RECORD
xMTF: A Formula-Free Model for Reinforcement-Learning-Based Multi-Task Fusion in Recommender Systems
Information Retrieval · Submitted 8 Apr 2025
abstract compiled · 2026-09-23 · no language-model review
The authors' main claim By expanding the MTF search space, xMTF outperforms existing methods in extensive offline and online experiments.
The compiled record
The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
Open the compiled record
The abstract, sentence by sentence
Recommender systems need to optimize various types of user feedback, e.g., clicks, likes, and shares. A typical recommender system handling multiple types of feedback has two components: a multi-task learning (MTL) module, predicting feedback such as click-through rate and like rate; and a multi-task fusion (MTF) module, integrating these predictions into a single score for item ranking. MTF is essential for ensuring user satisfaction, as it directly influences recommendation outcomes. Recently, reinforcement learning (RL) has been applied to MTF tasks to improve long-term user satisfaction. However, existing RL-based MTF methods are formula-based methods, which only adjust limited coefficients within pre-defined formulas. The pre-defined formulas restrict the RL search space and become a bottleneck for MTF. To overcome this, we propose a formula-free MTF framework. We demonstrate that any suitable fusion function can be expressed as a composition of single-variable monotonic functions, as per the Sprecher Representation Theorem. Leveraging this, we introduce a novel learnable monotonic fusion cell (MFC) to replace pre-defined formulas. We call this new MFC-based model eXtreme MTF (xMTF). Furthermore, we employ a two-stage hybrid (TSH) learning strategy to train xMTF effectively. By expanding the MTF search space, xMTF outperforms existing methods in extensive offline and online experiments.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | Recommender systems need to optimize various types of user feedback, e.g., clicks, likes, and shares. | background 1.00 | 0.05 | 0.01 | asserted 1.00 |
| s1 | A typical recommender system handling multiple types of feedback has two components: a multi-task learning (MTL) module, predicting feedback such as click-through rate and like rate; and a multi-task fusion (MTF) module, integrating these predictions into a single score for item ranking. | background 1.00 | 0.06 | 0.02 | asserted 1.00 |
| s2 | MTF is essential for ensuring user satisfaction, as it directly influences recommendation outcomes. | background 1.00 | 0.06 | 0.03 | asserted 1.00 |
| s3 | Recently, reinforcement learning (RL) has been applied to MTF tasks to improve long-term user satisfaction. | background 1.00 | 0.04 | 0.02 | asserted 1.00 |
| s4 | However, existing RL-based MTF methods are formula-based methods, which only adjust limited coefficients within pre-defined formulas. | gap 0.60 | 0.05 | 0.03 | asserted 1.00 |
| s5 | The pre-defined formulas restrict the RL search space and become a bottleneck for MTF. | gap 0.70 | 0.13 | 0.03 | asserted 1.00 |
| s6 | To overcome this, we propose a formula-free MTF framework. | contribution 0.99 | 0.75 | 0.07 | asserted 0.92 |
| s7 | We demonstrate that any suitable fusion function can be expressed as a composition of single-variable monotonic functions, as per the Sprecher Representation Theorem. | method 0.54 | 0.36 | 0.08 | asserted 1.00 |
| s8 | Leveraging this, we introduce a novel learnable monotonic fusion cell (MFC) to replace pre-defined formulas. | method 0.68 | 0.90 | 0.08 | asserted 0.99 |
| s9 | We call this new MFC-based model eXtreme MTF (xMTF). | contribution 0.65 | 0.86 | 0.05 | asserted 0.96 |
| s10 | Furthermore, we employ a two-stage hybrid (TSH) learning strategy to train xMTF effectively. | method 0.94 | 0.92 | 0.04 | asserted 1.00 |
| s11 | By expanding the MTF search space, xMTF outperforms existing methods in extensive offline and online experiments. | result 1.00 | 0.94 | 0.85 | asserted 1.00 |
What was examined, and what was not
- arXiv listing abstract v1
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
The paper
Original abstract
Recommender systems need to optimize various types of user feedback, e.g., clicks, likes, and shares. A typical recommender system handling multiple types of feedback has two components: a multi-task learning (MTL) module, predicting feedback such as click-through rate and like rate; and a multi-task fusion (MTF) module, integrating these predictions into a single score for item ranking. MTF is essential for ensuring user satisfaction, as it directly influences recommendation outcomes. Recently, reinforcement learning (RL) has been applied to MTF tasks to improve long-term user satisfaction. However, existing RL-based MTF methods are formula-based methods, which only adjust limited coefficients within pre-defined formulas. The pre-defined formulas restrict the RL search space and become a bottleneck for MTF. To overcome this, we propose a formula-free MTF framework. We demonstrate that any suitable fusion function can be expressed as a composition of single-variable monotonic functions, as per the Sprecher Representation Theorem. Leveraging this, we introduce a novel learnable monotonic fusion cell (MFC) to replace pre-defined formulas. We call this new MFC-based model eXtreme MTF (xMTF). Furthermore, we employ a two-stage hybrid (TSH) learning strategy to train xMTF effectively. By expanding the MTF search space, xMTF outperforms existing methods in extensive offline and online experiments.
Human review
How the literature uses it
What this paper does with the arXiv papers it cites
What it cites · 4 sentences without a confident label
Uses the method or tool of 1 paper
-
Specifically, we define a pairwise loss, i.e., the BPR loss [ref] , to ensure the consistency of the ranking order between $z̃_i^I$ and $z̃_i$:
BPR: Bayesian Personalized Ranking from Implicit Feedback · Inner Stage · 1.00
Cites as background 2 papers
-
Recently, RL-based approaches have been applied to MTF [ref] .
Value-aware Recommendation based on Reinforced Profit Maximization in E-commerce Systems · RL Modeling of MTF · 0.93
-
Recently, RL-based approaches have been applied to MTF [ref] .
2404.17589 · RL Modeling of MTF · 0.93
Discussion
Know something about this paper? Leave a comment, or highlight any passage above to comment on it.