DAIAN: Deep Adaptive Intent-Aware Network for CTR Prediction in Trigger-Induced Recommendation

TL;DR

The Deep Adaptive Intent-Aware Network (DAIAN) is proposed, that dynamically adapts to users' intent preferences and reinforces similarity by leveraging a hybrid enhancer with ID and semantic information, followed by adaptive selection based on varying intents.

Abstract

Recommendation systems are essential for personalizing e-commerce shopping experiences. Among these, Trigger-Induced Recommendation (TIR) has emerged as a key scenario, which utilizes a trigger item (explicitly represents a user's instantaneous interest), enabling precise, real-time recommendations. Although several trigger-based techniques have been proposed, most of them struggle to address the intent myopia issue, that is, a recommendation system overemphasizes the role of trigger items and narrowly focuses on suggesting commodities that are highly relevant to trigger items. Meanwhile, existing methods rely on collaborative behavior patterns between trigger and recommended items to identify the user's preferences, yet the sparsity of ID-based interaction restricts their effectiveness. To this end, we propose the Deep Adaptive Intent-Aware Network (DAIAN) that dynamically adapts to users' intent preferences. In general, we first extract the users' personalized intent representations by analyzing the correlation between a user's click and the trigger item, and accordingly retrieve the user's related historical behaviors to mine the user's diverse intent. Besides, sparse collaborative behaviors constrain the performance in capturing items associated with user intent. Hence, we reinforce similarity by leveraging a hybrid enhancer with ID and semantic information, followed by adaptive selection based on varying intents. Experimental results on public datasets and our industrial e-commerce datasets demonstrate the effectiveness of DAIAN.

Figures (7)

• Figure 1: Two primary recommendation scenarios on the Xianyu platform. The photo highlighted by a red solid frame represents User-Induced Recommendation. When users click on a trigger item and scroll down, they can transition to Trigger-Induced Recommendation.
• Figure 2: Differences in user intent preference modeling between existing TIR solutions and DAIAN. Black dashed boxes denote strongly-relevant items, purple dashed boxes represent relevant items, and orange dashed boxes highlight irrelevant items.
• Figure 3: An overall architecture of DAIAN can be broadly divided into three modules: (1) User Intent Modeling Module generates users’ personalized intent representations. (2) Diverse Intent Extraction Module retrieves the user’s historical behaviors to extract users' diverse intents. (3) Similarity-Enhanced Intent Network reinforces similarity and performs adaptive selection based on users' diverse intents.
• Figure 4: An Overview of our Three-Stage Training Strategy: the pre-training of UIM and DIE, followed by the integration of pre-trained users' personalized intent and diverse intent representations into recommendation model.
• Figure 5: (a) exhibits the Predicted vs. Ground-Truth intent distribution for specific-intent users, and (b) exhibits the Predicted vs. Ground-Truth intent distribution for broad-intent users.