MovieLens 1M Research Articles

Online Pricing with Reserve Price Constraint for Personal Data Markets

The society's insatiable appetites for personal data are driving the emergence of data markets, allowing data consumers to launch customized queries over the datasets collected by a data broker from data owners. In this paper, we study how the data broker can maximize its cumulative revenue by posting reasonable prices for sequential queries. We thus propose a contextual dynamic pricing mechanism with the reserve price constraint, which features the properties of ellipsoid for efficient online optimization and can support linear and non-linear market value models with uncertainty. In particular, under low uncertainty, the proposed pricing mechanism attains a worst-case cumulative regret logarithmic in the number of queries. We further extend our approach to support other similar application scenarios, including hospitality service and online advertising, and extensively evaluate all three use cases over MovieLens 20M dataset, Airbnb listings in U.S. major cities, and Avazu mobile ad click dataset, respectively. The analysis and evaluation results reveal that: (1) our pricing mechanism incurs low practical regret, while the latency and memory overhead incurred is low enough for online applications; and (2) the existence of reserve price can mitigate the cold-start problem in a posted price mechanism, thereby reducing the cumulative regret.

Utilizing an Autoencoder-Generated Item Representation in Hybrid Recommendation System

While collaborative filtering (CF) is the most popular approach for recommendation systems, it only makes use of the ratings given to items by users and neglects side information about user attributes or item features. In this work, a natural language processing (NLP) technique is applied to generate a more consistent version of Tag Genome, a side information which is associated with each movie in the MovieLens 20M dataset. Subsequently, we propose a 3-layer autoencoder to create a more compact representation of these tags which improves the performance of the system both in accuracy and in computational complexity. Finally, the proposed representation and the well-known matrix factorization techniques are combined into a unified framework that outperforms the state-of-the-art models by at least 2.87% and 3.36% in terms of RMSE and MAE, respectively.

Enriching Non-negative Matrix Factorization with Contextual Embeddings for Recommender Systems

Recommender Systems (RS) are used to give customized recommendations about specific items to users in a variety of applications including social web sites applications, media recommendation and commerce sites etc. Collaborative Filtering (CF) along with the Content Based Filtering (CBF) are two widely used methods which are being efficiently applied in RS implementation. CF suffers with sparseness problem where user-to-item ratings are amply sparse. On the other hand, CBF performance rely on methods of feature extraction for efficient use of items’ description. The sparseness of user-to-item ratings and features extraction impede the performance of RS. Quality of rating prediction and item recommendation further degrades due to the negative values present in users/items latent factors. This paper proposes a novel RS that is built upon the semantics based items’ content embedding model, enriched with contextual features extracted through Convolutional Neural Network (CNN). Non-negative Matrix Factorization (NMF), supplied with improvised embedding is used as CF technique. Embedding model captures the item details, thus resolving the sparsity, whereas, NMF caters for information loss due to negative values in latent factors. The proposed RS with contextually enriched NMF (Contx-NMF) simultaneously overcomes both the issues of sparseness and loss due to negative values, thus enhancing the rating prediction accuracy. The proposed model is evaluated on three public datasets (MovieLens 1M, MovieLens 10M)and Amazon Instant Video (AIV). The results demonstrate significant improvement in performance of Contx-NMF over state of the art RS models for sparse user-to-item ratings.

Mitigating long tail effect in recommendations using few shot learning technique

Recommender system has been established as an effective tool for users in providing personalized suggestions in many domains, especially in e-commerce. In these domains, recommendations are provided based on the feedback (ratings) given by the users. However, recommendations provided by the traditional approaches are biased towards the popular items (items that receive more number of ratings). As a result, unpopular items are left out and these items remain un-recommended and unsold. These unpopular items form a “long tail” in the product space, resulting in a huge loss to the e-commerce industry. However, diminishing this long tail effect is a highly challenging and non-trivial task due to limited available rating information. Recommending long tail items helps in improving the item liquidation and recommendation diversity as well.In this paper, we propose a novel framework to mitigate the long tail effect and overcome the limited ratings problem using few shot learning techniques. Siamese network, a type of few shot learning technique is found to be performing well in many domains with a limited number of instances in the recent past. In the proposed framework, vital statistics of each user are computed and this information is provided to deep siamese network. The trained siamese network is used to identify the long tail items that are similar to the liked items of each user. Finally, the identified long tail items are recommended to the appropriate users. We introduce three novel performance metrics to evaluate the long tail item recommendations. The proposed framework is evaluated on two real world datasets (MovieLens 1M and Netflix) and the results demonstrate that the proposed framework outperforms the traditional approaches and existing long-tail recommendation techniques.

Joint Neural Collaborative Filtering for Recommender Systems

We propose a Joint Neural Collaborative Filtering (J-NCF) method for recommender systems. The J-NCF model applies a joint neural network that couples deep feature learning and deep interaction modeling with a rating matrix. Deep feature learning extracts feature representations of users and items with a deep learning architecture based on a user-item rating matrix. Deep interaction modeling captures non-linear user-item interactions with a deep neural network using the feature representations generated by the deep feature learning process as input. J-NCF enables the deep feature learning and deep interaction modeling processes to optimize each other through joint training, which leads to improved recommendation performance. In addition, we design a new loss function for optimization that takes both implicit and explicit feedback, point-wise and pair-wise loss into account. Experiments on several real-world datasets show significant improvements of J-NCF over state-of-the-art methods, with improvements of up to 8.24% on the MovieLens 100K dataset, 10.81% on the MovieLens 1M dataset, and 10.21% on the Amazon Movies dataset in terms of HR@10. NDCG@10 improvements are 12.42%, 14.24%, and 15.06%, respectively. We also conduct experiments to evaluate the scalability and sensitivity of J-NCF. Our experiments show that the J-NCF model has a competitive recommendation performance with inactive users and different degrees of data sparsity when compared to state-of-the-art baselines.

Sanitizing and measuring privacy of large sparse datasets for recommender systems

Big Data is characterized by large volumes of highly dynamical data and is used for discovering hidden trends and correlations. However, as more data is collected, previous pieces of information can be put together to facilitate linkage of private records. In this context, when protecting the privacy of data subjects, the same attributes that are to be protected may be used for further re-identification, that is, sensitive attributes may be used as quasi-identifiers. For example, in high-dimensional data such as recommendations, transaction records or geo-located data, previously published transactions and locations may be used to uncover further private transactions and locations. In this paper, we propose a k-anonymization algorithm and a metric for privacy in databases in which all the attributes are quasi-identifiers as well as sensitive attributes. We apply our algorithm on high dimensional datasets for model-based and memory-based collaborative filtering, and use the metric to perform privacy comparisons between different methods of protection such as k-anonymity and differential privacy. We show the applicability of our method by performing tests on the large and sparse dataset (MovieLens 20M) of 20 million ratings that 138,493 users gave to 27,278 movies.

Bridging the gap between linked open data-based recommender systems and distributed representations

Recently, several methods have been proposed for introducing Linked Open Data (LOD) into recommender systems. LOD can be used to enrich the representation of items by leveraging RDF statements and adopting graph-based methods to implement effective recommender systems. However, most of those methods do not exploit embeddings of entities and relations built on knowledge graphs, such as datasets coming from the LOD. In this paper, we propose a novel recommender system based on holographic embeddings of knowledge graphs built from Wikidata, a free and open knowledge base that can be read and edited by both humans and machines. The evaluation performed on three standard datasets such as Movielens 1M, Last.fm and LibraryThing shows promising results, which confirm the effectiveness of the proposed method.

Improvising Singular Value Decomposition by KNN for Use in Movie Recommender Systems

Online recommender systems are an integral part of e-commerce. There are a plethora of algorithms following different approaches. However, most of the approaches except the singular value decomposition (SVD), do not provide any insight into the underlying patterns/concepts used in item rating. SVD used underlying features of movies but are computationally resource-heavy and performs poorly when there is data sparsity. In this article, we perform a comparative study among several pre-processing algorithms on SVD. In the experiments, we have used the MovieLens 1M dataset to compare the performance of these algorithms. KNN-based approach was used to find out K-nearest neighbors of users and their ratings were then used to impute the missing values. Experiments were conducted using different distance measures, such as Jaccard and Euclidian. We found that when the missing values were imputed using the mean of similar users and the distance measure was Euclidean, the KNN-based (K-Nearest Neighbour) approach of pre-processing the SVD was performing the best. Based on our comparative study, data managers can choose to employ the algorithm best suited for their business.

Detecting shilling attacks in recommender systems based on analysis of user rating behavior

The existing unsupervised methods for detecting shilling attacks are mostly based on the rating patterns of users, ignoring the rating behavior difference between genuine users and attack users, and these methods suffer from low accuracy in detecting various shilling attacks without a priori knowledge of the attacks. To address these limitations, we propose a novel unsupervised shilling attack detection model based on an analysis of user rating behavior. First, we identify the target item(s) and the corresponding intentions of the attack users by analyzing the deviation of rating tendencies on each item, and based on this analysis, a set of suspicious users is constructed. Second, we analyze the users’ rating behaviors from an interest preference and rating preference perspective. In particular, we measure the diversity and memory of users’ interest preferences by entropy and block entropy, respectively, and we analyze the memory of user rating preferences by a self-correlation analysis. Finally, we calculate the suspicious degree and spot attack users in the set of suspicious users based on measurements of user rating behavior. Experimental results on the Netflix dataset, the MovieLens 1M dataset and the sampled Amazon review dataset demonstrate the effectiveness of the proposed detection model.

A Stable Collaborative Filtering Algorithm for Long Tail Recommendation

As a relatively successful recommendation system, the collaborative filtering recommendation system (CFRS) has been widely used in e-commerce. However, the current CFRS is mainly based on mainstream or popular products to recommending similar items for users and is less efficiency in recommend the so called “Long Tail” products to meet the individual needs of users. Based on the Item-based system filtering recommendation algorithm, this paper proposes a collaborative filtering recommendation algorithm that implements long tail recommendation by using the item rating probability matrix and item rating reliability. Compared with the traditional collaborative filtering algorithm, the experimental result based on MovieLens 1M dataset shows that the proposed algorithm can deal with the data sparsity problem better, and is better for producing recommendation for the long tail products effect, and furthermore, it shows stability to a certain extent in producing recommendations under different situations of data sparseness.

HRS-CE: A hybrid framework to integrate content embeddings in recommender systems for cold start items

Recommender systems (RSs) provide the personalized recommendations to users for specific items in a wide range of applications such as e-commerce, media recommendations and social networking applications. Collaborative Filtering (CF) and Content Based (CB) Filtering are two methods which have been employed in implementing the recommender systems. CF suffers from Cold Start (CS) problem where no rating records (Complete Cold Start CSS) or very few records (Incomplete Cold Start ICS) are available for newly coming users and items. The performance of CB methods relies on good feature extraction methods so that the item descriptions can be used to measure items similarity as well as for user profiling. This paper addresses the CS problem by providing a novel way of integrating content embeddings in CF. The proposed algorithm (HRS-CE) generates the user profiles that depict the type of content in which a particular user is interested. The word embedding model (Word2Vec) is used to produce distributed representation of items descriptions. The higher representation for an item description, obtained using content embeddings, are combined with similarity techniques to perform rating predictions. The proposed method is evaluated on two public benchmark datasets (MovieLens 100k and MovieLens 20M). The results demonstrate that the proposed model outperforms the state of the art recommender system models for CS items.

Singular value decomposition based recommendation using imputed data

Among widely used recommendation methods, singular value decomposition (SVD) based approaches are the most successful ones. Although SVD-based methods are effective, they suffer from the problem of data sparsity, which could lead to poor recommendation quality. This paper proposes a novel imputation-based recommendation method, called the imputation-based SVD (ISVD), to solve the problem of data sparsity in SVD-based methods. Firstly, we propose a neighbor selection algorithm based on a similarity measure for users and items. In this algorithm, we set two thresholds to select effective neighbors for each user and item. Secondly, we generate the imputed data according to the neighbors’ ratings. Finally, we imputed these data into the SVD framework. By using imputed training data in SVD, our method can learn the prediction model accurately. We have tested our method on the MovieLens 100k, MovieLens 1M, Netflix and Filmtrust datasets. Experiment results show that our method outperforms the state-of-the-art ones. This study not only offers new insights into generating imputed data but also provides a guide to the alleviation of data sparsity in SVD-based methods.

Collaborative Filtering With User-Item Co-Autoregressive Models

Deep neural networks have shown promise in collaborative filtering (CF). However, existing neural approaches are either user-based or item-based, which cannot leverage all the underlying information explicitly. We propose CF-UIcA, a neural co-autoregressive model for CF tasks, which exploits the structural correlation in the domains of both users and items. The co-autoregression allows extra desired properties to be incorporated for different tasks. Furthermore, we develop an efficient stochastic learning algorithm to handle large scale datasets. We evaluate CF-UIcA on two popular benchmarks: MovieLens 1M and Netflix, and achieve state-of-the-art performance in both rating prediction and top-N recommendation tasks, which demonstrates the effectiveness of CF-UIcA.

Hybrid recommendations by content-aligned Bayesian personalized ranking

ABSTRACTIn many application domains of recommender systems, content-based (CB) information are available for users, objects or both. CB information plays an important role in the process of recommendation, especially in cold-start scenarios, where the volume of feedback data is low. However, CB information may come from several, possibly external, sources varying in reliability, coverage or relevance to the recommending task. Therefore, each content source or attribute possess a different level of informativeness, which should be taken into consideration during the process of recommendation. In this paper, we propose a Content-Aligned Bayesian Personalized Ranking Matrix Factorization method (CABPR), extending Bayesian Personalized Ranking Matrix Factorization (BPR) by incorporating multiple sources of content information into the BPR’s optimization procedure. The working principle of CABPR is to calculate user-to-user and object-to-object similarity matrices based on the content information and penalize differences in latent factors of closely related users’ or objects’. CABPR further estimates relevance of similarity matrices as a part of the optimization procedure. CABPR method is a significant extension of a previously published BPR_MCA method, featuring additional variants of optimization criterion and improved optimization procedure. Four variants of CABPR were evaluated on two publicly available datasets: MovieLens 1M dataset, extended by data from IMDB, DBTropes and ZIP code statistics and LOD-RecSys dataset extended by the information available from DBPedia. Experiments shown that CABPR significantly improves over standard BPR as well as BPR_MCA method w.r.t. several cold-start scenarios.

A Novel Deep Learning-Based Collaborative Filtering Model for Recommendation System.

The collaborative filtering (CF) based models are capable of grasping the interaction or correlation of users and items under consideration. However, existing CF-based methods can only grasp single type of relation, such as restricted Boltzmann machine which distinctly seize the correlation of user-user or item-item relation. On the other hand, matrix factorization explicitly captures the interaction between them. To overcome these setbacks in CF-based methods, we propose a novel deep learning method which imitates an effective intelligent recommendation by understanding the users and items beforehand. In the initial stage, corresponding low-dimensional vectors of users and items are learned separately, which embeds the semantic information reflecting the user-user and item-item correlation. During the prediction stage, a feed-forward neural networks is employed to simulate the interaction between user and item, where the corresponding pretrained representational vectors are taken as inputs of the neural networks. Several experiments based on two benchmark datasets (MovieLens 1M and MovieLens 10M) are carried out to verify the effectiveness of the proposed method, and the result shows that our model outperforms previous methods that used feed-forward neural networks by a significant margin and performs very comparably with state-of-the-art methods on both datasets.

A new method to find neighbor users that improves the performance of Collaborative Filtering

We propose a new neighbors finding method for CF based on subspace clustering.Based on subspaces, tree structures of neighbor users are drawn for the target user.Proposed method finds the best neighbors without any adjustable parameters.A new similarity method is proposed to compute the similarity value.Proposed method has been tested by Movielens 100K, Movielens 1M and Jester datasets. Recommender Systems (RS) are used to help people reduce the amount of time they spend to find the items they are looking for. One of the most successful techniques used in RS is called Collaborative Filtering (CF). It looks into the choices made by other users to find items that are most similar to the target user. Data sparsity and high dimensionality which are common in the RS domains have negatively affected the efficiency of CF. The current paper seeks to solve the mentioned problems through a neighbor user finding method which has been derived from the subspace clustering approach. In this method, the authors extract different subspaces of rated items under the categories of Interested, Neither Interested Nor Uninterested, and Uninterested. Based on subspaces, tree structures of neighbor users are drawn for the target user. Furthermore, a new similarity method is proposed to compute the similarity value. This new method has been tested via the Movielens 100K, Movielens 1M and Jester datasets in order to make a comparison with the traditional techniques. The results have indicated that the proposed method can enhance the performance of the Recommender Systems.

A Novel Preferential Diffusion Recommendation Algorithm Based on User’s Nearest Neighbors

Recommender system is a very efficient way to deal with the problem of information overload for online users. In recent years, network based recommendation algorithms have demonstrated much better performance than the standard collaborative filtering methods. However, most of network based algorithms do not give a high enough weight to the influence of the target user’s nearest neighbors in the resource diffusion process, while a user or an object with high degree will obtain larger influence in the standard mass diffusion algorithm. In this paper, we propose a novel preferential diffusion recommendation algorithm considering the significance of the target user’s nearest neighbors and evaluate it in the three real-world data sets: MovieLens 100k, MovieLens 1M, and Epinions. Experiments results demonstrate that the novel preferential diffusion recommendation algorithm based on user’s nearest neighbors can significantly improve the recommendation accuracy and diversity.

K-means Atkinson clustering approach for collaborative filtering-based recommendation system

The amount of information generated by web is growing rapidly every day and results in information overload. The information overload problem makes recommendation system necessary. Collaborative filtering is one of the most successful approaches to design a recommendation system. The key idea of this technique is based on the common interest of users. If the user has similar taste in past for a set of items, then they will share common taste in future. However, sparsity and scalability are major drawbacks of this prosperous approach affecting the quality of recommendations. In this paper, a clustering-based recommendation algorithm is proposed. The clustering technique has been used to form neighbourhoods (groups of users who have similar preferences) of active user. It exploits underlying data correlation structures to choose the initial centroid for k-means. The experimental results on three benchmark datasets, MovieLens 100k, MovieLens 1M and Jester, demonstrates that proposed method exhibits superior accuracy in comparison to the traditional k-means based recommender systems.

N-dimensional Markov random field prior for cold-start recommendation

A recommender system is a commonly used technique to improve user experience in e-commerce applications. One of the popular recommender methods is Matrix Factorization (MF) that learns the latent profile of both users and items. However, if the historical ratings are not available, the latent profile will draw from a zero-mean Gaussian prior, resulting in uninformative recommendations. To deal with this issue, we propose using an n-dimensional Markov random field as the prior of matrix factorization (called mrf-MF). In the Markov random field, the attribute (such as age, occupation of users and genre, release year of items) is considered as the site and the latent profile, the random variable. Through the prior, new users or items will be recommended according to its neighbors. The proposed model is suitable for three types of cold-start recommendation: (1) recommend new items to existing users; (2) recommend new users for existing items; (3) recommend new items to new users. The proposed model is assessed on two movie datasets, Movielens 100K and Movielens 1M. Experimental results show that it can effectively solve each of the three cold-start problems and outperforms several matrix factorization based methods.

Multi-linear interactive matrix factorization

Recommender systems, which can significantly help users find their interested items from the information era, has attracted an increasing attention from both the scientific and application society. One of the widest applied recommendation methods is the Matrix Factorization (MF). However, most of MF based approaches focus on the user-item rating matrix, but ignoring the ingredients which may have significant influence on users’ preferences on items. In this paper, we propose a multi-linear interactive MF algorithm (MLIMF) to model the interactions between the users and each event associated with their final decisions. Our model considers not only the user-item rating information but also the pairwise interactions based on some empirically supported factors. In addition, we compared the proposed model with three typical other methods: user-based collaborative filtering (UCF), item-based collaborative filtering (ICF) and regularized MF (RMF). Experimental results on two real-world datasets, MovieLens 1M and MovieLens 100k, show that our method performs much better than other three methods in the accuracy of recommendation. This work may shed some light on the in-depth understanding of modeling user online behaviors and the consequent decisions.