=Paper=
{{Paper
|id=Vol-1388/demo_paper1
|storemode=property
|title=LibRec: A Java Library for Recommender Systems
|pdfUrl=https://ceur-ws.org/Vol-1388/demo_paper1.pdf
|volume=Vol-1388
|dblpUrl=https://dblp.org/rec/conf/um/GuoZSY15
}}
==LibRec: A Java Library for Recommender Systems==
https://ceur-ws.org/Vol-1388/demo_paper1.pdf
LibRec: A Java Library for
Recommender Systems
Guibing Guo∗ , Jie Zhang† , Zhu Sun† , and Neil Yorke-Smith‡
∗
School of Information Systems, Singapore Management University, Singapore
†
School of Computer Engineering, Nanyang Technological University, Singapore
‡
Suliman S. Olayan School of Business, American University of Beirut, Lebanon
∗
gbguo@smu.edu.sg, † {zhangj,sunzhu}@ntu.edu.sg, ‡ nysmith@aub.edu.lb
Abstract. The large array of recommendation algorithms proposed over
the years brings a challenge in reproducing and comparing their perfor-
mance. This paper introduces an open-source Java library that imple-
ments a suite of state-of-the-art algorithms as well as a series of evalua-
tion metrics. We empirically find that LibRec performs faster than other
such libraries, while achieving competitive evaluative performance.
1 Introduction
Recommender systems have been developed for decades, and a large number
of algorithms have been proposed by the community. As more and more algo-
rithms are being designed, the concern of reproducibility of algorithm perfor-
mance grows [1]. Although multiple open-source frameworks exist for the pur-
pose of algorithm reproduction and comparison, many of them only implement
a set of classic algorithms which are now outdated [2]. Hence, we posit that more
efforts are warranted in comparing the current state-of-the-art algorithms with
new algorithms that are rapidly emerging.
This paper proposes an open-source Java library for recommender systems,
called LibRec 1 . The LibRec library implements a suite of state-of-the-art recom-
mendation algorithms as well as the traditional methods. In addition, a series of
evaluation metrics are implemented including diversity-based metrics which are
rarely enabled in other libraries. LibRec provides a platform for fair comparisons
among different algorithms in multiple aspects, given the fact that the evaluative
performance depends on data characteristic. It also provides a high flexibility for
expansion with new algorithms.
2 The LibRec Library
LibRec is GPL-licensed Java software2 (version 1.7 or higher required), which
can be easily deployed and executed in platforms including MS Windows, Linux,
1
http://www.librec.net. Version 1.3 of LibRec is described in this paper
2
Source code hosted in GitHub: https://github.com/guoguibing/librec
2 G. Guo et al.
IterativeRecommender SocialRecommender
Generic Interfaces Recommender
GraphicRecommender ContextRecommender
Data Structures Recommendation Algorithms
SparseMatrix Baseline Rating Prediction Item Ranking Extension
SparseVector
DenseMatrix GlobalAvg UserKNN RegSVD BPR GBPR AR
DenseVector UserAvg ItemKNN SVD++ BUCM SBPR PD
SymmMatrix ItemAvg BiasedMF RSTE LRMF WBPR NMF
DiagMatrix UserCluster BPMF SoRec CLiMF FISM Hybrid
DataDAO ItemCluster SocialMF SoReg LDA SLIM PRankD
DataSplitter MostPop TimeSVD TrustMF BHfree WRMF SlopeOne
DataConverter … TrustSVD … RankALS ... …
Fig. 1. The Class Structure of the LibRec Library
and Mac OS. It facilitates the study of the two classic problems of recommender
systems, namely rating prediction and item recommendation. The LibRec frame-
work consists of three major components, namely generic interfaces, data struc-
tures and recommendation algorithms, as illustrated in Figure 1.
2.1 Generic Interfaces
Generic interfaces define a set of abstract recommenders that can be extended
and implemented by specific recommendation algorithms. In LibRec version 1.3,
five generic recommenders are implemented. Recommender defines a general
recommender which is extended by baselines and some other algorithms such as
SlopeOne. IterativeRecommender defines a recommender usually based on iter-
ative learning techniques. Matrix factorization-based approaches (e.g., SVD++,
BiasedMF) are often derived from it. GraphicRecommender suits for the algo-
rithms based on probabilistic graphic models (e.g., LDA, BUCM). SocialRec-
ommender defines a recommender that incorporates social information, such as
SocialMF and TrustSVD. Lastly, ContextRecommender defines a recommender
that integrates additional contextual information into recommendations, for ex-
ample, temporal information by TimeSVD. Although social connections are in-
deed a form of contextual information, we determine to expand from IterativeRe-
commender rather than from ContexRecommender. This is because many social
recommenders have been proposed and the format of social connections is rela-
tively simple and consistent across different data sets. Additional generic inter-
faces can be defined by further development. By defining those interfaces, new
algorithms can be easily implemented by focusing on their own logics.
2.2 Data Structures
At least four data structures are widely and heavily used to implement recom-
mender systems, namely sparse matrices and vectors, and dense matrices and
vectors. Other structures include symmetric and diagonal matrices. The data
structures have a great influence on the execution time of recommenders. An
LibRec: A Java Library for Recommender Systems 3
important characteristic of LibRec is that it runs much faster than other coun-
terparts3 . Our implementations are mainly inspired by a Java matrix library
MTJ4 , one of the most powerful matrix libraries in Java. MTJ implements sev-
eral sparse matrix structures in a compressed row (column) storage for effective
row (column) operations. Since it is often necessary to operate both in rows and
columns, we choose to implement a sparse matrix class by keeping both com-
pressed row and column storages, whereby additional utility functions can be
easily added. Similarly, our sparse vector is an enhanced version of that in MTJ
by incorporating a number of functionalities. For dense matrix and vector, we
discard the implementations of MTJ which only stores data in a single array
and hence not suitable for large-scale data storage. In contrast, we implement
our own versions using two-dimensional Java arrays. Java caching techniques are
also adopted to further boost algorithm executions. In addition, LibRec includes
(1) DataDAO, a data access object (DAO) for input/output data operations;
(2) DataSplitter, a utility class to split a data set into the training and test sub-
sets; (3) DataConverter, a converter to transform a format of source data sets
into another; and (4) other data structures and classes.
2.3 Recommendation Algorithms
We identify and implement three kinds of recommendation algorithms: (1) base-
lines that make little use of personalized information; (2) core algorithms that are
state-of-the-art approaches based on user-item interactions and contextual infor-
mation; and (3) other algorithms. A list of enabled recommendation algorithms
with references are elaborated at: http://www.librec.net/tutorial.html.
Recommendation algorithms can be evaluated in different settings: Given
N (ratio) ratings, K-fold cross validation and cold start, to name a few. Three
kinds of evaluation metrics are implemented: (1) predictive error-based mea-
sures including (normalized) mean absolute error (MAE), root mean square er-
ror (RMSE) and mean prediction error (MPE); (2) ranking-based measures in-
cluding mean average precision (MAP), normalized discounted cumulative gain
(NDCG), mean reciprocal rank (MRR), area under the ROC curve (AUC), pre-
cision and recall, etc.; and (3) other novel measures: currently a similarity-based
diversity measure [3] is implemented. More measures will be added in future.
3 Comparison with Other Frameworks
Many open-source libraries are available including Mahout5 , Duine6 , Cofi7 , LensKit8
MyMediaLite9 and PREA10 . Lee et al. [2] provide a detailed comparison among
3
The comparison is elaborated at http://www.librec.net/example.html
4
Matrix Toolkits Java (MTJ): https://github.com/fommil/matrix-toolkits-java
5
Mahout: https://mahout.apache.org
6
Duine: http://www.duineframework.org
7
Cofi: http://www.nongnu.org/cofi/
8
LensKit: http://lenskit.org/
9
MyMediaLite: http://www.mymedialite.net
10
PREA: http://prea.gatech.edu
4 G. Guo et al.
these different frameworks, and report that Mahout, Duine and Cofi focus only
on memory-based algorithms and hence are outdated. LensKit provides only a
few classic recommendation algorithms. We give a comparison with more ad-
vanced packages, i.e., PREA and MyMediaLite. MyMediaLite is a well-known
recommendation library written in C#. Some toolkits, e.g., Rival11 and WrapRec12
are recently designed as a wrapper of MyMediaLite for better use (e.g., data split
and evaluation). PREA is a more recently released framework implemented in
Java. However, the two libraries become less active for further development.
The last update of MyMediaLite was September 2013 and that of PREA was
June 2014. Consequently, some newly proposed algorithms are not supported by
these libraries, e.g., TrustMF, FISM and TrustSVD as we do. We also note that
graphic recommenders are rarely provided by other libraries. Besides, LibRec
provides more baseline and extension algorithms than PREA and MyMediaLite,
such as UserCluster and NMF. To evaluate recommendation performance, PREA
only provides predictive error-based metrics while MyMediaLite does not provide
novel measures beyond accuracy. In contrast, our library provides novel measures
as well as the traditional accuracy-based measures. Lastly, we empirically demon-
strate that LibRec runs much faster than PREA and MyMediaLite while achiev-
ing competitive recommendation performance. The amount of performance gain
differs from algorithm to algorithm. A detailed comparison in training and eval-
uation time can be found at: http://www.librec.net/example.html. Another
important characteristic of our library is that LibRec configures recommenders
using a configuration file, while PREA and MyMydiaLite use command lines.
The advantages of a configuration file are (1) easier to configure all possible pa-
rameters; (2) portable and easier to reproduce algorithm performance; and (3)
easier to debug programs by using alternative parameter settings in one time.
4 Conclusion
LibRec contributes to the community of recommender systems by providing (1)
a much faster implementation of a set of recent state-of-the-art recommendation
algorithms; (2) a fair and easy comparison among recommendation algorithms
in terms of multi-aspect evaluation metrics; and (3) a platform for others to
contribute more source codes of other algorithms as an open-source library.
References
1. Ekstrand, M.D., Ludwig, M., Konstan, J.A., Riedl, J.T.: Rethinking the recom-
mender research ecosystem: reproducibility, openness, and lenskit. In: Proceedings
of the fifth ACM conference on Recommender systems (RecSys). 133–140 (2011)
2. Lee, J., Sun, M., Lebanon, G.: Prea: Personalized recommendation algorithms
toolkit. Journal of Machine Learning Research 13 2699–2703 (2012)
3. Smyth, B., McClave, P.: Similarity vs. diversity. In: Proceedings of the International
Conference on Case-Based Reasoning Research and Development. 347–361 (2001)
11
http://rival.recommenders.net
12
https://github.com/babakx/WrapRec