Recommendation as Language Processing (RLP): A Unified Pretrain, Personalized Prompt &amp; Predict Paradigm (P5)

Geng, Shijie; Liu, Shuchang; Fu, Zuohui; Ge, Yingqiang; Zhang, Yongfeng

doi:10.1145/3523227.3546767

Cited by 176 publications

(59 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Memory Network [10], Self-Attention [25,40], GCN [5,46], Machine Reasoning [7,24,38] and Foundation Models [19] to improve the model expressiveness, so that it can better capture the abundant and complex information from user/item features and interaction histories. The key insight behind all these methods is to accurately encode the long-range histories, but this paradigm does not optimize the long-term user rewards.…”

Section: Related Work 41 Sequential Recommendation and Session-based ...mentioning

confidence: 99%

Exploration and Regularization of the Latent Action Space in Recommendation

Liu,

Cai,

Sun

et al. 2023

Preprint

View full text Add to dashboard Cite

In recommender systems, reinforcement learning solutions have effectively boosted recommendation performance because of their ability to capture long-term user-system interaction. However, the action space of the recommendation policy is a list of items, which could be extremely large with a dynamic candidate item pool. To overcome this challenge, we propose a hyper-actor and critic learning framework where the policy decomposes the item list generation process into a hyper-action inference step and an effect-action selection step. The first step maps the given state space into a vectorized hyper-action space, and the second step selects the item list based on the hyper-action. In order to regulate the discrepancy between the two action spaces, we design an alignment module along with a kernel mapping function for items to ensure inference accuracy and include a supervision module to stabilize the learning process. We build simulated environments on public datasets and empirically show that our framework is superior in recommendation compared to standard RL baselines.

show abstract

Section: Related Work 41 Sequential Recommendation and Session-based ...mentioning

confidence: 99%

Exploration and Regularization of the Latent Action Space in Recommendation

Liu,

Cai,

Sun

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…For instance, LMRec-Sys [64] utilizes prompts to reconstitute some recommendation tasks as multi-token cloze tasks, aiming to address zero-shot and data efficiency issues. P5 [19] is the first attempt to integrate different recommendation tasks within a shared conditional language generation framework (i.e., T5 [42]). To effectively convert user interaction data into text sequences, a variety of prompts are desgined to accomodate the specific characteristics of each recommendation task.…”

Section: Language Model For Recommendationmentioning

confidence: 99%

“…In numerical evaluations, we employ Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) for rating prediction. And we adopt top-k Hit Ratio (HR@k), top-k Normalized Discounted Cumulative Gain (NDCG@k) for sequential recommendation and direct recommendation which are widely used in related works [19,67]. Specifically, we report results on HR@{1,5,10}, NCGG@{5,10} for evaluation.…”

Section: Metricsmentioning

confidence: 99%

“…In this experiment, we use the titles of the items as meta information, and set 𝑛 = 10 and 𝑘 = 3 due to the limitation of a maximum context length of 4096 tokens in ChatGPT. We ramdomly sample 100 records from the test set proposed by P5 [19] for evaluation.…”

Section: Implementation Detailsmentioning

confidence: 99%

“…To effectively convert user interaction data into text sequences, a variety of prompts [64] is designed to convert user interaction data into text sequences. Furthermore, P5 [19] and M6-Rec [11] focus on building a foundation model to support a wide range of recommendation tasks.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations