Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ClickPrompt: CTR Models are Strong Prompt Generators for Adapting Language Models to CTR Prediction

Jianghao Lin, Bo Chen, Hangyu Wang, Yunjia Xi, Yanru Qu, Xinyi Dai, Kangning Zhang, Ruiming Tang, Yong Yu, Weinan Zhang

TL;DR

This paper designs a prompt-augmented masked language modeling (PA-MLM) pretraining task, where PLM has to recover the masked tokens based on the language context, as well as the soft prompts generated by CTR model, to incorporate CTR models to generate interaction-aware soft prompts for PLMs.

Abstract

Click-through rate (CTR) prediction has become increasingly indispensable for various Internet applications. Traditional CTR models convert the multi-field categorical data into ID features via one-hot encoding, and extract the collaborative signals among features. Such a paradigm suffers from the problem of semantic information loss. Another line of research explores the potential of pretrained language models (PLMs) for CTR prediction by converting input data into textual sentences through hard prompt templates. Although semantic signals are preserved, they generally fail to capture the collaborative information (e.g., feature interactions, pure ID features), not to mention the unacceptable inference overhead brought by the huge model size. In this paper, we aim to model both the semantic knowledge and collaborative knowledge for accurate CTR estimation, and meanwhile address the inference inefficiency issue. To benefit from both worlds and close their gaps, we propose a novel model-agnostic framework (i.e., ClickPrompt), where we incorporate CTR models to generate interaction-aware soft prompts for PLMs. We design a prompt-augmented masked language modeling (PA-MLM) pretraining task, where PLM has to recover the masked tokens based on the language context, as well as the soft prompts generated by CTR model. The collaborative and semantic knowledge from ID and textual features would be explicitly aligned and interacted via the prompt interface. Then, we can either tune the CTR model with PLM for superior performance, or solely tune the CTR model without PLM for inference efficiency. Experiments on four real-world datasets validate the effectiveness of ClickPrompt compared with existing baselines.

ClickPrompt: CTR Models are Strong Prompt Generators for Adapting Language Models to CTR Prediction

TL;DR

This paper designs a prompt-augmented masked language modeling (PA-MLM) pretraining task, where PLM has to recover the masked tokens based on the language context, as well as the soft prompts generated by CTR model, to incorporate CTR models to generate interaction-aware soft prompts for PLMs.

Abstract

Click-through rate (CTR) prediction has become increasingly indispensable for various Internet applications. Traditional CTR models convert the multi-field categorical data into ID features via one-hot encoding, and extract the collaborative signals among features. Such a paradigm suffers from the problem of semantic information loss. Another line of research explores the potential of pretrained language models (PLMs) for CTR prediction by converting input data into textual sentences through hard prompt templates. Although semantic signals are preserved, they generally fail to capture the collaborative information (e.g., feature interactions, pure ID features), not to mention the unacceptable inference overhead brought by the huge model size. In this paper, we aim to model both the semantic knowledge and collaborative knowledge for accurate CTR estimation, and meanwhile address the inference inefficiency issue. To benefit from both worlds and close their gaps, we propose a novel model-agnostic framework (i.e., ClickPrompt), where we incorporate CTR models to generate interaction-aware soft prompts for PLMs. We design a prompt-augmented masked language modeling (PA-MLM) pretraining task, where PLM has to recover the masked tokens based on the language context, as well as the soft prompts generated by CTR model. The collaborative and semantic knowledge from ID and textual features would be explicitly aligned and interacted via the prompt interface. Then, we can either tune the CTR model with PLM for superior performance, or solely tune the CTR model without PLM for inference efficiency. Experiments on four real-world datasets validate the effectiveness of ClickPrompt compared with existing baselines.
Paper Structure (38 sections, 11 equations, 3 figures, 9 tables)

This paper contains 38 sections, 11 equations, 3 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Traditional CTR Prediction
  4. PLM-based CTR Prediction
  5. Methodology
  6. Overview of ClickPrompt
  7. Modality Transformation
  8. Prompt Generation
  9. Prompt Fusion
  10. Learning Strategy
  11. Prompt-augmented Masked Language Modeling
  12. Finetuning with PLM
  13. Finetuning without PLM
  14. Experiment
  15. Experiment Setups
  16. ...and 23 more sections

Figures (3)

  • Figure 1: The information flow of the collaborative and semantic knowledge in our proposed ClickPrompt framework. With the soft prompts serving as the bridges, the ID-based collaborative information is transmitted to the language model by forward propagation, and the text-based semantic information flows back into the CTR model via backpropagation.
  • Figure 2: The illustration of the model architecture and learning strategy for our proposed ClickPrompt framework.
  • Figure 3: The illustration of prompt strategy variants. We can either only insert the prompts at the first input layer (left), or perform layerwise soft prompts for PLM (right). Note that ClickPrompt adopts the latter prompt strategy.