Learning User Preferences Research Articles

An effective recommendation model based on deep representation learning

Recommender system has recently attracted a lot of attention in the information service community. Currently, most recommendation models use deep neural networks to learn user preferences for items and make the final recommendations. However, these current models have not effectively captured the deep semantic features of users and items and have not fully used the auxiliary information of the items. This may result in unsatisfactory recommendations for users’ projects. In order to solve the above problem, in this paper, a novel recommendation model called RM-DRL (Recommendation Model based on Deep Representation Learning) was proposed. It mainly consists of two modules: Information Preprocessing and Feature Representation. The former generates the user’s primitive feature vectors and the items used in the latter. The latter consists of two phases: Representation Learning for Item Features (RL-IF) and Representation Learning for User Features (RL-UF). The RL-IF takes the primitive feature vectors of the item as input and uses a multi-layer Convolutional Neural Network (CNN) to learn to accurately produce the semantic feature vector of the item through multi-task learning. In RL-UF, the user primitive feature vectors and semantic feature vectors of the user preference history, and the positive and negative items were taken as input, and a novel Attention-Integrated Gated Recurrent Unit (AIGRU) neural network was proposed to learn to accurately produce user semantic feature vector. After the Feature Representation module converges, the semantic feature vectors of the users and the items can be used to calculate the users’ preferences on the items via vector dot product. Extensive experiments on five real-world datasets show that RM-DRL remarkably outperforms state-of-the-art baselines in solving the recommendation problem.

Meta-path Augmented Sequential Recommendation with Contextual Co-attention Network

It is critical to comprehensively and efficiently learn user preferences for an effective sequential recommender system. Existing sequential recommendation methods mainly focus on modeling local preference from users’ historical behaviors, which largely ignore the global context information from the heterogeneous information network. This prevents a comprehensive user preference representation. To address these issues, we propose a joint learning approach to incorporate global context with local preferences efficiently. The proposed approach introduces meta-paths from a heterogeneous information network to capture the global context information, and the position-based self-attention mechanism is adopted to model the local preference representation efficiently. Compared with the methods that only consider the local preference, our proposed method takes the advantages of incorporating global context information, which extracts structural features that captures relevant semantics to construct users’ global preference representation for the sequential recommendation. We further adopt a co-attention mechanism to model complex interactions between global context and users’ historical behaviors for better user representations. Quantitative and qualitative experimental evaluations are conducted on nine large-scale Amazon datasets and a multi-modal Zhihu dataset. The promising results demonstrate the effectiveness of the proposed model.

Who Likes What? — SplitLBI in Exploring Preferential Diversity of Ratings

In recent years, learning user preferences has received significant attention. A shortcoming of existing learning to rank work lies in that they do not take into account the multi-level hierarchies from social choice to individuals. In this paper, we propose a multi-level model which learns both the common preference or utility function over the population based on features of alternatives to-be-compared, and preferential diversity functions conditioning on user categories. Such a multi-level model, enables us to simultaneously learn a coarse-grained social preference function together with a fine-grained personalized diversity. It provides us prediction power for the choices of new users on new alternatives. The key algorithm in this paper is based on Split Linearized Bregman Iteration (SplitLBI) algorithm which generates a dynamic path from the common utility to personalized preferential diversity, at different levels of sparsity on personalization. A synchronized parallel version of SplitLBI is proposed to meet the needs of fast analysis of large-scale data. The validity of the methodology are supported by experiments with both simulated and real-world datasets such as movie and dining restaurant ratings which provides us a coarse-to-fine grained preference learning.

Learning evolving user’s behaviors on location-based social networks

With the popularity of smart phones, users’ activities on location-based social networks (LBSNs) evolve faster than traditional social networks. Existing models focus on modeling users’ long-term preferences, leveraging social collaborative filtering to enhance prediction performance. However, the dynamic mobility mechanism of user’s check-in behaviors on LBSNs is seldom considered. In this paper, we propose a new dynamic model that considers both geo-aware user preferences and the social interaction excitation arising from social connections to learn the dynamic mobility mechanism of user’s behaviors on LBSNs. Geo-aware location features, such as semantic features, latent features and dynamic features, are utilized to characterize the location information and reveal the evolution of the geographical impact of location. These geo-aware location features enable us to exploit user’s personal preferences. Meanwhile, we integrate a user’s social connections and friends’ preferences for modeling social interaction excitations. Finally, we jointly incorporate geo-aware user preference learning and social interaction excitation modeling to create a conditional intensity function for temporal point processes with which to explore the dynamic mobility mechanism of evolving user’s check-in behaviors on LBSNs. Extensive experiments on several real-world check-in datasets confirm that our proposed algorithm performs better than existing state-of-the-art methods.

2D and 3D Geovisualization: Learning user Preferences in Landslide Vulnerability

Landslides often cause impacts on the environment, infrastructure, and society. The impacts of landslides can be minimized by creating disaster awareness. Landslide vulnerability mapping can be used as a dissemination media to increase disaster awareness. The mapping methods that can be used are 2D and 3D geovisualization. There is very few research in current literature explaining the user preferences on geovisualization 2D and 3D related to landslide vulnerability. In this paper, the user preferences of both 2D and 3D geovisualization will be evaluated. This study will focus to find out which geovisualization suits most users and their literacy spatial among those provided geovisualizations. From our results, 90% of users prefer 3D geovisualization over 2D. Furthermore, our analysis shows that 2D geovisualization has the advantage of being easily understood by users in all ages. Meanwhile, 3D geovisualization is better at increasing users' spatial literacy at all ages and levels of education in knowing the causes of landslide vulnerability. Appropriate geovisualization will provide information and knowledge that is useful for communities in regards of landslide vulnerability for better disaster awareness

Social Media Users Send Promotional Links to Strangers: Legitimate Promotion or Security Vulnerability?

Nowadays, many users make money by publishing content on social media platforms. In order to attract users' attention, they often take measures to promote themselves. The security vulnerabilities in social media platforms may provide convenience for their user promotion work. We call this type of vulnerability the user promotion security vulnerability (UPSV). UPSV may cause unfair competition and endanger the interests of legitimate users and the social media platforms. Therefore it has great research significance to find and fix this vulnerability. In this paper, we propose a UPSV which widely exists in the function of sending messages to strangers of in-app chatting of many social media platforms. We first analyzed this vulnerability in some apps, and then YY app (China's largest live streaming platform) was chosen as the research object to verify the actual effect of the vulnerability on illegitimate user promotion. We took the method of promoting a target YY streamer through sending promotional links to viewers, and to improve promotion effect, we used the method of user preference learning to select viewers for promotion. The experimental results show that among the promoted viewers, more than 44% entered the target streamers' channels to watch live streaming, more than 21% followed the target steamers, and more than 13% gave gifts to the target steamers. It fully proves that this UPSV is real, exploitable and harmful, and we also proposed some fix suggestions to help the platforms to fix it.

Music recommender using deep embedding-based features and behavior-based reinforcement learning

With the rapid increase of digital music on online music platforms, it has become difficult for users to find unknown but interesting songs. Although many collaborative filtering or content based recommendation methods have been proposed, they have various relatively serious some problems, including cold start, diversity of recommendations. etc. Therefore, we propose a reinforcement personal music recommendation system (RPMRS) to address these problems. RPMRS comprises two main components. First, deep representation of audio and lyrics extracted by WaveNet and Word2Vec models, respectively, and apply a proposed content based recommendation method from these. Second, we employ reinforcement learning is to learn user preferences from their song playing log. Experimental results confirm, that hybrid features are superior to audio or lyrics based features for content recommendation, largely because independent audio features significantly outperform lyrics features; and reinforcement learning improves personalized recommendations. Overall, the proposed RPMRS provides dynamic and personalized music recommendations for the user.

Multi-facet user preference learning for fine-grained item recommendation

Existing recommendation methods mainly learn user preference from historical user-item interaction data, while ignoring the extent of interactions, i.e., diverse user experience and user intention. To be more specific, for new users with little or no experience, they may turn to popular items preferred by majority users. In terms of those medium-level users who already have interacted with some items, they may require and expect items to meet their personal preferences. As pro-active users are likely to leave rich behavior (action) feedback (e.g., view, like) on the items they interacted with, the system will have good chance to better interpret users’ intention, and thus generate more accurate and elaborate recommendations to hit their preferences. In this paper, we propose a generic Multi-facet User Preference Learning (MUPL) framework for fine-grained item recommendation. By considering diverse user experience and intention, MUPL captures user preference in the level of group-, individual- and action-facet. Besides, the importance of user preference in different facets can be automatically learnt by MUPL. Extensive experiments on two real-world datasets (Xing, Sobazaar) demonstrate the superiority of our proposed approach over other state-of-the-art methods.

The complexity of exact learning of acyclic conditional preference networks from swap examples

Learning of user preferences, as represented by, for example, Conditional Preference Networks (CP-nets), has become a core issue in AI research. Recent studies investigate learning of CP-nets from randomly chosen examples or from membership and equivalence queries. To assess the optimality of learning algorithms as well as to better understand the combinatorial structure of classes of CP-nets, it is helpful to calculate certain learning-theoretic information complexity parameters. This article focuses on the frequently studied case of exact learning from so-called swap examples, which express preferences among objects that differ in only one attribute. It presents bounds on or exact values of some well-studied information complexity parameters, namely the VC dimension, the teaching dimension, and the recursive teaching dimension, for classes of acyclic CP-nets. We further provide algorithms that exactly learn tree-structured and general acyclic CP-nets from membership queries. Using our results on complexity parameters, we prove that our algorithms, as well as another query learning algorithm for acyclic CP-nets presented in the literature, are near-optimal.

Hierarchical Attention Network for Visually-Aware Food Recommendation

Food recommender systems play an important role in assisting users to identify the desired food to eat. Deciding what food to eat is a complex and multi-faceted process, which is influenced by many factors such as the ingredients, appearance of the recipe, the user's personal preference on food, and various contexts like what had been eaten in the past meals. In this work, we formulate the food recommendation problem as predicting user preference on recipes based on three key factors that determine a user's choice on food, namely, 1) the user's (and other users') history; 2) the ingredients of a recipe; and 3) the descriptive image of a recipe. To address this challenging problem, we develop a dedicated neural network based solution Hierarchical Attention based Food Recommendation (HAFR) which is capable of: 1) capturing the collaborative filtering effect like what similar users tend to eat; 2) inferring a user's preference at the ingredient level; and 3) learning user preference from the recipe's visual images. To evaluate our proposed method, we construct a large-scale dataset consisting of millions of ratings from AllRecipes.com. Extensive experiments show that our method outperforms several competing recommender solutions like Factorization Machine and Visual Bayesian Personalized Ranking with an average improvement of 12%, offering promising results in predicting user preference for food. Codes and dataset will be released upon acceptance.

Operation Strategy of Smart Thermostats That Self-Learn User Preferences

Smart thermostats can automatically adjust indoor temperature based on user preferences to save electricity bills without significantly comprising comfort. However, current smart thermostats usually require users to master programming or require a significant amount of user behavior observations to enable automatic control, which is demanding and adverse to their popularization. In this paper, we propose a practical method that enables a smart thermostat to track user preferences and derive the optimal temperature setting schedule. We assume that users are rational and aim to minimize their overall costs. First, we propose a Bayesian-inference-based method that can quickly learn user preferences with a limited number of user behavior observations. We then generate the optimal temperature setting schedule via a stochastic expected value model. Finally, we propose an operation strategy under which a thermostat can work automatically and continuously. The “virtual user” case study indicates that the proposed method can quickly yield a satisfying probabilistic estimate of user preferences based on even only 10 observations. The “real user” case study demonstrates that the method can dynamically track user preferences and continuously generate optimal temperature setting schedules to reduce overall costs an average of 12 percent. Based on the proposed method, users can conveniently enjoy a customized temperature zone with a lower overall cost.

Gyroscope-Based Continuous Human Hand Gesture Recognition for Multi-Modal Wearable Input Device for Human Machine Interaction.

Human hand gestures are a widely accepted form of real-time input for devices providing a human-machine interface. However, hand gestures have limitations in terms of effectively conveying the complexity and diversity of human intentions. This study attempted to address these limitations by proposing a multi-modal input device, based on the observation that each application program requires different user intentions (and demanding functions) and the machine already acknowledges the running application. When the running application changes, the same gesture now offers a new function required in the new application, and thus, we can greatly reduce the number and complexity of required hand gestures. As a simple wearable sensor, we employ one miniature wireless three-axis gyroscope, the data of which are processed by correlation analysis with normalized covariance for continuous gesture recognition. Recognition accuracy is improved by considering both gesture patterns and signal strength and by incorporating a learning mode. In our system, six unit hand gestures successfully provide most functions offered by multiple input devices. The characteristics of our approach are automatically adjusted by acknowledging the application programs or learning user preferences. In three application programs, the approach shows good accuracy (90–96%), which is very promising in terms of designing a unified solution. Furthermore, the accuracy reaches 100% as the users become more familiar with the system.

LORI: A Learning-to-Rank-Based Integration Method of Location Recommendation

Location recommendation method is an important application in a location-based social network. At present, it is a trend to integrate different recommendation methods since they have their own advantages in capturing different preferences of users and an integrated method can generally provide a better performance than every individual. However, the existing integration policies do not learn user preferences in their integration processes so that they cannot make full use of the advantage of each method. Therefore, we propose a novel integration method: learning-to-rank-based integration . In our method, a confidence coefficient is applied for each user in the integration process, and these coefficients can well optimize recommendation performance. A learning-to-rank-based algorithm is designed to train the confidence coefficients. A group of experiments are done on a real large-scale check-in data set, and the results demonstrate that our method outperforms the state-of-the-art ones.

User Preference Learning-Based Edge Caching for Fog Radio Access Network

In this paper, the edge caching problem in fog radio access network (F-RAN) is investigated. By maximizing the overall cache hit rate, the edge caching optimization problem is formulated to find the optimal policy. Content popularity in terms of time and space is considered from the perspective of regional users. We propose an online content popularity prediction algorithm by leveraging the content features and user preferences, and an offline user preference learning algorithm by using the online gradient descent (OGD) method and the follow the (proximally) regularized leader (FTRL-Proximal) method. Our proposed edge caching policy not only can promptly predict the future content popularity in an online fashion with low complexity, but also can track the content popularity with spatial and temporal popularity dynamic in time without delay. Furthermore, we design two learning-based edge caching architectures. Moreover, we theoretically derive the upper bound of the popularity prediction error, the lower bound of the cache hit rate, and the regret bound of the overall cache hit rate of our proposed edge caching policy. Simulation results show that the overall cache hit rate of our proposed policy is superior to those of the traditional policies and asymptotically approaches the optimal performance.

Automated Learning of User Preferences for Selection of High Quality 3D Designs

3D object design is one of the most sought-after topics due to its application in many domains including Medical imaging, Virtual Reality, Computer Graphics, Animation etc. Manual 3D object design is inherently an expensive and labour intensive task. Several techniques such as GAN (Generative Adversarial Nets), SFM (Structure from Motion) and MVS (Multi-View Stereo) are employed to automate 3D design. 3D-GAN is one such method that allows the fully automated design of 3D objects using deep artificial neural networks. However, such generated 3D objects often suffer from distortions, obscure/missing parts, and other artefacts. For a large number of generated models, it becomes time-consuming for the human expert to select models of high quality among the set of generated ones. In this paper, we tackle this problem through machine learning powered 3D model quality assessment algorithm, that mimics the users preferences. For our experiments, we generate 3D objects using 3D-GAN and investigate several machine learning algorithms including Support Vector Machines, Gradient Boosting, Deep Neural Networks to classify generated models based on their visual appealability. We find that our model successfully mimics user choices with 90% accuracy, thus eliminating the need for human intervention in the subjective quality analysis, once user preferences have been learned. The proposed system can be seamlessly integrated into existing automated 3D design systems further augmenting the capabilities of human designers.

Image Recommendation With Reciprocal Social Influence

Image recommendation plays an important role for exploring user potential interests in large-scale image sharing websites (e.g., Flickr and Instagram). Social relationships have been exploited to learn user preference, and shown their effectiveness. We argue that their performance improvement tends to be limited, as most existing approaches only consider the side of social influence from friends to a user. However, social influence is reciprocal per se as the preference of friends will be also influenced by the user herself. In this paper, we propose a deep neural network for image recommendation (dubbed RSIM) by leveraging reciprocal social influence, and optimize the preferences of users and friends simultaneously. Specifically, we split images into three types: positive image by an active user, social image by her social friends, and negative image by neither of them. We contend that a user prefers positive image to social image, which is in turn better than negative image for relative preference learning. Two neural networks are designed to capture user and image representations by tags and visual features, respectively. The proposed model is evaluated on a real dataset crawled from Flickr. The experimental results show that better performance can be reached than the state-of-the-art social image recommendation models in terms of precision.

Next POI Recommendation via Graph Embedding Representation From H-Deepwalk on Hybrid Network

With the rapid development of location-based social networks (LBSNs), point of interest (POI) recommendation has become more and more popular personalized service. Cold start problem and poor interpretability are two main challenges of the traditional recommendation system. In this paper, we propose a novel social and sequence-aware next POI recommendation model. Our model utilizes an improved DeepWalk on heterogeneous network to learn better POI representation which contains social relationship and geographical influence. In this way, the cold start problem and interpretability problem can be solved to some extend. As for the user preference learning, we use LSTM (Long Short-Term Memory) mechanism to infer user interest, taking both long term and short term intentions into consideration. The model is trained in a metric learning framework. The experimental results on two real LBSN datasets show our model has outstanding performance in terms of AUC, Recall, and ACC. Furthermore, we can obtain 20% performance improvement on spare dataset which indicates that our model can well solve the cold start problem.

Caching Policy for Cache-Enabled D2D Communications by Learning User Preference

Prior works in a designing caching policy do not distinguish content popularity with user preference. In this paper, we illustrate the caching gain by exploiting individual user behavior in sending requests. After showing the connection between the two concepts, we provide a model for synthesizing user preference from content popularity. We then optimize the caching policy with the knowledge of user preference and activity level to maximize the offloading probability for cache-enabled device-to-device communications, and develop a low-complexity algorithm to find the solution. In order to learn user preference, we model the user request behavior resorting to probabilistic latent semantic analysis, and learn the model parameters by the expectation maximization algorithm. By analyzing a Movielens data set, we find that the user preferences are less similar, and the activity level and topic preference of each user change slowly over time. Based on this observation, we introduce a prior knowledge-based learning algorithm for user preference, which can shorten the learning time. Simulation results show a remarkable performance gain of the caching policy with user preference over existing policy with content popularity, both with realistic data set and synthetic data validated by the real data set.

Decomposition Strategies for Constructive Preference Elicitation

We tackle the problem of constructive preference elicitation, that is the problem of learning user preferences over verylarge decision problems, involving a combinatorial space of possible outcomes. In this setting, the suggested configuration is synthesized on-the-fly by solving a constrained optimization problem, while the preferences are learned iteratively by interacting with the user. Previous work has shown that Coactive Learning is a suitable method for learning userpreferences in constructive scenarios. In Coactive Learning the user provides feedback to the algorithm in the form of an improvement to a suggested configuration. When the problem involves many decision variables and constraints, this type of interaction poses a significant cognitive burden on the user. We propose a decomposition technique for large preference-based decision problems relying exclusively on inference and feedback over partial configurations. This has the clear advantage of drastically reducing the user cognitive load. Additionally, part-wise inference can be (up to exponentially) less computationally demanding than inference over full configurations. We discuss the theoretical implications of working with parts and present promising empirical results on one synthetic and two realistic constructive problems.

Constructive Preference Elicitation Over Hybrid Combinatorial Spaces

Peference elicitation is the task of suggesting a highly preferred configuration to a decision maker. The preferences are typically learned by querying the user for choice feedback over pairs or sets of objects. In its constructive variant, new objects are synthesized "from scratch" by maximizing an estimate of the user utility over a combinatorial (possibly infinite) space of candidates. In the constructive setting, most existing elicitation techniques fail because they rely on exhaustive enumeration of the candidates. A previous solution explicitly designed for constructive tasks comes with no formal performance guarantees, and can be very expensive in (or unapplicable to) problems with non-Boolean attributes. We propose the Choice Perceptron, a Perceptron-like algorithm for learning user preferences from set-wise choice feedback over constructive domains and hybrid Boolean-numeric feature spaces. We provide a theoretical analysis on the attained regret that holds for a large class of query selection strategies, and devise a heuristic strategy that aims at optimizing the regret in practice. Finally, we demonstrate its effectiveness by empirical evaluation against existing competitors on constructive scenarios of increasing complexity.