Translation-based Recommendation

By Ruining He, Wang-Cheng Kang, Julian McAuley. ACM Conference on Recommender Systems, 2017

477 citations recorded by the source corpus.

Modeling the complex interactions between users and items as well as amongst items themselves is at the core of designing successful recommender systems. One classical setting is predicting users' personalized sequential behavior (or 'next-item' recommendation), where the challenges mainly lie in modeling 'third-order' interactions between a user, her previously visited item(s), and the next item to consume. Existing methods typically decompose these higher-order interactions into a combination of pairwise relationships, by way of which user preferences (user-item interactions) and sequential patterns (item-item interactions) are captured by separate components. In this paper, we propose a unified method, TransRec, to model such third-order relationships for large-scale sequential prediction. Methodologically, we embed items into a 'transition space' where users are modeled as translation vectors operating on item sequences. Empirically, this approach outperforms the state-of-the-art on a wide spectrum of real-world datasets. Data and code are available at https://sites.google.com/a/eng.ucsd.edu/ruining-he/.

Subfields: Artificial Intelligence, Information Systems

View Paper: https://www.semanticscholar.org/paper/d5a823053663a77869d4b38a8026c51ebc2e975e

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. Xiangnan He
2. Xiang Wang
3. Tat-seng Chua
4. Yongfeng Zhang
5. Wayne Xin Zhao
6. Xu Chen
7. Yi Tay
8. Julian McAuley

In Information Systems

1. Xiangnan He
2. Xiang Wang
3. Yongfeng Zhang
4. Tat-seng Chua
5. Xu Chen
6. Wayne Xin Zhao
7. Meng Wang
8. Julian McAuley

Citations in the corpus

• Sequential Recommender Systems: Challenges, Progress and Prospects (2019, background): “an interaction score between users and items, or successive users’ actions [18; 19], while other works directly utilize them to calculate a metric like the Euclidean distance as the interaction score [5]. This model has shown great potential in recent years due to its simplicity, efficiency and effi”
• AMER: Automatic Behavior Modeling and Interaction Exploration in Recommender System (2020, background): “…• STAMP [45] simultaneously captures user’s long-term general interests and short-term current interests with attention module on MLP. STAMP is a typical sequential recommender system without RNN or CNN, representing the MF-based sequential recommender system such as FPMC[58] and TransRec [22].”
• Hierarchical Gating Networks for Sequential Recommendation (2019, methodology): “In [7], the translation-based method is proposed for sequential recommendation.”
• Time Interval Aware Self-Attention for Sequential Recommendation (2020, methodology): “To preprocess we follow the procedure from [7, 14, 21].”
• Reinforcement Knowledge Graph Reasoning for Explainable Recommendation (2019, methodology): “TransRec [6] invokes translation-based embeddings for sequential recommendation.”
• BERT4Rec: Sequential Recommendation with Bidirectional Encoder Representations from Transformer (2019, methodology): “Besides the first-order MCs, high-order MCs are also adopted to consider more previous items [10, 11].”
• A Survey on Accuracy-Oriented Neural Recommendation: From Collaborative Filtering to Information-Rich Recommendation (2021, methodology): “Given the item embedding matrix Q, each user’s embedding pu can be approximated as: qi þ pu qj [46].”
• Learning to Embed Categorical Features without Embedding Tables for Recommendation (2020, methodology): “As in [13], we treat all ratings as observed feedback, and sort the feedback according to timestamps.”
• Unifying Knowledge Graph Learning and Recommendation: Towards a Better Understanding of User Preferences (2019, methodology): “ationlearning.Itassumesthattheselection of items satisfies translational relations in the latent vector space, where the relations are either regarded as related to users in sequential recommendation [11], or modeled implicitly via Memory-based Attention [37]. We thus improve this type of method by ”
• Learning over Knowledge-Base Embeddings for Recommendation (2018, methodology): “ from the knowledge base for recommendation, but the whole knowledge of an item is learned as a single item vector and the model did not preserve the knowledge-base structure for reasoning; He et al. [25] leveraged translation-based embedding for recommendation by modeling items as entities and user”
• Large Language Models are Zero-Shot Rankers for Recommender Systems (2023, methodology): “5-turbo to (1) LLMs that have not been instruction-tuned, like LLaMA [28], and (2) LLMs that have been fine-tuned on instructions, including Vicuna [3] and text-davinci-003 2 , and then instruct these LLMs to perform the ranking task.”
• Next-item Recommendation with Sequential Hypergraphs (2020, methodology): “Pioneering works adopt Markov Chains [28] and translation-based [14] methods to model the transition between items that a user interacted with sequentially.”
• Controllable Multi-Interest Framework for Recommendation (2020, background): “TransRec [16] embeds items into a vector space where users are modeled as vectors operating on item sequences for large-scale sequential prediction.”
• Personalized Top-N Sequential Recommendation via Convolutional Sequence Embedding (2018, methodology): “th cold-start recommendation is usually treated as a separate issue in the literature [6, 7, 21, 32]. n is 5,15,10,10 for MovieLens, Gowalla, Foursquare, and Tmall. The Amazon data previously used in [5, 6] was not used due to its SI (0.0026 for ‘O˝ce Products’ category, 0.0019 for ‘Clothing, Shoes,”
• Latent Relational Metric Learning via Memory-based Attention for Collaborative Ranking (2017, methodology): “ast number of possibilities of learning weighted (linearly combined) representations. Finally, we note that another translation-based recommendation model, TransRec was recently proposed by He et al. [6] in which the authors proposed to use translations to model sequential data. While TransRec also ”
• Self-Attentive Sequential Recommendation (2018, methodology): “We adopt two common Top-N metrics, Hit Rate@10 and NDCG@10, to evaluate recommendation performance [14], [19].”