Abstract
News recommender systems are essential for helping users to efficiently and effectively find out those interesting news from a large amount of news. Most of existing news recommender systems usually learn topic-level representations of users and news for recommendation, and neglect to learn more informative aspect-level features of users and news for more accurate recommendation. As a result, they achieve limited recommendation performance. Aiming at addressing this deficiency, we propose a novel Aspect-driven News Recommender System (ANRS) built on aspect-level user preference and news representation learning. Here, news aspect is fine-grained semantic information expressed by a set of related words, which indicates specific aspects described by the news. In ANRS, news aspect-level encoder and user aspect-level encoder are devised to learn the fine-grained aspect-level representations of user's preferences and news characteristics respectively, which are fed into click predictor to judge the probability of the user clicking the candidate news. Extensive experiments are done on the commonly used real-world dataset MIND, which demonstrate the superiority of our method compared with representative and state-of-the-art methods.
Introduction
In the realm of news recommendation, understanding user interests with high precision is paramount. Historically, state-of-the-art models have relied on coarse-grained "topic" information (e.g., "Sports", "Health") to match users with articles. However, this approach suffers from three major flaws:
- Topics are too broad: An article tagged with "Auto" might be specifically about "Ford" and "Australia," which are more descriptive "aspects."
- User preferences are nuanced: A user might be interested in a specific brand like "Ford" but indifferent to the broader "Auto" category.
- Labels are expensive: Manually annotating thousands of daily articles with accurate topic labels is time-consuming and often impossible.
To address these issues, we propose using aspects—fine-grained semantic details expressed by sets of related words (e.g., "beef, pasta, tomato" instead of just "food"). These aspects can be extracted automatically from text, bypassing the need for manual labeling.
The ANRS Framework
The proposed Aspect-driven News Recommender System (ANRS) consists of three main modules designed to capture granular information:
1. News Aspect-level Encoder
This module processes news titles, abstracts, and categories using Convolutional Neural Networks (CNNs) and word-level attention. The core innovation is the aspect-level feature extractor. Inspired by autoencoders, it uses an attention mechanism to identify key aspect words. It reconstructs news embeddings via a linear combination of learnable aspect embeddings, initialized using K-means clustering of word embeddings. A custom loss function ensures these aspect embeddings remain diverse.
2. User Aspect-level Encoder
To model user preferences, this module analyzes reading history. It applies sentence-level attention over the aspect-level representations of previously browsed news, aggregating them into a single, personalized preference vector.
3. Click Predictor
The system calculates the probability of a user clicking a candidate article by taking the inner product of the user's preference representation and the candidate news's representation.
Experimental Evaluation
We evaluated ANRS on the MIND (Microsoft News Dataset), using both MIND-small (94k users) and MIND-large (1M users, 24M click logs).
Performance Comparison
ANRS was compared against several baselines, including traditional matrix factorization (BiasMF, FM) and neural models (NRMS, TANR, LSTUR). ANRS consistently outperformed all baselines. On MIND-large, ANRS achieved an AUC of 0.6826 and an MRR of 0.3350, notably higher than NRMS (AUC 0.6701) and TANR (AUC 0.6611).
Ablation Study
To verify the impact of aspect-level features, we tested a version of the model (ANRS-a) without the aspect-level feature extractors. The significant drop in performance (AUC fell to 0.6761) confirms that the aspect-driven approach is the primary driver of the system's success.
Conclusion and Future Work
Our research demonstrates that fine-grained aspect information is significantly more powerful than traditional coarse topic labels for personalized news recommendation. The unsupervised extraction of these aspects allows for high performance without the need for expensive manual labeling.
Future work will focus on incorporating the full body text of news articles, which likely contains even richer aspect information, and exploring more advanced architectures like Transformers to further refine aspect extraction.
Summary updated