arXiv subjects

Arjen P. de Vries

Publications and source records attributed to Arjen P. de Vries.

23 records · Page 2Linked to original sources

arXiv2016

Exploring Deep Space: Learning Personalized Ranking in a Semantic Space

Recommender systems leverage both content and user interactions to generate recommendations that fit users' preferences. The recent surge of interest in deep learning presents new opportunities for exploiting these two sources of information. To recommend items we propose to first learn a user-independent high-dimensional semantic space in which items are positioned according to their substitutability, and then learn a user-specific transformation function to transform this space into a ranking according to the user's past preferences. An advantage of the proposed architecture is that it can be used to effectively recommend items using either content that describes the items or user-item ratings. We show that this approach significantly outperforms state-of-the-art recommender systems on the MovieLens 1M dataset.

cs.IR↗

arXiv2016

Efficient Parallel Learning of Word2Vec

Since its introduction, Word2Vec and its variants are widely used to learn semantics-preserving representations of words or entities in an embedding space, which can be used to produce state-of-art results for various Natural Language Processing tasks. Existing implementations aim to learn efficiently by running multiple threads in parallel while operating on a single model in shared memory, ignoring incidental memory update collisions. We show that these collisions can degrade the efficiency of parallel learning, and propose a straightforward caching strategy that improves the efficiency by a factor of 4.

cs.CL↗

arXiv2013

Runtime Optimizations for Prediction with Tree-Based Models

Tree-based models have proven to be an effective solution for web ranking as well as other problems in diverse domains. This paper focuses on optimizing the runtime performance of applying such models to make predictions, given an already-trained model. Although exceedingly simple conceptually, most implementations of tree-based models do not efficiently utilize modern superscalar processor architectures. By laying out data structures in memory in a more cache-conscious fashion, removing branches from the execution flow using a technique called predication, and micro-batching predictions using a technique called vectorization, we are able to better exploit modern processor architectures and significantly improve the speed of tree-based models over hard-coded if-else blocks. Our work contributes to the exploration of architecture-conscious runtime implementations of machine learning algorithms.

cs.DB↗

arXiv2011

SERIMI - Resource Description Similarity, RDF Instance Matching and Interlinking

The interlinking of datasets published in the Linked Data Cloud is a challenging problem and a key factor for the success of the Semantic Web. Manual rule-based methods are the most effective solution for the problem, but they require skilled human data publishers going through a laborious, error prone and time-consuming process for manually describing rules mapping instances between two datasets. Thus, an automatic approach for solving this problem is more than welcome. In this paper, we propose a novel interlinking method, SERIMI, for solving this problem automatically. SERIMI matches instances between a source and a target datasets, without prior knowledge of the data, domain or schema of these datasets. Experiments conducted with benchmark collections demonstrate that our approach considerably outperforms state-of-the-art automatic approaches for solving the interlinking problem on the Linked Data Cloud.

cs.DB↗

arXiv2011

Personalised Travel Recommendation based on Location Co-occurrence

We propose a new task of recommending touristic locations based on a user's visiting history in a geographically remote region. This can be used to plan a touristic visit to a new city or country, or by travel agencies to provide personalised travel deals. A set of geotags is used to compute a location similarity model between two different regions. The similarity between two landmarks is derived from the number of users that have visited both places, using a Gaussian density estimation of the co-occurrence space of location visits to cluster related geotags. The standard deviation of the kernel can be used as a scale parameter that determines the size of the recommended landmarks. A personalised recommendation based on the location similarity model is evaluated on city and country scale and is able to outperform a location ranking based on popularity. Especially when a tourist filter based on visit duration is enforced, the prediction can be accurately adapted to the preference of the user. An extensive evaluation based on manual annotations shows that more strict ranking methods like cosine similarity and a proposed RankDiff algorithm provide more serendipitous recommendations and are able to link similar locations on opposite sides of the world.

cs.IR↗