Beyond Relevance: Improving User Engagement by Personalization for Short-Video Search

Wentian Bao,Hu Liu,Kai Zheng,Chao Zhang,Shunyu Zhang,Enyun Yu,Wenwu Ou,Yang Song

2024-09-17

Abstract:Personalized search has been extensively studied in various applications, including web search, e-commerce, social networks, etc. With the soaring popularity of short-video platforms, exemplified by TikTok and Kuaishou, the question arises: can personalization elevate the realm of short-video search, and if so, which techniques hold the key? In this work, we introduce $\text{PR}^2$, a novel and comprehensive solution for personalizing short-video search, where $\text{PR}^2$ stands for the Personalized Retrieval and Ranking augmented search system. Specifically, $\text{PR}^2$ leverages query-relevant collaborative filtering and personalized dense retrieval to extract relevant and individually tailored content from a large-scale video corpus. Furthermore, it utilizes the QIN (Query-Dominate User Interest Network) ranking model, to effectively harness user long-term preferences and real-time behaviors, and efficiently learn from user various implicit feedback through a multi-task learning framework. By deploying the $\text{PR}^2$ in production system, we have achieved the most remarkable user engagement improvements in recent years: a 10.2% increase in CTR@10, a notable 20% surge in video watch time, and a 1.6% uplift of search DAU. We believe the practical insights presented in this work are valuable especially for building and improving personalized search systems for the short video platforms.

Information Retrieval

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to improve user engagement through personalization in short - video search. Specifically, the authors explore how to improve the search experience on short - video platforms through personalization techniques, thereby enhancing key metrics such as users' click - through rates, viewing time, and the number of active search users. They propose a new system named PR2, which aims to achieve this goal through personalized retrieval and ranking methods. The PR2 system not only considers the relevance between the query and the video but also combines the user's long - term interests and real - time behaviors to provide more customized search results.

### Analysis of the Core Problems in the Paper

1. **Background and Motivation**:
- The popularity of short - video applications (such as TikTok and Kuaishou) makes personalized search particularly important.
- Traditional search methods mainly rely on the relevance between the query and the document but ignore the user context, resulting in potentially unsatisfactory search results, especially when the query is ambiguous or the user's intentions are diverse.

2. **Research Questions**:
- **Can user engagement in short - video search be improved through personalization techniques?**
- **Which techniques are the most effective in personalized short - video search?**

3. **Solutions**:
- **PR2 System**: PR2 (Personalized Retrieval and Ranking augmented search system) is a comprehensive solution aimed at enhancing the user experience in short - video search through personalized retrieval and ranking methods.
- **Personalized Retrieval**:
- **Query - Related Collaborative Filtering (QRCF)**: Find videos related to the current query through the user's historical viewing records.
- **Personalized Dense Retrieval (PDR)**: Encode the user and query information into embedding vectors and use the dense retrieval method to find relevant and personalized candidate videos.
- **Personalized Ranking**:
- **QIN (Query - Dominant Interest Network)**: A novel ranking model that can effectively utilize the user's long - term and short - term behaviors and learn from various user feedbacks through a multi - task learning framework.

4. **Experimental Verification**:
- The authors conducted an A/B test on the Kuaishou platform to verify the effectiveness of the PR2 system.
- The experimental results show that the PR2 system significantly improves the user's click - through rate (CTR@10 increased by 10.2%), video viewing time (increased by 20%), and the number of daily active search users (increased by 1.6%).

### Formula Analysis

1. **Relevant Behavior Selection**:
\[
B_{\text{rel}}=\{b_i\mid \text{top}_k(\cos(E_{b_i}, E_q), K), \cos(E_{b_i}, E_q)\geq\epsilon\}
\]
- \(B_{\text{rel}}\): Set of relevant behaviors
- \(b_i\): Video watched by the user
- \(E_{b_i}\): Embedding vector of the video
- \(E_q\): Embedding vector of the query
- \(\cos(E_{b_i}, E_q)\): Cosine similarity between the video and the query
- \(\epsilon\): Relevance threshold
- \(k\): Number of selected behaviors

2. **Video - Video Similarity Calculation**:
- **Memory - Based Method**:
\[
s(i, j)=\sum_{u\in S_i\cap S_j}\frac{1}{\alpha + |I_u\cap I_v|}
\]
- \(S_i\): Search session in which the user clicks on item \(i\)
- \(I_u\): All items clicked in session \(u\)
- \(\alpha\): Smoothing coefficient
- **Embedding - Based Method**: