Long-Tail Recommendation Based on Reflective Indexing

Szwabe, Andrzej; Ciesielczyk, Michal; Misiorek, Pawel

doi:10.1007/978-3-642-25832-9_15

Cited by 7 publications

(5 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this paper, we present the first application of the proposed framework for the scenario of one-relational training sets (consisting of propositions of the likes relation only). As a consequence of using one-relational training sets for each compared method, the observed recommendation quality for the experiments presented in this paper is lower than the quality observed in experiments on the ML100k dataset including all the ratings (i.e., in [5] or [16]).…”

Section: Datasetscontrasting

confidence: 59%

“…Moreover, all the methods have been analyzed from the informationtheoretic perspective. In order to provide reliable and realistic results, the experiments involve using datasets of various sparsity (various training-set/testing-set ratios) [13] and various heavy-tailness (i.e., various spectral features of input data) [16].…”

Section: Methodsmentioning

confidence: 99%

“…In the present paper, we propose a solution that is fully automatic due to the application of the probabilistically-grounded relational data representation and processing model of HPMPP. Our system is able to achieve a high recommendation quality for datasets of different sparsity and spectral characteristics (i.e., different heavy-tailness [16]).…”

Section: Related Workmentioning

confidence: 99%

“…For this reason, we have made our experiments more realistic by reflecting the observation made by the authors of [5] that recommending the most popular items is rarely beneficial for the users. Following [16], we have removed a specified number of the most popular items to produce a long-tailed dataset. The percentage of items remaining in the dataset after removing items from the short head has been defined as lt.…”

Section: Datasetsmentioning

confidence: 99%

“…The main assumption with regard to the dataset preparation was that extracted RDF triples should represent data of both the likes and dislikes relations. Following the approach presented in [16] (which was based on the MovieLens dataset analysis presented in [5]), we have subtracted the arithmetic average of the user average rate and the item average rate from each known rate. All the users' ratings below zero have been interpreted (both for the training set and the test set) as facts: (user X, dislikes, movie Y ) and those above zero as facts: (user X, likes, movie Y ).…”

Section: Datasetsmentioning

confidence: 99%

See 4 more Smart Citations

Holistic Entropy Reduction for Collaborative Filtering

Szwabe

Misiorek

Janasiewicz

et al. 2014

Foundations of Computing and Decision Sciences

Self Cite

View full text Add to dashboard Cite

We propose a collaborative filtering (CF) method that uses behavioral data provided as propositions having the RDF-compliant form of (user X, likes, item Y) triples. The method involves the application of a novel self-configuration technique for the generation of vector-space representations optimized from the information-theoretic perspective. The method, referred to as Holistic Probabilistic Modus Ponendo Ponens (HPMPP), enables reasoning about the likelihood of unknown facts. The proposed vector-space graph representation model is based on the probabilistic apparatus of quantum Information Retrieval and on the compatibility of all operators representing subjects, predicates, objects and facts. The dual graph-vector representation of the available propositional data enables the entropy-reducing transformation and supports the compositionality of mutually compatible representations. As shown in the experiments presented in the paper, the compositionality of the vector-space representations allows an HPMPP-based recommendation system to identify which of the unknown facts having the triple form (user X, likes, item Y) are the most likely to be true in a way that is both effective and, in contrast to methods proposed so far, fully automatic.

show abstract

Section: Datasetscontrasting

confidence: 59%

Section: Methodsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Datasetsmentioning

confidence: 99%

Section: Datasetsmentioning

confidence: 99%

See 3 more Smart Citations