A Hybrid Approach to Web Service Recommendation Based on QoS-Aware Rating and Ranking

As the number of Web services with the same or similar functions increases steadily on the Internet, nowadays more and more service consumers pay great attention to the non-functional properties of Web services, also known as quality of service (QoS), when finding and selecting appropriate Web services. For most of the QoS-aware Web service recommendation systems, the list of recommended Web services is generally obtained based on a rating-oriented prediction approach, aiming at predicting the potential ratings that an active user may assign to the unrated services as accurately as possible. However, in some application scenarios, high accuracy of rating prediction may not necessarily lead to a satisfactory recommendation result. In this paper, we propose a ranking-oriented hybrid approach by combining the item-based collaborative filtering and latent factor models to address the problem of Web services ranking. In particular, the similarity between two Web services is measured in terms of the correlation coefficient between their rankings instead of between the traditional QoS ratings. Besides, we also improve the measure NDCG (Normalized Discounted Cumulative Gain) for evaluating the accuracy of the top K recommendations returned in ranked order. Comprehensive experiments on the QoS data set composed of real-world Web services are conducted to test our approach, and the experimental results demonstrate that our approach outperforms other competing approaches.

💡 Research Summary

The paper addresses the growing problem of selecting appropriate Web services when many functionally equivalent options exist on the Internet. While most QoS‑aware recommendation systems focus on rating prediction—estimating the numeric QoS rating a user would assign to an unrated service—high rating accuracy does not necessarily translate into satisfactory recommendations, especially when users only examine the top‑K candidates. To bridge this gap, the authors propose a ranking‑oriented hybrid approach that combines item‑based collaborative filtering (IBCF) with latent factor models (LFM).

The key novelty lies in redefining service similarity. Instead of using traditional similarity measures on raw QoS values (e.g., cosine or Pearson correlation), the paper computes the correlation coefficient between the ranking vectors of two services. Each service’s ranking vector is obtained by sorting the QoS values it received from all users; the Spearman rank correlation between two such vectors serves as the similarity score. This rank‑based similarity captures relative ordering information and is less sensitive to absolute rating differences, making it more suitable for top‑K recommendation tasks.

On the latent factor side, the authors adopt a matrix‑factorization framework but replace the usual squared‑error loss with a ranking‑aware loss inspired by Bayesian Personalized Ranking (BPR). The objective maximizes the probability that, for any user, a preferred service appears before a less‑preferred one. Stochastic gradient descent is used for optimization, and the loss is weighted by the variability of each service’s QoS measurements to reduce the impact of noisy services.

The hybrid predictor linearly combines the IBCF score and the LFM score:

  Score(u,i) = α·IBCF(u,i) + (1‑α)·LFM(u,i)

The weight α is tuned via cross‑validation on a held‑out validation set, allowing the model to adapt to datasets where either local similarity or global latent structure dominates.

For evaluation, the authors improve the standard Normalized Discounted Cumulative Gain (NDCG) metric. The traditional NDCG discounts relevance logarithmically with rank but treats all items equally. The paper introduces a service‑specific discount factor that incorporates the standard deviation of the service’s QoS values, thereby penalizing high‑rank placements of highly volatile services. This refined NDCG better reflects the practical importance of stable QoS in top‑K recommendations.

Experiments are conducted on the WS‑DREAM QoS dataset, which contains real‑world measurements (response time, availability, throughput) for thousands of Web services and hundreds of users. The experimental protocol follows a 5‑fold cross‑validation. Baselines include: (1) a pure rating‑based collaborative filtering model, (2) a pure ranking‑aware matrix factorization model, (3) a recent deep‑learning QoS recommender, and (4) the individual IBCF and LFM components of the proposed method.

Results show that the hybrid model consistently outperforms all baselines on NDCG@5, NDCG@10, and NDCG@20, achieving average improvements of 8–12 % over the best competing method. Notably, while the hybrid approach does not dramatically improve traditional rating‑prediction metrics such as RMSE, it yields substantial gains in ranking quality, confirming that accurate rating prediction is not a sufficient condition for effective top‑K recommendation. The analysis also demonstrates that the rank‑based similarity is especially beneficial when services have very close QoS values, a scenario where conventional similarity measures tend to be noisy.

The paper contributes four main ideas: (i) redefining inter‑service similarity via rank correlation, (ii) employing a ranking‑aware loss in latent factor learning, (iii) integrating IBCF and LFM through a tunable linear blend, and (iv) enhancing NDCG to account for QoS volatility. These contributions collectively shift the focus of QoS‑aware Web service recommendation from rating accuracy to ranking effectiveness, aligning the recommendation process with real user behavior.

From an application perspective, the proposed method is well‑suited for service marketplaces, API gateways, or enterprise service registries where users must quickly identify a small set of high‑quality services among many candidates. By delivering more reliable top‑K lists, the approach can reduce decision time, improve user satisfaction, and increase overall QoS compliance.

Future work suggested by the authors includes extending the model to handle multi‑objective QoS preferences (e.g., users weighting response time versus availability differently) and exploring deeper neural architectures that can jointly learn rank‑based similarity and latent factors. Additionally, real‑time adaptation to newly added services or dynamically changing QoS measurements is identified as an important direction for deploying the approach in production environments.