A Unified Multi-task Learning Framework for Multi-goal Conversational Recommender Systems

Recent years witnessed several advances in developing multi-goal conversational recommender systems (MG-CRS) that can proactively attract users’ interests and naturally lead user-engaged dialogues with multiple conversational goals and diverse topics. Four tasks are often involved in MG-CRS, including Goal Planning, Topic Prediction, Item Recommendation, and Response Generation. Most existing studies address only some of these tasks. To handle the whole problem of MG-CRS, modularized frameworks are adopted where each task is tackled independently without considering their interdependencies. In this work, we propose a novel Unified MultI-goal conversational recommeNDer system, namely UniMIND. In specific, we unify these four tasks with different formulations into the same sequence-to-sequence (Seq2Seq) paradigm. Prompt-based learning strategies are investigated to endow the unified model with the capability of multi-task learning. Finally, the overall learning and inference procedure consists of three stages, including multi-task learning, prompt-based tuning, and inference. Experimental results on two MG-CRS benchmarks (DuRecDial and TG-ReDial) show that UniMIND achieves state-of-the-art performance on all tasks with a unified model.

Introduction. Conversational Recommender Systems (CRS) aim to make recommendations by learning users’ preferences through interactive conversations [22, 26, 55]. CRS has become one of the trending research topics for recommender systems and is gaining increasing attention, due to its natural advantage of explicitly acquiring users’ real-time preferences and providing a user-engaged recommendation procedure. Based on different scenarios, various CRS have been proposed, either from the perspective of recommender systems, being an enhanced interactive recommender system [7, 22, 44], or from the perspective of dialogue systems, being a variation of goal-oriented conversational systems [18, 23, 26]. Most of these CRS assume that users always know what they want and the system passively and solely targets at making the successful recommendation on users’ desired items.

Discussion / Conclusion. 7.4 Error Analysis and Limitations Despite the effectiveness of the proposed UniMIND framework for MG-CRS, we would like to better understand the failure modes of UniMIND for further improvement in future studies. After analyzing those cases with low human evaluation scores, we identify the following limitations and discuss the potential solutions: • Low Recommendation Success Rate. All the baselines and UniMIND fail to reach a promising recommendation performance on TG-ReDial as shown in Table 8, due to the sparsity of the user-item interactions. Since the historical interaction data is not utilized in UniMIND, one possible direction is to study how to incorporate this kind of data into the Seq2Seq framework for improving the recommendation performance. • Informativeness. As shown in Table 4, there is still a gap between the generated and the ground-truth response on Informativeness. In order to diversify and enrich the information 8 CONCLUSIONS In this work, we propose a novel unified multi-task learning framework for multi-goal conversational recommender systems, namely UniMIND.