Cross-Domain Recommendation via Deep Domain Adaptation

Kanagawa, Heishiro; Kobayashi, Hayato; Sato, Nobuyuki; Tagami, Yukihiro; Suzuki, Taiji

doi:10.1007/978-3-030-15719-7_3

Cited by 80 publications

(34 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…There are several models for cross-domain recommendation without overlap. Kanagawa et al [11] introduced an interesting task -recommending items from the source domain to users from the target domain based on text. To do so, textual features are extracted by an auto-encoder and aligned by domain adaptation.…”

Section: Cross-domain Recommendation Withoutmentioning

confidence: 99%

“…The basic models introduced in [11,23] are all text-based models. Kanagawa et al [11] devised an auto-encoder and Kanagawa et al [11] leveraged LSTM to extract user and item textual representations. Zheng et al [30] gathered reviews for each user and item, and extract textual features from these reviews by a Convolutional Neural Network (CNN) and Chen et al [2] further added an attention mechanism.…”

Section: Text-based Recommendationmentioning

confidence: 99%

“…Wang et al [23] proposed Long Short-Term Memory (LSTM) to construct user and item textual representations, and aligned them of the two domains for transfer learning. As we can see, these models [11,23] achieve domain adaptation in the textual space since embeddings are difficult to align without overlap. In this case, cross-domain recommendation in [11,23] is more close to a Natural Language Processing (NLP) task rather than a recommendation task.…”

Section: Cross-domain Recommendation Withoutmentioning

confidence: 99%

See 2 more Smart Citations

Semi-supervised Collaborative Filtering by Text-enhanced Domain Adaptation

Lin

et al. 2020

Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery &Amp; Data Mining

View full text Add to dashboard Cite

Data sparsity is an inherent challenge in the recommender systems, where most of the data is collected from the implicit feedbacks of users. This causes two difficulties in designing effective algorithms: first, the majority of users only have a few interactions with the system and there is no enough data for learning; second, there are no negative samples in the implicit feedbacks and it is a common practice to perform negative sampling to generate negative samples. However, this leads to a consequence that many potential positive samples are mislabeled as negative ones and data sparsity would exacerbate the mislabeling problem. To solve these difficulties, we regard the problem of recommendation on sparse implicit feedbacks as a semi-supervised learning task, and explore domain adaption to solve it. We transfer the knowledge learned from dense data to sparse data and we focus on the most challenging case âĂŤ there is no user or item overlap. In this extreme case, aligning embeddings of two datasets directly is rather sub-optimal since the two latent spaces encode very different information. As such, we adopt domain-invariant textual features as the anchor points to align the latent spaces. To align the embeddings, we extract the textual features for each user and item and feed them into a domain classifier with the embeddings of users and items. The embeddings are trained to puzzle the classifier and textual features are fixed as anchor points. By domain adaptation, the distribution pattern in the source domain is transferred to the target domain. As the target part can be supervised by domain adaptation, we abandon negative sampling in target dataset to avoid label noise.

show abstract

Section: Cross-domain Recommendation Withoutmentioning

confidence: 99%

Section: Text-based Recommendationmentioning

confidence: 99%

Section: Cross-domain Recommendation Withoutmentioning

confidence: 99%

See 1 more Smart Citation

Semi-supervised Collaborative Filtering by Text-enhanced Domain Adaptation

Lin

et al. 2020

Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery &Amp; Data Mining

View full text Add to dashboard Cite

show abstract

“…Lian et al [26] propose to incorporate content information into the multi-view neural network. Kanagawa et al [21] go further in this direction and formulate cross-domain recommendation as extreme multi-class classification, where only content features are used for adapting a classifier trained in the source domain to the target domain. These methods focus on learning better representations for users and items, while emphasizing less learning their interaction patterns, which is important for voice shopping.…”

Section: Transfer Learning For Recommendationmentioning

confidence: 99%

“…In the recommendation literature, transfer learning has been implemented by extending recommendation models, e.g., factorization models [28,29,33] or neural networks [14,21,26], with shared user 1 representations for cross-domain recommendation tasks. While being different in the underlying recommendation models (see Section 2 for a detailed discussion), both classes of methods are designed for transferring user representations with less emphasis on transferring the interaction patterns, which has to be carefully considered in our Web-to-Voice transfer context.…”

Section: Introductionmentioning

confidence: 99%

Web-to-Voice Transfer for Product Recommendation on Voice

Zhang

Yang

2020

Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval

View full text Add to dashboard Cite

While product recommendation algorithms on the Web are wellsupported by a vast amount of interaction data, the same is not true on Voice. A promising approach to mitigate the issue is transfer learning, i.e., transferring the knowledge of customers' shopping behaviors learned from their shopping activities on the Web to Voice. Such a Web-to-Voice transfer is challenging due to customers' distinct shopping behaviors on Voice: customers are inclined to purchase more low-consideration products and are more likely to purchase certain products repeatedly. This paper presents TransV, a novel Web-to-Voice neural transfer network that allows for effective transfer of customers' shopping patterns from the Web to Voice, while taking into account customers' distinct purchase patterns on Voice. Our method extends the state-of-the-art self-attention neural architecture with a multi-level tri-factorization neural component, which allows to explicitly capture the similarity and dissimilarity of customers' shopping patterns on the Web and Voice. To model repeated purchases, TransV adopts a recency-based copy mechanism that considers the impact of the recency of historical purchases on customers' behavior of repeated purchases. Extensive validation on multiple real-world datasets, including two cross-platform datasets from Amazon.com and Amazon Alexa, shows that our method is able to improve voice-based recommendation substantially by 26.8% as compared with non-transfer learning methods. CCS CONCEPTS • Information systems → Recommender systems; • Computing methodologies → Transfer learning; Neural networks.

show abstract

Leverage side information for top‐N recommendation with latent Gaussian process

Wang

Yang

et al. 2019

Concurrency and Computation

View full text Add to dashboard Cite

Currently, Recommender Systems (RS) have been ubiquitously applied to various online applications and obtained tremendous success due to their capability to overcome information overload; however, the available Side Information (SI), such as demographics and attributions of items, is always neglected. Actually, SI could reflect user's interests and preference, and even influence user's decision over items; therefore, it would be greatly helpful to leverage side information to improve the performance of RS, which is already partly certificated in previous research studies. Motivated by this, in this article, a novel multi-task learning-based recommendation system referred to as LGP-RS is proposed, which takes utility functions mapping from the non-linear side information to low-rank feature space with Gaussian process (GP) priors, for users and items, respectively. In addition, Laplace approximation method is employed to approximate the posterior distribution for the utility functions, and gradient-based method is employed to learn the hyperparameters for GP. This flexible approach is able to capture the non-linear high-order interactions within side information, reduce the prediction uncertainty, and provide personalized top-N recommendation. Experimental analysis over three real-world datasets demonstrates that LGP-RS could deal with both explicit and implicit side information simultaneously, and significantly outperforms state-of-the-art approaches.

show abstract

Cross-Domain Recommendation via Deep Domain Adaptation

Cited by 80 publications

References 27 publications

Semi-supervised Collaborative Filtering by Text-enhanced Domain Adaptation

Semi-supervised Collaborative Filtering by Text-enhanced Domain Adaptation

Web-to-Voice Transfer for Product Recommendation on Voice

Leverage side information for top‐N recommendation with latent Gaussian process

Contact Info

Product

Resources

About