Enhancing CTR Prediction in Recommendation Domain with Search Query Representation

TL;DR

This paper proposes a framework that learns from user search query embeddings within the context of user preferences in the recommendation domain, and outperforms state-of-the-art models in the recommendation domain.

Abstract

Many platforms, such as e-commerce websites, offer both search and recommendation services simultaneously to better meet users' diverse needs. Recommendation services suggest items based on user preferences, while search services allow users to search for items before providing recommendations. Since users and items are often shared between the search and recommendation domains, there is a valuable opportunity to enhance the recommendation domain by leveraging user preferences extracted from the search domain. Existing approaches either overlook the shift in user intention between these domains or fail to capture the significant impact of learning from users' search queries on understanding their interests. In this paper, we propose a framework that learns from user search query embeddings within the context of user preferences in the recommendation domain. Specifically, user search query sequences from the search domain are used to predict the items users will click at the next time point in the recommendation domain. Additionally, the relationship between queries and items is explored through contrastive learning. To address issues of data sparsity, the diffusion model is incorporated to infer positive items the user will select after searching with certain queries in a denoising manner, which is particularly effective in preventing false positives. Effectively extracting this information, the queries are integrated into click-through rate prediction in the recommendation domain. Experimental analysis demonstrates that our model outperforms state-of-the-art models in the recommendation domain.

Figures (5)

• Figure 1: The overall framework of our proposed QueryRec model. Query representation from the search domain is fully learned by next-item prediction and diffusion-augmented contrastive learning modules. This refined representation is combined with other features from the recommendation domain and used in a CTR prediction model. For the next item prediction, the query sequences in chronological order up to time point $t$ are encoded with the self-attention layer and pointwise feed-forward network (FFN), and then predict the item the user will click in the recommendation domain at time point $t+1$. In diffusion-augmented contrastive learning, the diffusion model is first trained with records with items clicked after searching with certain queries. $x_0$ are probabilities of items that the users would click after a certain query. Considering the change the users could still be interested in the items that have not been exposed, a value from $[-1, 1]$ is also assigned to non-exposed items. After the reverse process of the diffusion model, we get $\Tilde{x}_0$, which is the original item clicked history of in the search domain. The trained diffusion model is used to learn the item distributions for queries without corresponding clicked items. The top-K items are treated as positive samples for contrastive learning, which aims to capture query-item relationships.
• Figure 2: Histogram of JS Divergence Values of Selected User Interests.
• Figure 3: Statistic analysis on the performance of contrastive learning (CL) with Mask v.s. CL with diffusion-augmented positive samples.
• Figure 4: Statistic analysis on the impact of diffusion initialization values.
• Figure 5: Data statistics. The comparison of queries with corresponding item clicks to queries with items clicked. Left: KuaiSAR-small; Right: KuaiSAR-large.