What Makes an Ideal Quote? Recommending "Unexpected yet Rational" Quotations via Novelty

TL;DR

NovelQR, a novelty-driven quotation recommendation framework that formalizes quote recommendation as choosing contextually novel but semantically coherent quotations, operationalizes with NovelQR, a generative label agent that interprets each quotation and its surrounding context into multi-dimensional deep-meaning labels, enabling label-enhanced retrieval.

Abstract

Quotation recommendation aims to enrich writing by suggesting quotes that complement a given context, yet existing systems mostly optimize surface-level topical relevance and ignore the deeper semantic and aesthetic properties that make quotations memorable. We start from two empirical observations. First, a systematic user study shows that people consistently prefer quotations that are ``unexpected yet rational'' in context, identifying novelty as a key desideratum. Second, we find that strong existing models struggle to fully understand the deep meanings of quotations. Inspired by defamiliarization theory, we therefore formalize quote recommendation as choosing contextually novel but semantically coherent quotations. We operationalize this objective with NovelQR, a novelty-driven quotation recommendation framework. A generative label agent first interprets each quotation and its surrounding context into multi-dimensional deep-meaning labels, enabling label-enhanced retrieval. A token-level novelty estimator then reranks candidates while mitigating auto-regressive continuation bias. Experiments on bilingual datasets spanning diverse real-world domains show that our system recommends quotations that human judges rate as more appropriate, more novel, and more engaging than other baselines, while matching or surpassing existing methods in novelty estimation.

Figures (16)

• Figure 1: An ideal quote should not only fit the context, but also be novel, adding aesthetic value to writing. As shown in the third example, the best quote often feels unexpected at first, but makes perfect sense in context.
• Figure 2: Empirical result. (a) The evaluation results of the only-quote (left) and enhanced-quote (right) scene. All models perform significantly better with enhanced inputs, demonstrating the effectiveness of guided prompt in deep meaning understanding. (b) In user studies, (left) they perceive ideal quotations as "unexpected yet rational"(), while current models tend to produce clichéd-but-high-fit ones (); (right) across various writing scenarios, novelty consistently emerges as a key dimension of quotation quality.
• Figure 3: Overview of our novelty-driven quotation recommendation framework: (1) Label Enhancement, where the generative label agent enhances understanding of the quotation knowledge base (KB) and user-given context; (2) Rationality Retrieval, which "retrieves then filters" quotations using deep meanings and labels; and (3) Novelty Reranking, which highlights the continuation bias, and introduces the method to mitigate it and estimate novelty.
• Figure 4: Semantic embedding visualization (T-SNE) of retrieved quotations using different methods. Label-enhanced shows tighter clustering and better semantic consistency compared to Quote-based retrieval.
• Figure 5: The Correlation between our LLM-as-judge evaluation and human scores. To avoid overlapping points, random jitters were added to ratings.