Social networks include millions of users constantly looking for new relationships for personal or professional purposes. Social network sites recommend friends based on relationship features and content information. A significant part of information shared every day is spread in Hashtags. None of the existing content-based recommender systems uses the semantic of hashtags while suggesting new friends. Currently, hashtags are considered as strings without looking at their meanings. Social network sites group together people sharing exactly the same hashtags and never semantically close ones. We think that hashtags encapsulate some people interests. In this paper, we propose a framework showing how a recommender system can benefit from hashtags to enrich users' profiles. This framework consists of three main components: (1) constructing user's profile based on shared hashtags, (2) matching method that computes semantic similarity between profiles, (3) grouping semantically close users using clustering technics. The proposed framework has been tested on a Twitter dataset from the Stanford Large Network Dataset Collection consisting of 81306 profiles.
Recommender systems (RSs) provide an effective way of alleviating the information overload problem by selecting personalized choices. Online social networks and user-generated content provide diverse sources for recommendation beyond ratings, which present opportunities as well as challenges for traditional RSs. Although social matrix factorization (Social MF) can integrate ratings with social relations and topic matrix factorization can integrate ratings with item reviews, both of them ignore some useful information. In this paper, we investigate the effective data fusion by combining the two approaches, in two steps. First, we extend Social MF to exploit the graph structure of neighbors. Second, we propose a novel framework MR3 to jointly model these three types of information effectively for rating prediction by aligning latent factors and hidden topics. We achieve more accurate rating prediction on two real-life datasets. Furthermore, we measure the contribution of each data source to the proposed framework.
The mood of a song is a highly relevant feature for exploration and recommendation in large collections of music. These collections tend to require automatic methods for predicting such moods. In this work, we show that listening-based features outperform content-based ones when classifying moods: embeddings obtained through matrix factorization of listening data appear to be more informative of a track mood than embeddings based on its audio content. To demonstrate this, we compile a subset of the Million Song Dataset, totalling 67k tracks, with expert annotations of 188 different moods collected from AllMusic. Our results on this novel dataset not only expose the limitations of current audio-based models, but also aim to foster further reproducible research on this timely topic.
The recommendation system is an important commercial application of machine learning, where billions of feed views in the information flow every day. In reality, the interaction between user and item usually makes user's interest changing over time, thus many companies (e.g. ByteDance, Baidu, Alibaba, and Weibo) employ online learning as an effective way to quickly capture user interests. However, hundreds of billions of model parameters present online learning with challenges for real-time model deployment. Besides, model stability is another key point for online learning. To this end, we design and implement a symmetric fusion online learning system framework called WeiPS, which integrates model training and model inference. Specifically, WeiPS carries out second level model deployment by streaming update mechanism to satisfy the consistency requirement. Moreover, it uses multi-level fault tolerance and real-time domino degradation to achieve high availability requirement.
A key challenge in the application of evolutionary algorithms in practice is the selection of an algorithm instance that best suits the problem at hand. What complicates this decision further is that different algorithms may be best suited for different stages of the optimization process. Dynamic algorithm selection and configuration are therefore well-researched topics in evolutionary computation. However, while hyper-heuristics and parameter control studies typically assume a setting in which the algorithm needs to be chosen while running the algorithms, without prior information, AutoML approaches such as hyper-parameter tuning and automated algorithm configuration assume the possibility of evaluating different configurations before making a final recommendation. In practice, however, we are often in a middle-ground between these two settings, where we need to decide on the algorithm instance before the run ("oneshot" setting), but where we have (possibly lots of) data available on which we can base an informed decision. We analyze in this work how such prior performance data can be used to infer informed dynamic algorithm selection schemes for the solution of pseudo-Boolean optimization problems. Our specific use-case considers a family of genetic algorithms.
Recommender systems are an essential component of e-commerce marketplaces, helping consumers navigate massive amounts of inventory and find what they need or love. In this paper, we present an approach for generating personalized item recommendations in an e-commerce marketplace by learning to embed items and users in the same vector space. In order to alleviate the considerable cold-start problem present in large marketplaces, item and user embeddings are computed using content features and multi-modal onsite user activity respectively. Data ablation is incorporated into the offline model training process to improve the robustness of the production system. In offline evaluation using a dataset collected from eBay traffic, our approach was able to improve the Recall@k metric over the Recently-Viewed-Item (RVI) method. This approach to generating personalized recommendations has been launched to serve production traffic, and the corresponding scalable engineering architecture is also presented. Initial A/B test results show that compared to the current personalized recommendation module in production, the proposed method increases the surface rate by $\sim$6\% to generate recommendations for 90\% of listing page impressions.
Because of the superior feature representation ability of deep learning, various deep Click-Through Rate (CTR) models are deployed in the commercial systems by industrial companies. To achieve better performance, it is necessary to train the deep CTR models on huge volume of training data efficiently, which makes speeding up the training process an essential problem. Different from the models with dense training data, the training data for CTR models is usually high-dimensional and sparse. To transform the high-dimensional sparse input into low-dimensional dense real-value vectors, almost all deep CTR models adopt the embedding layer, which easily reaches hundreds of GB or even TB. Since a single GPU cannot afford to accommodate all the embedding parameters, when performing distributed training, it is not reasonable to conduct the data-parallelism only. Therefore, existing distributed training platforms for recommendation adopt model-parallelism. Specifically, they use CPU (Host) memory of servers to maintain and update the embedding parameters and utilize GPU worker to conduct forward and backward computations. Unfortunately, these platforms suffer from two bottlenecks: (1) the latency of pull \& push operations between Host and GPU; (2) parameters update and synchronization in the CPU servers. To address such bottlenecks, in this paper, we propose the ScaleFreeCTR: a MixCache-based distributed training system for CTR models. Specifically, in SFCTR, we also store huge embedding table in CPU but utilize GPU instead of CPU to conduct embedding synchronization efficiently. To reduce the latency of data transfer between both GPU-Host and GPU-GPU, the MixCache mechanism and Virtual Sparse Id operation are proposed. Comprehensive experiments and ablation studies are conducted to demonstrate the effectiveness and efficiency of SFCTR.
Our goal is to answer real-world tourism questions that seek Points-of-Interest (POI) recommendations. Such questions express various kinds of spatial and non-spatial constraints, necessitating a combination of textual and spatial reasoning. In response, we develop the first joint spatio-textual reasoning model, which combines geo-spatial knowledge with information in textual corpora to answer questions. We first develop a modular spatial-reasoning network that uses geo-coordinates of location names mentioned in a question, and of candidate answer POIs, to reason over only spatial constraints. We then combine our spatial-reasoner with a textual reasoner in a joint model and present experiments on a real world POI recommendation task. We report substantial improvements over existing models with-out joint spatio-textual reasoning.
Many collaborative recommender systems leverage social correlation theories to improve suggestion performance. However, they focus on explicit relations between users and they leave out other types of information that can contribute to determine users' global reputation; e.g., public recognition of reviewers' quality. We are interested in understanding if and when these additional types of feedback improve Top-N recommendation. For this purpose, we propose a multi-faceted trust model to integrate local trust, represented by social links, with various types of global trust evidence provided by social networks. We aim at identifying general classes of data in order to make our model applicable to different case studies. Then, we test the model by applying it to a variant of User-to-User Collaborative filtering (U2UCF) which supports the fusion of rating similarity, local trust derived from social relations, and multi-faceted reputation for rating prediction. We test our model on two datasets: the Yelp one publishes generic friend relations between users but provides different types of trust feedback, including user profile endorsements. The LibraryThing dataset offers fewer types of feedback but it provides more selective friend relations aimed at content sharing. The results of our experiments show that, on the Yelp dataset, our model outperforms both U2UCF and state-of-the-art trust-based recommenders that only use rating similarity and social relations. Differently, in the LibraryThing dataset, the combination of social relations and rating similarity achieves the best results. The lesson we learn is that multi-faceted trust can be a valuable type of information for recommendation. However, before using it in an application domain, an analysis of the type and amount of available trust evidence has to be done to assess its real impact on recommendation performance.
Nonnegative matrix factorization (NMF) is a relatively new unsupervised learning algorithm that decomposes a nonnegative data matrix into a parts-based, lower dimensional, linear representation of the data. NMF has applications in image processing, text mining, recommendation systems and a variety of other fields. Since its inception, the NMF algorithm has been modified and explored by numerous authors. One such modification involves the addition of auxiliary constraints to the objective function of the factorization. The purpose of these auxiliary constraints is to impose task-specific penalties or restrictions on the objective function. Though many auxiliary constraints have been studied, none have made use of data-dependent penalties. In this paper, we propose Zellner nonnegative matrix factorization (ZNMF), which uses data-dependent auxiliary constraints. We assess the facial recognition performance of the ZNMF algorithm and several other well-known constrained NMF algorithms using the Cambridge ORL database.