Jinoh Oh

Convolutional Matrix Factorization for Document Context-Aware Recommendation

Sparseness of user-to-item rating data is one of the major factors that deteriorate the quality of recommender system. To handle the sparsity problem, several recommendation techniques have been proposed that additionally consider auxiliary information to improve rating prediction accuracy. In particular, when rating data is sparse, document modeling-based approaches have improved the accuracy by additionally utilizing textual data such as reviews, abstracts, or synopses. However, due to the inherent limitation of the bag-of-words model, they have difficulties in effectively utilizing contextual information of the documents, which leads to shallow understanding of the documents. This paper proposes a novel context-aware recommendation model, convolutional matrix factorization (ConvMF) that integrates convolutional neural network (CNN) into probabilistic matrix factorization (PMF). Consequently, ConvMF captures contextual information of documents and further enhances the rating prediction accuracy. Our extensive evaluations on three real-world datasets show that ConvMF significantly outperforms the state-of-the-art recommendation models even when the rating data is extremely sparse. We also demonstrate that ConvMF successfully captures subtle contextual difference of a word in a document. Our implementation and datasets are available at http://dm.postech.ac.kr/ConvMF.

Protecting location privacy using location semantics

As the use of mobile devices increases, a location-based service (LBS) becomes increasingly popular because it provides more convenient context-aware services. However, LBS introduces problematic issues for location privacy due to the nature of the service. Location privacy protection methods based on k-anonymity and l-diversity have been proposed to provide anonymized use of LBS. However, the k-anonymity and l-diversity methods still can endanger the users privacy because location semantic information could easily be breached while using LBS. This paper presents a novel location privacy protection technique, which protects the location semantics from an adversary. In our scheme, location semantics are first learned from location data. Then, the trusted-anonymization server performs the anonymization using the location semantic information by cloaking with semantically heterogeneous locations. Thus, the location semantic information is kept secure as the cloaking is done with semantically heterogeneous locations and the true location information is not delivered to the LBS applications. This paper proposes algorithms for learning location semantics and achieving semantically secure cloaking.

Novel Recommendation Based on Personal Popularity Tendency

Recently, novel recommender systems have attracted considerable attention in the research community. Recommending popular items may not always satisfy users. For example, although most users likely prefer popular items, such items are often not very surprising or novel because users may already know about the items. Also, such recommender systems hardly satisfy a group of users who prefer relatively obscure items. Existing novel recommender systems, however, still recommend mainly popular items or degrade the quality of recommendation. They do so because they do not consider the balance between novelty and preference-based recommendation. This paper proposes an efficient novel-recommendation method called Personal Popularity Tendency Matching (PPTM) which recommends novel items by considering an individuals Personal Popularity Tendency (or PPT). Considering PPT helps to diversify recommendations by reasonably penalizing popular items while improving the recommendation accuracy. We experimentally show that the proposed method, PPTM, is better than other methods in terms of both novelty and accuracy.

Fast and Robust Parallel SGD Matrix Factorization

Matrix factorization is one of the fundamental techniques for analyzing latent relationship between two entities. Especially, it is used for recommendation for its high accuracy. Efficient parallel SGD matrix factorization algorithms have been developed for large matrices to speed up the convergence of factorization. However, most of them are designed for a shared-memory environment thus fail to factorize a large matrix that is too big to fit in memory, and their performances are also unreliable when the matrix is skewed. This paper proposes a fast and robust parallel SGD matrix factorization algorithm, called MLGF-MF, which is robust to skewed matrices and runs efficiently on block-storage devices (e.g., SSD disks) as well as shared-memory. MLGF-MF uses Multi-Level Grid File (MLGF) for partitioning the matrix and minimizes the cost for scheduling parallel SGD updates on the partitioned regions by exploiting partial match queries processing}. Thereby, MLGF-MF produces reliable results efficiently even on skewed matrices. MLGF-MF is designed with asynchronous I/O permeated in the algorithm such that CPU keeps executing without waiting for I/O to complete. Thereby, MLGF-MF overlaps the CPU and I/O processing, which eventually offsets the I/O cost and maximizes the CPU utility. Recent flash SSD disks support high performance parallel I/O, thus are appropriate for executing the asynchronous I/O. From our extensive evaluations, MLGF-MF significantly outperforms (or converges faster than) the state-of-the-art algorithms in both shared-memory and block-storage environments. In addition, the outputs of MLGF-MF is significantly more robust to skewed matrices. Our implementation of MLGF-MF is available at http://dm.postech.ac.kr/MLGF-MF as executable files.

Improving top-K recommendation with truster and trustee relationship in user trust network

Due to the data sparsity problem, social network information is often additionally used to improve the performance of recommender systems. While most existing works exploit social information to reduce the rating prediction error, e.g., RMSE, a few had aimed to improve the top-k ranking prediction accuracy. This paper proposes a novel top-k ranking oriented recommendation method, TRecSo, which incorporates social information into recommendation by modeling two different roles of users as trusters and trustees while considering the structural information of the network. Empirical studies on real-world datasets demonstrate that TRecSoźleads to a remarkable improvement compared with previous methods in top-k recommendation.

When to recommend: A new issue on TV show recommendation

Abstract Recommender systems have gained much attention in both research and industry communities, and have been actively researched for the last decade. However, recommendation techniques for TV shows have not been actively researched despite TV’s importance. It is because TV show recommendation has two unique and notable characteristics: (1) items (i.e., TV shows) are available only for a certain time period and (2) user cannot watch two different shows at the same time. Due to the different characteristics, TV recommender system should be able to recommend item in online time , and deciding the recommendation timing becomes an important issue for TV show recommender system. Developing such a system raises several technical challenges: (1) Since the time conditions of TV shows such as watching time and remaining time affect on how much the user is attracted to the show, recommendation must consider the time conditions as well as users’ preferences on items. (2) The cost of inaccurate recommendations (or inaccurate timing) is higher than other domains, because a recommendation involves blocking a part of screen. This paper proposes a novel recommender system for TV shows called ShowTime , which determines the timing as well as the items for recommendation. In our extensive experiments on a real-world data, the proposed TV show recommender system, ShowTime , demonstrates promising results in terms of accuracy and the cost management.

Efficient feature weighting methods for ranking

Feature weighting or selection is a crucial process to identify an important subset of features from a data set. Removing irrelevant or redundant features can improve the generalization performance of ranking functions in information retrieval. Due to fundamental differences between classification and ranking, feature weighting methods developed for classification cannot be readily applied to feature weighting for ranking. A state of the art feature selection method for ranking, called GAS, has been recently proposed, which exploits importance of each feature and similarity between every pair of features. However, GAS must compute the similarity scores of all pairs of features, thus it is not scalable for high-dimensional data and its performance degrades on nonlinear ranking functions. This paper proposes novel algorithms, RankWrapper and RankFilter, which is scalable for high-dimensional data and also performs reasonably well on nonlinear ranking functions. RankWrapper and RankFilter are designed based on the key idea of Relief algorithm. Relief is a feature selection algorithm for classification, which exploits the notions of hits (data points within the same class) and misses (data points from different classes) for classification. However, there is no such notion of hits or misses in ranking. The proposed algorithms instead utilize the ranking distances of nearest data points in order to identify the key features for ranking. Our extensive experiments show that RankWrapper and RankFilter generate higher accuracy overall than the GAS and traditional Relief algorithms adapted for ranking, and run substantially faster than the GAS on high dimensional data.

Deep Hybrid Recommender Systems via Exploiting Document Context and Statistics of Items

The sparsity of user-to-item rating data is one of the major obstacles to achieving high rating prediction accuracy of model-based collaborative filtering (CF) recommender systems. To overcome the obstacle, researchers proposed hybrid methods for recommender systems that exploit auxiliary information together with rating data. In particular, document modeling-based hybrid methods were recently proposed that additionally utilize description documents of items such as reviews, abstracts, or synopses in order to improve the rating prediction accuracy. However, they still have two following limitations on further improvements: (1) They ignore contextual information such as word order or surrounding words of a word because their document modeling methods use bag-of-words model. (2) They do not explicitly consider Gaussian noise differently in modeling latent factors of items based on description documents together with ratings although Gaussian noise depend on statistics of items.In this paper, we propose a robust document context-aware hybrid method, which integrates convolutional neural network (CNN) into probabilistic matrix factorization (PMF) with the statistics of items to both capture contextual information and consider Gaussian noise differently. Our extensive evaluations on three real-world dataset show that our variant recommendation models based on our proposed method significantly outperform the state-of-the-art recommendation models.

Processing time-dependent shortest path queries without pre-computed speed information on road networks

Shortest path (or least travel time path) identification has been actively studied for direct application to road networks. In addition, the processing of time-dependent shortest-path queries, which use past traffic data to compute the speed variations of road segments, has been investigated in order to incorporate speed variations over time. However, speed information pre-computed from static past traffic data is often invalid because road traffic is inherently dynamic. This paper addresses a new problem in processing a Dynamic Time-Dependent Shortest Path (DTDSP) query, which considers the current road status without assuming pre-determined speed patterns on roads. By dynamically adjusting the speed patterns of roads instead of fixing them based on past traffic data, the recommended paths, which reflect the current road status, are more effective in distributing the road traffic and thus reducing the travel time. To process DTDSP queries, we first propose a Continuous Piece-wise Linear Speed Pattern (CPLSP) model to compute the vehicle speed patterns, which is more flexible and realistic than previously adopted piece-wise constant speed pattern models. Using dynamically computed CPLSPs, we process a DTDSP query in two phases: (1) the least travel time path is found for the query and (2) the speed patterns of the following vehicles, which are affected by the participation of the new vehicle on the road network, are updated. We propose efficient algorithms for finding the least travel time path of a new query (vehicle) and for updating the speed patterns of the existing vehicles. Experiments on real data sets show that our query processing algorithms effectively distribute road traffic, and thus, significantly reduce both global and individual travel times.

TRecSo: Enhancing Top-k Recommendation With Social Information

Due to the data sparsity problem, social network information is often additionally used to improve the performance of recommender system. While most existing works exploit social information to reduce the rating prediction error, e.g., RMSE, a few had aimed to improve the top-k ranking prediction accuracy. This paper proposes a novel top-k oriented recommendation method, TRecSo, which incorporates social information into recommendation by modeling two different roles of users as trusters and trustees while considering the structural information of the network. Empirical studies on real-world datasets demonstrate that TRecSo leads to remarkable improvement compared to previous methods in top-k recommendation.