2016, background): “The core of a recommendation system is a personalized algorithm for identifying the preference of users based on their feedback towards items [Bao and Zhang, 2014].”
• Explanation Mining: Post Hoc Interpretability of Latent Factor Models for Recommendation Systems (2018, background): “Whenanexternaltextualdatasource is available (e.g. product reviews [36] or news articles [15]), topic modelling can be used to discover the themes present in the text [9, 15, 31] which can then be combined with the rating predictions to provide an interpretation of the latent factors.”
• A Context-Aware User-Item Representation Learning for Item Recommendation (2017, methodology): “orization (MF) framework to derive item preferences. RMR [20] uses similar technique to derive topic factors from text but they use Gaussian mixtures to model ratings instead of MF framework. TopicMF [3] uses MF technique to jointly model the interaction data and topics from review text. In their tr”
• Relational Stacked Denoising Autoencoder for Tag Recommendation (2015, methodology): “The third class of methods (Wu et al. 2009; Wang and Blei 2011; Yang, Zhang, and Wang 2013; Zhao et al. 2013; Bao, Fang, and Zhang 2014; Chen et al. 2014), also the most popular and effective ones, consists of hybrid methods.”
• MMALFM (2018, methodology): “ed rating prediction models, which assume a one-to-one correspondence relationship between the latent topics (learned from reviews) and the latent factors (learned from rating), such as the models in [2, 41, 45, 63]. Besides, our model could learn an aspect importance for each user-item pair, namely”
• Transfer Meets Hybrid: A Synthetic Approach for Cross-Domain Collaborative Filtering with Text (2019, methodology): “Combining matrix factorization and topic modelling technique (Topic MF) is an effective way to integrate ratings with item contents [2, 29, 33].”
• Aspect-Aware Latent Factor Model: Rating Prediction with Ratings and Reviews (2018, methodology): “HFT [26] and TopicMF [2] link the latent topics and latent factors by using a defined transform function.”
• A Capsule Network for Recommendation and Explaining What You Like and Dislike (2019, methodology): “Many recommender systems have been developed by exploiting the semantic information covered in reviews [1, 3–6, 15, 18, 20, 24, 26, 27, 29, 37, 39] against using naturally sparse user-item ratings, leading to signficant performance gain.”
• Joint Representation Learning for Top-N Recommendation with Heterogeneous Information Sources (2017, background): “One trend is to augment numerical ratings with textual reviews for recommendation [8], which includes topic modeling [3, 26, 27, 34], sentiment analysis [9, 13, 39, 42–44], and (deep) neural network [2, 36, 41, 45] approaches for review modeling.”