Rating Matrices Research Articles

TEE-Graph: efficient privacy and ownership protection for cloud-based graph spectral analysis.

Big graphs like social network user interactions and customer rating matrices require significant computing resources to maintain. Data owners are now using public cloud resources for storage and computing elasticity. However, existing solutions do not fully address the privacy and ownership protection needs of the key involved parties: data contributors and the data owner who collects data from contributors. We propose a Trusted Execution Environment (TEE) based solution: TEE-Graph for graph spectral analysis of outsourced graphs in the cloud. TEEs are new CPU features that can enable much more efficient confidential computing solutions than traditional software-based cryptographic ones. Our approach has several unique contributions compared to existing confidential graph analysis approaches. (1) It utilizes the unique TEE properties to ensure contributors' new privacy needs, e.g., the right of revocation for shared data. (2) It implements efficient access-pattern protection with a differentially private data encoding method. And (3) it implements TEE-based special analysis algorithms: the Lanczos method and the Nystrom method for efficiently handling big graphs and protecting confidentiality from compromised cloud providers. The TEE-Graph approach is much more efficient than software crypto approaches and also immune to access-pattern-based attacks. Compared with the best-known software crypto approach for graph spectral analysis, PrivateGraph, we have seen that TEE-Graph has 103-105 times lower computation, storage, and communication costs. Furthermore, the proposed access-pattern protection method incurs only about 10%-25% of the overall computation cost. Our experimentation showed that TEE-Graph performs significantly better and has lower costs than typical software approaches. It also addresses the unique ownership and access-pattern issues that other TEE-related graph analytics approaches have not sufficiently studied. The proposed approach can be extended to other graph analytics problems with strong ownership and access-pattern protection.

Community-Based Matrix Factorization (CBMF) Approach for Enhancing Quality of Recommendations.

Matrix factorization is a long-established method employed for analyzing and extracting valuable insight recommendations from complex networks containing user ratings. The execution time and computational resources demanded by these algorithms pose limitations when confronted with large datasets. Community detection algorithms play a crucial role in identifying groups and communities within intricate networks. To overcome the challenge of extensive computing resources with matrix factorization techniques, we present a novel framework that utilizes the inherent community information of the rating network. Our proposed approach, named Community-Based Matrix Factorization (CBMF), has the following steps: (1) Model the rating network as a complex bipartite network. (2) Divide the network into communities. (3) Extract the rating matrices pertaining only to those communities and apply MF on these matrices in parallel. (4) Merge the predicted rating matrices belonging to communities and evaluate the root mean square error (RMSE). In our experimentation, we use basic MF, SVD++, and FANMF for matrix factorization, and the Louvain algorithm is used for community division. The experimental evaluation on six datasets shows that the proposed CBMF enhances the quality of recommendations in each case. In the MovieLens 100K dataset, RMSE has been reduced to 0.21 from 1.26 using SVD++ by dividing the network into 25 communities. A similar reduction in RMSE is observed for the datasets of FilmTrust, Jester, Wikilens, Good Books, and Cell Phone.

Bidirectional Trust-Enhanced Collaborative Filtering for Point-of-Interest Recommendation.

A personalized point-of-interest (POI) recommender system is of great significance to facilitate the daily life of users. However, it suffers from some challenges, such as trustworthiness and data sparsity problems. Existing models only consider the trust user influence and ignore the role of the trust location. Furthermore, they fail to refine the influence of context factors and fusion between the user preference and context models. To address the trustworthiness problem, we propose a novel bidirectional trust-enhanced collaborative filtering model, which investigates the trust filtering from the views of users and locations. To tackle the data sparsity problem, we introduce temporal factor into the trust filtering of users as well as geographical and textual content factors into the trust filtering of locations. To further alleviate the sparsity of user-POI rating matrices, we employ a weighted matrix factorization fused with the POI category factor to learn the user preference. To integrate the trust filtering models and the user preference model, we develop a fused framework with two kinds of integrating methods in relation to the different impacts of factors on the POIs that users have visited and the POIs that users have not visited. Finally, we conduct extensive experiments on Gowalla and Foursquare datasets to evaluate our proposed POI recommendation model, and the results show that our proposed model improves by 13.87% at precision@5 and 10.36% at recall@5 over the state-of-the-art model, which demonstrates that our proposed model outperforms the state-of-the-art method.

Closed-Form Models of Accuracy Loss due to Subsampling in SVD Collaborative Filtering

We postulate and analyze a nonlinear subsampling accuracy loss (SSAL) model based on the root mean square error (RMSE) and two SSAL models based on the mean square error (MSE), suggested by extensive preliminary simulations. The SSAL models predict accuracy loss in terms of subsampling parameters like the fraction of users dropped (FUD) and the fraction of items dropped (FID). We seek to investigate whether the models depend on the characteristics of the dataset in a constant way across datasets when using the SVD collaborative filtering (CF) algorithm. The dataset characteristics considered include various densities of the rating matrix and the numbers of users and items. Extensive simulations and rigorous regression analysis led to empirical symmetrical SSAL models in terms of FID and FUD whose coefficients depend only on the data characteristics. The SSAL models came out to be multi-linear in terms of odds ratios of dropping a user (or an item) vs. not dropping it. Moreover, one MSE deterioration model turned out to be linear in the FID and FUD odds where their interaction term has a zero coefficient. Most importantly, the models are constant in the sense that they are written in closed-form using the considered data characteristics (densities and numbers of users and items). The models are validated through extensive simulations based on 850 synthetically generated primary (pre-subsampling) matrices derived from the 25M MovieLens dataset. Nearly 460 000 subsampled rating matrices were then simulated and subjected to the singular value decomposition (SVD) CF algorithm. Further validation was conducted using the 1M MovieLens and the Yahoo! Music Rating datasets. The models were constant and significant across all 3 datasets.

Recommendation System on Travel Destination based on Geotagged Data

Tourism research has benefitted from the worldwide spread and development of social networking services. People nowadays are more likely to rely on internet resources to plan their vacations. Thus, travel recommendation systems are designed to sift through the mammoth amount of data and identify the ideal travel destinations for the users. Moreover, it is shown that the increasing availability and popularity of geotagged data significantly impacts the destination decision. However, most current research concentrates on reviews and textual information to develop the recommendation model. Therefore, the proposed travel recommendation model examines the collective behaviour and connections between users based on geotagged data to provide personalized suggestions for individuals. The model was developed using the user-based collaborative filtering technique. The matrix factorization model was selected as the collaborative filtering technique to compute user similarities due to its adaptability in dealing with sparse rating matrices. The recommendation model generates prediction values to recommend the most appropriate locations. Finally, the model performance of the proposed model was assessed against the popularity and random models using the test design established using Mean Average Precision (MAP), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). The findings indicated that the proposed matrix factorization model has an average MAP of 0.83, with RMSE and MAE values being 1.36 and 1.24, respectively. The proposed model got significantly higher MAP values and the lowest RMSE and MAE values compared to the two baseline models. The comparison shows that the proposed model is effective in providing personalized suggestions to users based on their past visits.

Deep transfer learning with multimodal embedding to tackle cold-start and sparsity issues in recommendation system.

Recommender systems (RSs) have become increasingly vital in the modern information era and connected economy. They play a key role in business operations by generating personalized suggestions and minimizing information overload. However, the performance of traditional RSs is limited by data sparseness and cold-start issues. Though deep learning-based recommender systems (DLRSs) are very popular, they underperform when considering rating matrices with sparse entries. Despite their performance improvements, DLRSs also suffer from data sparsity, cold start, serendipity, and generalizability issues. We propose a multistage model that uses multimodal data embedding and deep transfer learning for effective and personalized product recommendations, and is designed to overcome data sparsity and cold-start issues. The proposed model includes two phases. In the first—offline—phase, a deep learning technique is implemented to learn hidden features from a large image dataset (targeting new item cold start), and a multimodal data embedding is used to produce dense user feature and item feature vectors (targeting user cold start). This phase produces three different similarity matrices that are used as inputs for the second—online—phase to generate a list of top-n relevant items for a target user. We analyzed the accuracy and effectiveness of the proposed model against the existing baseline RSs using a Brazilian E-commerce dataset. The results show that our model scored 0.5882 for MAE and 0.4011 for RMSE which is lower than baseline RSs which indicates that the model achieved an improved accuracy and was able to minimize the typical cold start and data sparseness issues during the recommendation process.

UI Design and Optimization Method for Museum Display Based on User Behavior Recommendation

In view of the lack of rich display methods in the display design of museums, it is impossible to enhance the interest of visitors. This paper proposes a museum object recommendation method based on collaborative filtering, which simplifies the display design, improves the recommendation effect, and alleviates the scalability problem. Firstly, the algorithm of recommendation system combines the advantages of memory collaborative filtering and uses smoothing processing to improve the efficiency of recommendation and achieve the best consistency. Then, the cross-domain collaborative filtering rating matrix generation model is used to establish the correlation between multiple rating matrices by finding the shared hidden clustering rating matrix, which also improves the recommendation effect. Finally, the conclusion shows that we can use single user behavior data such as forgetting mechanism to recommend to users. SVD makes full use of the interaction data of various behaviors, and NMF algorithm makes full use of the data of various user behaviors, which can effectively solve the existing problems. The stochastic gradient descent is applied to the SVD algorithm to accelerate the convergence speed of the model, improve the performance of the model, and effectively improve the accuracy of score prediction.

On Producing Accurate Rating Predictions in Sparse Collaborative Filtering Datasets

The typical goal of a collaborative filtering algorithm is the minimisation of the deviation between rating predictions and factual user ratings so that the recommender system offers suggestions for appropriate items, achieving a higher prediction value. The datasets on which collaborative filtering algorithms are applied vary in terms of sparsity, i.e., regarding the percentage of empty cells in the user–item rating matrices. Sparsity is an important factor affecting rating prediction accuracy, since research has proven that collaborative filtering over sparse datasets exhibits a lower accuracy. The present work aims to explore, in a broader context, the factors related to rating prediction accuracy in sparse collaborative filtering datasets, indicating that recommending the items that simply achieve higher prediction values than others, without considering other factors, in some cases, can reduce recommendation accuracy and negatively affect the recommender system’s success. An extensive evaluation is conducted using sparse collaborative filtering datasets. It is found that the number of near neighbours used for the prediction formulation, the rating average of the user for whom the prediction is generated and the rating average of the item concerning the prediction can indicate, in many cases, whether the rating prediction produced is reliable or not.

Improving nonnegative matrix factorization with advanced graph regularization

Nonnegative Matrix Factorization (NMF) produces interpretable solutions for many applications including collaborative filtering. Typically, regularization is needed to address issues such as overfitting and interpretability, especially for collaborative filtering where the rating matrices are sparse. However, the existing regularizers are typically constructed from the factorization results instead of the rating matrices. Intuitively, we regard these existing regularizers as representing either user factors or item factors and anticipate that a more holistic regularizer could improve the effectiveness of NMF. To this end, we propose a graph regularizer based on a linear projection of the rating matrix, and call the resulting method: Linear Projection and Graph Regularized Nonnegative Matrix Factorization (LPGNMF). We develop two iterative methods to minimize the cost function and derive two update rules named LPGNMF and F-LPGNMF. Additionally, we prove the value of the objective function decreases with LPGNMF and converges to a fixed point with F-LPGNMF. Finally, we test these methods against a number of NMF algorithms on different data sets and show both LPGNMF and F-LPGNMF always achieve smaller errors based on two different error measures.

Rec-CFSVD++: Implementing Recommendation System Using Collaborative Filtering and Singular Value Decomposition (SVD)++

In recommender systems, Collaborative Filtering (CF) plays an essential role in promoting recommendation services. The conventional CF approach has limitations, namely data sparsity and cold-start. The matrix decomposition approach is demonstrated to be one of the effective approaches used in developing recommendation systems. This paper presents a new approach that uses CF and Singular Value Decomposition (SVD)[Formula: see text] for implementing a recommendation system. Therefore, this work is an attempt to extend the existing recommendation systems by (i) finding similarity between user and item from rating matrices using cosine similarity; (ii) predicting missing ratings using a matrix decomposition approach, and (iii) recommending top-N user-preferred items. The recommender system’s performance is evaluated considering Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). Performance evaluation is accomplished by comparing the systems developed using CF in combination with six different algorithms, namely SVD, SVD[Formula: see text], Co-Clustering, KNNBasic, KNNBaseline, and KNNWithMeans. We have experimented using MovieLens 100[Formula: see text]K, MovieLens 1[Formula: see text]M, and BookCrossing datasets. The results prove that the proposed approach gives a lesser error rate when cross-validation ([Formula: see text]) is performed. The experimental results show that the lowest error rate is achieved with MovieLens 100[Formula: see text]K dataset ([Formula: see text], [Formula: see text]). The proposed approach also alleviates the sparsity and cold-start problems and recommends the relevant items.

Attention Collaborative Autoencoder for Explicit Recommender Systems

Recently, various deep learning-based models have been applied in the study of recommender systems. Some researches have combined the classic collaborative filtering method with deep learning frameworks in order to obtain more accurate recommendations. However, these models either add additional features, but still recommend in the original linear manner, or only extract the global latent factors of the rating matrices in a non-linear way without considering some local special relationships. In this paper, we propose a deep learning framework for explicit recommender systems, named Attention Collaborative Autoencoder (ACAE). Based on the denoising autoencoder, our model can extract the global latent factors in a non-linear fashion from the sparse rating matrices. In ACAE, attention units are introduced during back propagation, enabling discovering potential relationships between users and items in the neighborhood, which makes the model obtain better results in the rating prediction tasks. In addition, we propose how to optimize the training process of the model by proposing a new loss function. Experiments on two public datasets demonstrate the effectiveness of ACAE and its outperformance of competitive baselines.

Tag-informed collaborative topic modeling for cross domain recommendations

Collaborative topic modeling is powerful to alleviate data sparsity in recommender systems owing to the incorporation of collaborative filtering and topic models. However sufficient textual data is not always available. On the other hand, tags serving as supplementary description of items can reflect users’ interests in item attributes. But previous works only mine the effect of tags on ratings in one domain and ignore that in related domains items can be related in attributes. Tags encode similar properties of items and can be transferred across domains to mutually benefit recommendations for both domains. In this study we propose a TagCDCTR (Tag-informed Cross Domain Collaborative Topic Regression) model, which exploits shared tags as bridges to link related domains through an extended collaborative topic modeling framework. The model exploits the inter-domain relations by encoding cross domain item–item similarity based on common tags and jointly learning a shared set of topics from all domains together. Collectively factorizing the rating matrices of multiple domains into common user latent factors and domain-specific item latent factors, so that the learned item latent factors are linked through the inter-domain relations, helping to capture the items more comprehensively. The rich information reused in multiple domains alleviates data sparsity and the semantic advantage of topics and tags provides a better interpretability of recommendations. The experiments conducted on three datasets demonstrate that TagCDCTR outperforms state-of-the-art collaborative-topic-based models and cross-domain-based models.

A new prediction method for recommendation system based on sampling reconstruction of signal on graph

Recommendation technology is widely used in various e-commerce platforms. Accurately predicting user’s preference is the most important goal of recommendation technology. One of the core difficulties of recommendation technology that the rating matrices are seriously sparse. However, the unknown entries in the rating matrix actually contain a lot of useful information for prediction, which are usually discarded in traditional methods. Based on the idea of semi-supervised learning, this paper models the recommendation problem as a signal reconstruction problem on a graph. The new model utilizes both the information of the unlabeled samples and the location information, and thus achieves an excellent predictive performance. Meanwhile, to reduce the computational complexity a strategy is designed skillfully to approximately solve the model. Experimental results shows that the proposed method significantly outperforms the reference methods in predictive accuracy and is robust to the diversity of data sets.

Knowledge transfer by domain-independent user latent factor for cross-domain recommender systems

Collaborative filtering is one of the most promising used techniques in recommender systems to provide personalized recommendations. However, it suffers from the data sparsity problem due to insufficient user preference data. The promising solution to this problem is knowledge exploitation from other related source domains through transfer learning mechanism. Although several works have been proposed to enhance prediction performance of the target domain by leveraging the knowledge of source domains, extracting domain-independent knowledge from source domains and preventing the negative knowledge both are challenging and open problems for cross-domain recommender systems framework.In this paper, we propose a method namely Knowledge Transfer by Domain-independent User Latent Factor for Cross-domain Recommender Systems. The propose method uses matrix tr-factorization by which a rating matrix decomposes into three sub-matrices. Advantage of tri-factorization is to be able to prevent domain effect knowledge transfer. This allows to enhance the overall performance of prediction in the target domain. We collectively factorize both domain rating matrices by matrix tri-factorization with the constraint of shared user latent factor across domains. The constraint holds to be true due to users fully overlap across domains. The loss functions of matrix tri-factorization are combined with a tuning parameter that is able to prevent negative transferring of a source domain. The combined loss function is optimized by stochastic gradient descent algorithm. The superiority of our proposed method over existing state-of-the-art methods is demonstrated by experimental results which have been done on real-world Amazon co-purchasing metadata.

Dynamic BIM component recommendation method based on probabilistic matrix factorization and grey model

With rapid advances in building information modeling (BIM), a huge amount of BIM components has been built to increase design efficiency. Meanwhile, finding the appropriate BIM component in the huge library has become a challenge. Besides the methods of case-based reasoning (CBR) or multi-attribute decision model (MADM), the probabilistic matrix factorization (PMF) method of a recommendation system can be an efficient alternative. However, the user behavior patterns (i.e., the rating matrices) are changing with time to influence the recommendation precision. Therefore, this study aims to enhance the dynamic recommendation ability for BIM components by proposing a hybrid probabilistic matrix factorization method (PMF-GMn). The latent user preference matrix and the latent BIM component feature matrix can be generated by the PMF method from the rating matrix. Then, the predicted latent matrices can be obtained by the optimized grey model. Finally, the predicted latent matrices are further combined into the predicted rating matrix to recommend the appropriate BIM components. An illustrative example of the prefabricated building design is used to demonstrate the feasibility. This experiment is implemented by inviting twenty users to use the proposed SharePBIM platform for five months. The statistical results indicated that PMF-GMn can provide better performance than PMF in both two criteria of RMSE and Recall@k.

PrivateGraph: Privacy-Preserving Spectral Analysis of Encrypted Graphs in the Cloud

Big graphs, such as user interactions in social networks and customer rating matrices in collaborative filters, possess great values for both businesses and research. They are not only big but often keep evolving, which requires a large amount of computing resources to maintain. With the wide deployment of public cloud resources, owners of big graphs may want to use cloud resources to obtain storage and computation scalability. However, privacy and ownership of the graphs in the cloud has become a major concern. In this paper, we study privacy-preserving algorithms for one of the important graph analysis techniques—graph spectral analysis for outsourced graph in the cloud. The core operation: eigendecomposition of large matrix is also important to many data mining algorithms. We consider a cloud-centric framework with three collaborative parties: data contributors, data owner, and cloud provider. Graphs are represented as matrices such as adjacency matrix and Laplacian matrix, the elements of which are encrypted and submitted by distributed contributors. The data owner then interacts with the cloud-side programs to conduct spectral analysis, while protecting data privacy from the honest-but-curious cloud provider. For a $N\times N$N×N graph matrix, we aim to design algorithms with the cloud handling expensive storage and computation in $O(N^2)$O(N2) complexity, while data owner and data contributors’ algorithms take only $O(N)$O(N). To achieve this goal, we develop the privacy-preserving versions of the two approximate eigendecomposition algorithms: the Lanczos algorithm and the Nystrom algorithm, considering two encryption methods: additive homomorphic encryption (AHE) methods and somewhat homomorphic encryption (SHE) methods. Both dense and sparse matrices are studied, while sparse matrices also involve a differentially private data submission protocol to allow the trade-off between data sparsity and privacy. Experimental results show that the Nystro algorithm with sparse encoding can dramatically reduce data owners’ costs. SHE-based methods have lower computational time while AHE-based methods have lower communication/storage costs.

Low-Complexity Decentralized Recommendation System With Similarity Constraints

In recommendation systems (RSs), nowadays, not only the traditional user-item rating matrices but also more additional information like contents, contexts, trust friends and other auxiliary information are available to enhance the performance of RS, leading to content-aware, context-aware, trust-aware RS, etc. Thus, it provides much potential to take into consideration the additional information in RS. Hence, we focus on a general low-rank matrix factorization (LRMF) model with similarity constraints and propose a decentralized algorithm based on alternating direction method of multipliers (ADMM) to relieve the computation burden in each server while preserving privacy. What’s more, we utilize low-complexity skills in numerical analysis to reduce the computational complexity, based on the exploitation of the special form of the problem. Finally, simulations are performed to validate the effectiveness of our algorithms.

Moisture flux adjustments in RegCM4 for improved simulation of Indian summer monsoon precipitation

The complexity of the Indian summer monsoon precipitation makes it’s prediction a challenging task as it is not only influenced by the large-scale flows but also by the micro-scale features. In a dynamical model, precipitation is resulted from the formation of clouds. The cloud formation and its processes occur at a micro scale. Current state-of-the-art dynamical models lack proper representation of the cloud processes, particularly at high resolutions for which the cloud processes are parameterized, thereby poorly resolving the precipitation. This study aims at examining the impact of the cloud parameters on the simulation of Indian summer monsoon precipitation in Regional Climate Model version 4 (RegCM4). The autoconversion coefficient which determines the conversion of cloud water into precipitation in the Explicit Moisture Convergence scheme is adjusted in the RegCM4. The impact of autoconversion is experimented with ten different values and it is found that it has a significant effect on the simulation of precipitation during summer monsoon season. The experiments are conducted by changing autoconversion from 1.5 × 10−4 to 7.5 × 10−4/s along with the default value of 2.5 × 10−4/s. On changing the autoconversion values from 60 to 300% of the default value, the precipitation pattern improves over most parts of India. The model simulates the rainfall better when the autoconversion coefficient is changed to 7.5 × 10−4/s. With the best outcome with the adjusted autoconversion and control configuration, the model is simulated for seventeen monsoon seasons and the analyses of RegCM4 simulated vertically integrated moisture transport, convective available potential energy and atmospheric moisture budget suggest that the model efficacy is enhanced in higher autoconversion value than the control one. Statistical evaluations using bias, correlation coefficients, comprehensive rating matrices and skill score confirm the suitability of higher autoconversion rate for summer monsoon simulations. The model with adjusted autoconversion coefficient (at 7.5 × 10−4/s) has improved the representation of seasonal precipitation distribution and its year-to-year variation including other derived features. The rainfall pattern is improved over North West India and North East India especially, the monsoon core regions. The mean seasonal rainfall is in phase 94% of the time with the modeler-adjusted moisture as compared to 82% in the control in the long term simulation.

RegCM4 model sensitivity to horizontal resolution and domain size in simulating the Indian summer monsoon

The sensitivity to model resolution and domain size of the regional climate model RegCM4 is examined in simulating a deficit (1987), excess (1988), and normal (1989) Indian summer monsoon (ISM) rainfall. The initial and boundary conditions are prescribed by the European Centre for Medium-Range Weather Forecasts (ECMWF) Interim reanalysis at 1.5° × 1.5° (EIN15). The model simulated precipitation is compared with India Meteorological Department (IMD) gridded precipitation data (0.25° × 0.25°) and other parameters are compared with EIN15. In the present study, eight different horizontal resolutions such as 60, 55, 50, 45, 40, 36, 27 and 18 km are used whereas four different domain sizes (D01, D02, D03, and D04; largest to smallest) centered at the Indian landmass are considered to investigate the model performance.Results illustrate that the RegCM4 has the capability to depict the important semi-permanent features of the ISM, however, skills vary with the resolutions and domain sizes. Experiments with different resolutions reveal that the lower (850 hPa) and upper (200 hPa) tropospheric circulations are stronger over the Arabian Sea and Bay of Bengal in the 36 and 40 km grid-spacing simulations and closer to EIN15, however, it is weaker in the high-resolution simulations at 27 and 18 km grid-spacing. The biases in vertically integrated moisture transport (VIMT) are positive in the model simulations with resolutions from 36 to 60 km, but, it is negative with 27 and 18 km grid resolution. The simulated precipitation intensity increases as one moves from coarse to higher resolution till grid spacing 36 km. The model efficiency reduces to simulate higher precipitation intensity over the maximum precipitation zones of India during summer monsoon season when grid spacing 27 km and 18 km are used. The overall performance is better with the use of 36–40 km grid spacing than other resolutions in simulating ISM and associated rainfall. The sensitivity of domain sizes is examined by confining the model resolution at 40 km. The model simulated meteorological parameters, as well as derived parameters (such as pattern and intensity of precipitation, geopotential height, relative humidity, air temperature, wind, and CAPE), are represented better in D02 than other domains. The present study proposes a Comprehensive Rating Matrices (CRM) to evaluate the model performance considering all experiments (horizontal resolution and domain size) and different homogeneous regions, including errors obtained using different statistical methods. The CRM and statistical based skill score (SS) suggest that the performance of RegCM4 is the best at 40 km horizontal resolution followed by 36 km and 45 km grid-spacing and the D02 domain is suitable for ISM simulation.

Evaluating recommendation and search in the labor market

This study evaluates the most popular recommender system algorithms for use on both sides of the labor market: job recommendation and job seeker recommendation. Recent research shows the drawbacks of focusing solely on predictive power when evaluating recommender systems, which become especially prominent in job- and job seeker recommendation, where aspects such as reciprocity and item spread are two other vital performance metrics for the quality of recommendations. Besides evaluating using these extra metrics, we compare recommendation with search using free text search engines. We measure what is gained, and what is lost when consuming items (jobs and job seekers) retrieved using search versus items presented via a recommender system. Based on insights in date recommendation literature, we propose changes to rating matrix construction aimed at mitigating the drawbacks of recommendation in the labor market. Our results, obtained from extensive experimentation on three datasets gathered from the Flemish public employment services, show that popular recommender algorithms perform significantly worse than user search in terms of reciprocity. Furthermore, we show that by swapping the rating matrices between two sides of a reciprocal recommender context, we can outperform user search in terms of reciprocity with limited trade off in predictive power. The insights from this research can help actors in the labor market to better understand the positioning of recommendation versus search, and to provide better job recommendations and job seeker recommendations.