PaRT: Enhancing Proactive Social Chatbots with Personalized Real-Time Retrieval Conference: Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval; July 13–18, 2025; Padua, Italy Proceedings of the 48th International ACM SIGIR Conference on Re
Zihan Niu ♡ , Zheyong Xie ♠ , Shaosheng Cao ♠ † , Chonggang Lu ♠ , Zheyu Ye ♠ Tong Xu ♡ † , Zuozhu Liu ♢ † , Yan Gao ♠ , Jia Chen ♠ , Zhe Xu ♠ , Yi Wu ♠ , Yao Hu ♠ · 2025
Read PaRT: Enhancing Proactive Social Chatbots with Personalized Real-Time Retrieval Conference: Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval; July 13–18, 2025; Padua, Italy Proceedings of the 48th International ACM SIGIR Conference on Re by Zihan Niu ♡ , Zheyong Xie ♠ , Shaosheng Cao ♠ † , Chonggang Lu ♠ , Zheyu Ye ♠ Tong Xu ♡ † , Zuozhu Liu ♢ † , Yan Gao ♠ , Jia Chen ♠ , Zhe Xu ♠ , Yi Wu ♠ , Yao Hu ♠ (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-20624.txt.
Introduction
Social chatbots wait for users to start, which cuts chats short, and Zihan Niu and co-authors give them a way to open and steer topics, in work from twenty twenty-five. Their frame first folds the user profile and chat past into a model that sharpens the user's aim. Fresh topics then grow from that aim.
Those topics guide a live fetch from a social app, and the fetched text grounds the bot's next lines. The frame ran in a live product for more than a month. Mean chat length rose by more than a fifth. The team frames the gain as proof that bots which plan topics can hold people longer than bots which just reply.
Catalog record
| CREATOR | Zihan Niu ♡ , Zheyong Xie ♠ , Shaosheng Cao ♠ † , Chonggang Lu ♠ , Zheyu Ye ♠ Tong Xu ♡ † , Zuozhu Liu ♢ † , Yan Gao ♠ , Jia Chen ♠ , Zhe Xu ♠ , Yi Wu ♠ , Yao Hu ♠ |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 20 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-20624
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
PaRT: Enhancing Proactive Social Chatbots with Personalized Real-Time Retrieval Conference: Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval; July 13–18, 2025; Padua, Italy Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’25), July 13–18, 2025, Padua, Italy DOI: 10.1145/3726302.3731946 ISBN: 979-8-4007-1592-1/2025/07 Thanks: † Corresponding authors. CCS: Information systems Personalization CCS: Information systems Language models Zihan Niu ♡ , Zheyong Xie ♠ , Shaosheng Cao ♠ † , Chonggang Lu ♠ , Zheyu Ye ♠ Tong Xu ♡ † , Zuozhu Liu ♢ † , Yan Gao ♠ , Jia Chen ♠ , Zhe Xu ♠ , Yi Wu ♠ , Yao Hu ♠ Abstract . Social chatbots have become essential intelligent companions in daily scenarios ranging from emotional support to personal interaction. However, conventional chatbots with passive response mechanisms usually rely on users to initiate or sustain dialogues by bringing up new topics, resulting in diminished engagement and shortened dialogue duration. In this paper, we present PaRT, a novel framework enabling context-aware proactive dialogues for social chatbots through personalized real-time retrieval and generation. Specifically, PaRT first integrates user profiles and dialogue context into a large language model (LLM), which is initially prompted to refine user queries and recognize their underlying intents for the upcoming conversation. Guided by refined intents, the LLM generates personalized dialogue topics, which then serve as targeted queries to retrieve relevant passages from RedNote. Finally, we prompt LLMs with summarized passages to generate knowledge-grounded and engagement-optimized responses. Our approach has been running stably in a real-world production environment for more than 30 days, achieving a 21.77% improvement in the average duration of dialogues. 1. Introduction Recent advancements in Large Language Models (LLMs) ( Achiam et al. 2023 ; Bai et al. 2023 ; Touvron et al. 2023 ) have significantly boosted the development of social chatbots ( Zhou et al. 2020 ; Adiwardana et al. 2020 ) , enabling them to exhibit increasingly intelligent and human-like behavior, gathering widespread attention. Despite this progress, current social chatbots primarily emphasize providing comprehensive and emotional responses to user requests ( Ruotsalo et al. 2023 ; Wang et al. 2024a ) , neglecting the importance of actively engaging with users in conversation ( Deng et al. 2023 ; Deng et al. 2024 ; Liao et al. 2023 ) . For instance, traditional chatbots relying solely on passive dialogue strategies ( Deng et al. 2023 ; Liao et al. 2023 ) often limit the depth and natural extension of conversation, requiring users to continually initiate and drive the conversation, resulting in decreased user engagement and shortened interaction duration. Figure 1. The interface of PaRT. From left to right, we sequentially present the greeting interface, the dialogue page, and the user profile interface. The greeting scenario is initiated by chatbot at the beginning, while the dialogue scenario aims to proactively guide ongoing conversation. Figure 2. The figure provides an overview of PaRT. It illustrates the different dialogue experiences based on traditional chatbot (left) and our PaRT method (right). Proactive dialogues have emerged as a promising solution to enhance user engagement in conversational AI. Based on established definitions of proactivity ( Grant and Ashford 2008 ; Deng et al. 2023 ) , these systems actively initiate or switch topics during interactions rather than maintaining passive responsiveness to users. For instance, they may initiate conversations with domain-specific prompts (e.g., "Which tourist attractions interest you most?") or dynamically shift topics when detecting waning user interest ( Zhang et al. 2023 ; Hu et al. 2022 ) . Current implementations of proactive dialogues often directly prompt LLMs for topic generation, resulting in two critical limitations: 1) Overgeneralized content. Direct prompting usually produces generic topics misaligned with user preferences, degrading interaction quality; 2) Knowledge boundaries: LLMs’ inherent limitation in knowledge constrains their ability to sustain domain-specific and up-to-date dialogues with real-time contextual information ( Vakili Tahami et al. 2020 ; Zhang et al. 2024a ) . To address these challenges, we propose PaRT , which enhances P ro a ctive social chatbots with personalized real-time R e T rieval. Our proposed PaRT adopts a unified framework that addresses two essential proactive conversation scenarios, as illustrated in Figure 1 . Both scenarios are supported by three key components: user profiling, an intent-guided query refiner, and retrieval-augmented generation. In particular, the user profiling module lays the foundation by constructing a detailed user profile through proactive questioning and memory extraction ( Zhong et al. 2024 ; Lu et al. 2023 ; Deng et al. 2025 ) . Subsequently, the intent-guided query refiner analyzes the dialogue context to identify user intent, which includes natural transitions, explicit retrieval (direct information requests), and implicit retrieval (signals for topic change). Based on the identified intent and user profile ( Zhang et al. 2024b ; Wang et al. 2022 ) , it then refines the current query into a personalized one. This refined query serves as a concrete representation of the topic for the upcoming conversation. Leveraging the tailored queries, the retrieval-augmented generation module performs retrieval from RedNote 1 1 1 https://www.xiaohongshu.com , and summarizes these retrieved passages to mitigate interference from irrelevant information ( Lyu et al. 2024 ; Xu et al. 2023 ) . Then we prompt LLMs with summarized text to generate responses that are well-aligned with the personalized query ( Lewis et al. 2020 ; Jiang et al. 2023 ) . Both offline and online experimental results demonstrate that PaRT significantly enhances generation quality and improves user engagement. In summary, our contributions are as follows: (1) To the best of our knowledge, this is the first work proposing the enhancement of proactive social chatbots with personalized real-time retrieval, which significantly improves generation quality and effectively enhances user engagement by initiating and guiding conversation topics. (2) Our method achieves an effective proactive dialogue experience with personalized real-time retrieval by leveraging user profiling, intent-guided query refiner and retrieval-augmented generation. (3) The proposed PaRT has been successfully deployed in a real-world production environment for over 30 days, achieving remarkable results by increasing the average dialogue duration by 21.77%. 2. Details of the Proposed PaRT The architecture of PaRT is illustrated in Figure 2 . In this section, we present a detailed explanation of PaRT, including its overall workflow and three core modules: user profiling, intent-guided query refiner, and retrieval-augmented generation. 2.1. User Profiling User profiling module is first introduced to enhance chatbots’ understanding of user preferences By introducing memory mechanisms ( Zhong et al. 2024 ; Lu et al. 2023 ) , critical information from dialogue history is summarized and stored in user profiles. Additionally, proactive greetings serve as an effective method for profile modeling. We have developed a comprehensive static question bank from which the chatbot can randomly select questions to initiate conversations. Furthermore, personalized greetings based on existing user profiles and retrieval-augmented generation (detailed in Section 2.4) facilitate the development of more comprehensive user profiles. As illustrated in Figure 2 , conversations begin with a greeting, while user preferences are continuously updated to maintain real-time user profiles. 2.2. Intent-guided Query Refiner As conversations evolve, effective chatbots should proactively address users’ needs or pivot topics when engagement wanes. Figure