Volume 12 Supplement 3
A structural SVM approach for reference parsing
© The Author(s) 2011
Published: 09 June 2011
Automated extraction of bibliographic data, such as article titles, author names, abstracts, and references is essential to the affordable creation of large citation databases. References, typically appearing at the end of journal articles, can also provide valuable information for extracting other bibliographic data. Therefore, parsing individual reference to extract author, title, journal, year, etc. is sometimes a necessary preprocessing step in building citation-indexing systems. The regular structure in references enables us to consider reference parsing a sequence learning problem and to study structural Support Vector Machine (structural SVM), a newly developed structured learning algorithm on parsing references.
In this study, we implemented structural SVM and used two types of contextual features to compare structural SVM with conventional SVM. Both methods achieve above 98% token classification accuracy and above 95% overall chunk-level accuracy for reference parsing. We also compared SVM and structural SVM to Conditional Random Field (CRF). The experimental results show that structural SVM and CRF achieve similar accuracies at token- and chunk-levels.
When only basic observation features are used for each token, structural SVM achieves higher performance compared to SVM since it utilizes the contextual label features. However, when the contextual observation features from neighboring tokens are combined, SVM performance improves greatly, and is close to that of structural SVM after adding the second order contextual observation features. The comparison of these two methods with CRF using the same set of binary features show that both structural SVM and CRF perform better than SVM, indicating their stronger sequence learning ability in reference parsing.
Bibliographic references, typically cited at the end of scientific articles, provide much valuable information. Parsing these references is an essential step for building citation-indexing systems. Many well-known citation-indexing systems, such as CiteSeer , ISI Web of Knowledge  and Google Scholar , could have implemented complex reference parsing algorithms, though detailed reports about their algorithms and performance have not been found in the literature. As the authors of CiteSeer mention in , the reliable parsing of references may still be considered an open problem. MEDLINE®, the flagship database of the U.S. National Library of Medicine, contains over 18 million citations to the medical journal literature and is a critical source of information for biomedical research and clinical medicine. With the rapid increase of journal literature indexed by MEDLINE every year, it is essential to have automated methods to extract bibliographic data, including article titles, author names, affiliations, abstracts, and many others.
While references are not included in MEDLINE citations, they are indispensable for detecting several other items. For example, creating the Comment-On/Comment-In field for MEDLINE (identifying pairs of articles, with one article commenting on the other) requires matching references to the citing text . In addition, assigning Medical Subject Heading (MeSH) terms , an essential step in indexing the article, may also benefit from analyzing the MeSH terms assigned to the cited articles, which requires parsing the references to those articles. Reliable reference parsing is therefore an important step for automatically creating citations for MEDLINE.
In this work, our goal is to extract the following 7 entities from the references: Citation Number (<N>), Author Names (<A>), Article Title (<T>), Journal Title (<J>), Volume (<V>), Pagination (<P>) and Publication Year (<Y>). All remaining words in the reference are labeled as Other (<O>). The notation inside each parenthesis is the abbreviated entity label.
Examples of references following different styles in medical journal articles
(a) 19 S. Miyazaki, K. Takahashi, M. Shiraki, T. Saito, Y. Tezuka and K. Kasuya, Properties of a poly(3-hydroxybbutyrate) depolymerase from Penicillium funiculosum, J. Polym. Environ. 8 (2002), pp. 175–182.
<N>19</N> <A>S. Miyazaki, K. Takahashi, M. Shiraki, T. Saito, Y. Tezuka, K. Kasuya,</A> <T>Properties of a poly(3-hydroxybbutyrate) depolymerase from Penicillium funiculosum,</T> <J>J. Polym. Environ.</J> <V>8</V> <Y>(2002),</Y> <P>pp. 175–182.</P>
(b) Sofuoglu and Kosten, 2005 M. Sofuoglu and T.R. Kosten, Novel approaches to the treatment of cocaine addiction, CNS Drugs 19 (2005), pp. 13–25. Full Text via CrossRef | Abstract + References in Scopus | Cited By in Scopus
<N>Sofuoglu and Kosten, 2005</N> <A>M. Sofuoglu and T.R. Kosten,</A> <T>Novel approaches to the treatment of cocaine addiction,</T> <J>CNS Drugs</J> <V>19</V> <Y>(2005),</Y> <P>pp. 13–25.</P> <O>Full Text via CrossRef | Abstract + References in Scopus | Cited By in Scopus</O>
(c) Czarnetzki, A. B., and C. C. Tebbe. 2004. Diversity of bacteria associated with Collembola: a cultivation-independent survey based on PCR-amplified 16S rRNA genes. FEMS Microbiol. Ecol. 49:217-227.[CrossRef]
<A>Czarnetzki, A. B., and C. C. Tebbe.</A> <Y>2004.</Y> <T>Diversity of bacteria associated with Collembola: a cultivation-independent survey based on PCR-amplified 16S rRNA genes.</T> <J>FEMS Microbiol. Ecol.</J> <V>49:</V> <P>217-227.</P> <O>[CrossRef]</O>
(d) Rios R, Carneiro I, Arce VM, and Devesa J. Myostatin is an inhibitor of myogenic differentiation. Am J Physiol Cell Physiol 282: C993–C999, 2002. [Abstract/Free Full Text]
<A>Rios R, Carneiro I, Arce VM, and Devesa J.</A> <T>Myostatin is an inhibitor of myogenic differentiation.</T> <J>Am J Physiol Cell Physiol</J> <V>282:</V> <P>C993–C999,</P> <Y>2002.</Y> <O>[Abstract/Free Full Text]</O>
(e) 12. T.J. McCarthy et al., Chem. Biol. 12, 1221 (2005). [CrossRef] [ISI] [Medline]
<N>12.</N> <A>T.J. McCarthy et al.,</A> <J>Chem. Biol.</J> <V>12,</V> <P>1221</P > <Y>(2005).</Y> <O>[CrossRef] [ISI] [Medline]</O>
(f) 18 J. Cavanagh, W.J. Fairbrother, A.G. Palmer and N.J. Skelton, Protein NMR Spectroscopy, Academic Press, San Diego, CA (1996).
<N>18</N> <A>J. Cavanagh, W.J. Fairbrother, A.G. Palmer and N.J. Skelton,</A> <J>Protein NMR Spectroscopy,</J> <O>Academic Press, San Diego, CA</O> <Y>(1996)</Y>
(g) Anonymous. 2005. Microbiology of food and animal feeding stuffs. Polymerase chain reaction (PCR) for the detection of food-borne pathogens. Requirements for amplification and detection for qualitative methods. Draft International Standard ISO/FDIS 20838:2005. DIN, Berlin, Germany.
<A>Anonymous.</A> <Y>2005.</Y> <T>Microbiology of food animal feeding stuffs. Polymerase chain reaction (PCR) for the detection of food-borne pathogens. Requirements for amplification detection for qualitative methods.</T> <O>Draft International Standard ISO/FDIS 20838</O> <Y>2005.</Y> <O>DIN, Berlin, Germany.</O>
Early research in reference parsing involved rule-based methods, which usually depend on knowledge that is manually crafted and based on a domain expert’s observation. This domain knowledge is organized as templates or hierarchical frameworks, which summarize the recognizable patterns formed by the data or the surrounding text, and the rules associated with those recognizable patterns. After the knowledge representation is built, various algorithms can be used to match the text to the knowledge representation, and to extract data according to the rules. These matching algorithms include template mining [7, 8], INFOMAP [9, 10] and BLAST (Basic Local Alignment Search Tool), a tool originally designed for gene sequence alignment .
Rule-based methods can be very successful when the references are from a small or moderate number of journals. This is because journal publishers usually require authors to strictly follow predefined citation styles, conduct careful editorial checking and correction before publishing. However, when a large number of journals are involved, it can be very challenging to build a sound knowledge representation due to the large variety of, and sometimes conflicting, citation styles. Rule-based methods also require domain experts to design the rules and maintain them over time, and therefore lack adaptability and are difficult to tune.
Machine learning approaches have recently attracted increased attention because they automatically learn the knowledge from training samples and therefore exhibit good adaptability. For example, Parmentier and Belaïd have developed a concept network to hierarchically represent and recognize structured data from bibliographic citations . Besagni et al. took a bottom-up approach based on Part-of-Speech (PoS) tagging . Basic tags, which are easily recognized, are first grouped into homogeneous classes. Confusing tokens are then classified by either a set of PoS correction rules or a structure model generated from well-detected tokens.
Reference parsing is essentially a sequence processing task and therefore statistical sequence models, e.g., Hidden Markov Model (HMM) and Conditional Random Field (CRF), as successful machine learning tools for information retrieval, have also been studied for parsing references. For example, Takasu applied HMM for metadata extraction from erroneous references . Another frequently adopted machine learning method for information extraction is the Support Vector Machine (SVM) classifier. Okada et al. combined SVM and HMM for bibliographic component extraction . In our previous research, we developed and compared a SVM-based method with one based on CRF .
Since collecting ground-truth training samples can be labor-intensive, unsupervised approaches have also been proposed. For example, Cortze et al. proposed an unsupervised approach, called FLUX-CiM, which is based on a frequency-tuned lexicon and includes four stages: blocking, matching, binding and joining .
As pointed out in a recent article, despite over a decade of research, reference parsing is still an unsolved task for several reasons, including data-entry errors, the wide variability of citation formats, lack of (or enforcement of) standards, large-scale citation data, and so on .
In this paper, we describe an extension of our previous work on reference parsing, reported in . We adopted the recently proposed structural SVM method and compared it to conventional SVM. Our experiments on 1800 ground-truth labeled references show that the structural SVM method achieves over 98% token-level accuracy and over 95% chunk-level accuracy. In addition, we compared SVM and structural SVM to Conditional Random Field (CRF), another state-of-the-art sequence learning method. We observe that structural SVM and CRF achieve about the same accuracies at token- and chunk-levels. Both methods show the advantage of stronger sequence learning ability over SVM.
Mathematical description of structural SVM
Structural Support Vector Machine (Structural SVM), introduced by Tsochantaridis et al., is a supervised learning method designed for predicting complex structured outputs, such as sequences, trees and graphs . Given a training sample of input-output pairs (x1,y1),…(x n ,y n )∈ X ×Y drawn from an unknown distribution, structural SVM addresses the general problem of learning a mapping f : X → Y from input patterns x ∈ X to discrete outputs y ∈ Y that has low prediction errors. The idea is to learn a discriminant function F from which we can derive a prediction by maximizing F over Y given a specific input is a linear combination of some joint feature representations of inputs and outputs, where w is a parameter vector and ψ is a feature vector relating x and y. The flexibility in designing ψ allows structural SVM to model many problems as diverse as natural language parsing, multiclass classification, sequence learning, etc.
The constraints are built upon the condition that given a training sample (x i ,y i ,), the value of for the correct prediction y i should be greater than those for all other incorrect predictions y . Each training sample is associated with |y| - 1 constraints which share the same slack variable ξ i . The introduction of ξ i allows structural SVM to learn a large soft margin with small misclassification errors, which makes structural SVM more general to solve those classification problems where different classes are not strictly separable even in high feature space. The objective function is penalized by adding non-zero slack variables, ξ i , each of which measures the degree of misclassification of a sample x i . Therefore, the optimization becomes a trade-off between a large margin and a small error penalty. ∑ξ i gives an upper bound for the empirical risk on the training set, and the constant C is a regularization term that controls the trade-off between training error minimization and margin maximization. Training structural SVM is computationally expensive due to the large number of margin constraints. By an equivalent 1-slack reformulation of the n-slack structural SVM, Joachims et al. proposed a “l-slack cutting-plane” method which significantly reduces the computation time, thereby making the training on large databases feasible . Both SVM and structural SVM are discriminative models. They learn optimal linear-separable hyperplanes with maximum-margin between classes. Structural SVM conducts global optimization on the whole structure, while SVM optimizes locally on individual tokens. Structural SVM is more general than SVM in its capability of learning interdependent and structured outputs. It has shown promising results for building highly complex, but still accurate discriminative models in the areas of classification with taxonomies, protein sequence alignment, and natural language context-free grammar parsing.
Features extracted from each token in a reference
1.Author Name Feature
Is the word in Author Name dictionary?
2. Article Title Feature
Is the word in Article Title dictionary?
3. Journal Title Feature
Is the word in Journal Title dictionary?
4. Pagination Pattern
Is the word in pagination formation, e.g., 200-5, H100-H105?
5. Name Initial Pattern
Is the word in name initial pattern, e.g., J.Z., J.-Z.?
6. Four Digit Year Pattern
Is the word in four digit year pattern, e.g., 2005? It must be not before 1500, and not later than the current year.
7. et, al
Is the word “et” or “al”, or “et.”, or “al.”?
8. pp., p.
Is the word “pp.”, or “p.”, or “pp”, or “p”?
9. Ended With “.”
Does the word end with “.”?
10. Upper Case First Char
Is the first character of the word upper case?
11. Letter Only
Does the word contain letters only?
12. Digit Only
Does the word contain digits only?
13. Digit and Letter
Does the word contain both digits and letters?
14. Digit and Letter Only
Does the word contain digits and letters only?
15. Normalized position
The position of the word normalized by the total number of words in the reference.
Features from neighboring tokens are very informative as they exploit the contextual dependencies between tokens. There are two kinds of contextual features: the observation features extracted from the neighboring tokens and the labels assigned to those tokens. We call the first one “contextual observation features” and the second “contextual label features”. Since in reference parsing, structural SVM is implemented as a sequence learning algorithm, the joint feature presentation function ψ(x, y) includes two kinds of features: state transition features and observation features extracted from individual tokens within a sequence. State transition features utilize contextual label information to model the dependencies between adjacent labels. Having these similar types of feature representations as Hidden Markov Models, structural SVM designed specifically for sequence labeling is sometimes called SVMHMM. In addition to contextual label features, we also combine contextual observation features from neighboring tokens for sequence classification.
Results and discussion
The total number of words and chunks for each of the 8 entities in references for evaluation
Total number of words
Total number of chunks
Evaluation of structural SVM
For our experiments, we use the SVMHMM library, an implementation of structural SVM for sequence labeling , and the linear-kernel since other kernels, e.g. radial basic function (RBF), can be extremely computation intensive. To compare this with SVM, we use LibSVM , a library developed at National Taiwan University for word classification. Here linear kernel function is also adopted to facilitate a fair comparison. All the meta-parameters in both SVM and structural SVM are determined with cross-validation on training samples.
Token classification accuracy obtained by SVM and structural SVM
Features from token itself (15 features)
Features from the token and its two neighbors (45)
Features from the token and its four neighbors (75)
Chunk-level accuracies of SVM method
Features from token itself
Features from the token and its two neighbors
Features from the token and its four neighbors
Chunk-level accuracies of structural SVM method
Features from token itself
Features from the token and its two neighbors
Features from the token and its four neighbors
We first use only the 15 observation features from the token itself. Since SVM does not use contextual features, it provides a baseline performance by analyzing only the token itself. As expected, the performance is relatively low: the token classification accuracy is 93.03% and the overall chunk accuracy is only 79.12%. Although structural SVM does not use contextual observation features, it does use the contextual label features. The overall accuracies at token-level and chunk-level are 98.41% and 95.35%, respectively, which are much better than those of the SVM method. This clearly indicates the value of contextual label features in structural SVM.
We then add the observation features from the immediate left and right neighbors (the first order contextual observation features). The corresponding token classification accuracy and overall chunk-level accuracy of SVM significantly increase to 98.20% and 94.27%. This indicates that the first order contextual observation features are very important for SVM classification. After combining observation features from one further left and one further right neighbors, the corresponding token-level and chunk-level accuracies increase to 98.65% and 95.59%. This indicates that the second order contextual observation features are still helpful, but less so than the first order ones. For the structural SVM method, when the first order contextual observation features are added, the overall accuracies at token-level and chunk-level increase to 98.91% and 96.81%, respectively. The accuracy improvement is not so substantial as that compared to the SVM method, which may imply that the contextual observation features and contextual label features share redundant discriminative information. After including the second order contextual observation features, there is virtually no performance gain for the structural SVM method, even though it uses extra contextual label features.
Comparison with Conditional Random Field (CRF)
We also compared our methods to CRF, another state-of-the-art sequence learning algorithm . Because only binary features can be used in CRF models, we removed the normalized position feature from the feature vector used in previous evaluation. We then repeated some experiments using the same set of binary features in SVM, structural SVM and CRF methods for fair comparisons. We use SimpleTagger, a sequence tagging tool for CRF implementation in MALLET  for our CRF experiments.
Token classification accuracy obtained by SVM, structural SVM and CRF
Features from the token and its four neighbors (70 features)
Chunk-level accuracies of SVM, structural SVM and CRF
The accuracies in CRF experiments are a little different from those reported in . That is because in , additional large number of word features is extracted from each token and used in the classification. Adding those word features significantly increases the feature dimensionality, which causes difficulties in training SVM and structural SVM. On the other hand, adding those thousands of word features in CRF improves accuracy only slightly, indicating the non-importance of word features. Basically, we use the first 14 binary features described in Table 2 for a fair comparison.
We have compared SVM and structural SVM as methods for parsing references that appear in medical journal articles. One important difference between the two methods is that the SVM uses only the contextual observation features, while structural SVM uses these as well as contextual label features. Although SVM performance improves greatly and is close to that of structural SVM when the second order contextual observation features are used, structural SVM achieves higher overall token-level and chunk-level accuracies than the SVM method. Both methods achieve above 98% token classification accuracy and an overall chunk-level accuracy of over 95%. Compared to the CRF, we find that the structural SVM achieves similar performance. However, both methods perform better than SVM, showing the advantage of their stronger sequence learning ability.
Reference parsing is considered a sequence learning problem due to the strong regular internal structure in each reference. Additionally, we note that references cited in any one article generally follow the same style. Further exploiting this consistency in consecutive references to improve the performance of reference parsing will be the subject of future work.
This research was supported by the Intramural Research Program of the National Institutes of Health (NIH), National Library of Medicine, and Lister Hill National Center for Biomedical Communications.
This article has been published as part of BMC Bioinformatics Volume 12 Supplement 3, 2011: Machine Learning for Biomedical Literature Analysis and Text Retrieval. The full contents of the supplement are available online at http://www.biomedcentral.com/1471-2105/12?issue=S3.
- Lawrence S, Giles CL, Bollacker K: Digital libraries and autonomous citation indexing. IEEE Computer. 1999, vol. 32, 6: 67-71.View ArticleGoogle Scholar
- ISI Web of Knowledge. [http://www.isiwebofknowledge.com/]
- Google Scholar. [http://scholar.google.com/]
- Lee D, Kang J, Mitra P, Giles CL, On BW: Are your citations clean?. Communications of the ACM. 2007, 50 (12): 33-38. 10.1145/1323688.1323690.View ArticleGoogle Scholar
- Kim I, Le DX, Thoma GR: Identification of "comment-on sentences" in online biomedical documents using support vector machines. Proc. of SPIE conference on Document Recognition and Retrieval. 2007, 68150X: 1-9.Google Scholar
- Aronson AR, Bodenreider O, Chang HF, Humphrey SM, Mork JG, Nelson SJ, Rindflesch TC, Wilbur WJ: The NLM indexing initiative. Proc. of AMIA Symp. 2000, 17-21.Google Scholar
- Chowdhury G: Template mining for information extraction from digital documents. Library Trends. 1999, 48 (1): 182-208.Google Scholar
- Ding Y, Chowdhury G, Foo S: Template mining for the extraction of citation from digital documents. Proc. of the 2nd Asian Digital Library Conference. 1999, 47-62.Google Scholar
- Day MY, Tsai TH, Sung CL, Lee CW, Wu SH, Ong CS, Hsu WL: A knowledge-based approach to citation extraction. IEEE Int’l Conf. on Information Reuse and Integration. 2005, 50-55.Google Scholar
- Day MY, Tsai TH, Sung CL, Hsieh CC, Lee CW, Wu SH, Wu KP, Ong CS, Hsu WL: Reference metadata extraction using a hierarchical knowledge representation framework. Decision Support Systems. 2007, 43 (1): 152-167. 10.1016/j.dss.2006.08.006.View ArticleGoogle Scholar
- Huang IA, Ho JM, Kao HY, Lin WC: Extracting citation metadata from online publication lists using BLAST. Proc. of the Eighth Pacific-Asia Conference on Knowledge Discovery and Data Mining. 2004, 26-28.Google Scholar
- Parmentier F, Belaïd A: Logical structure recognition of scientific bibliographic references. Proc. of ICDAR. 1997, 2: 1072-1076.Google Scholar
- Besagni D, Belaïd A, Benet N: A segmentation method for bibliographic references by contextual tagging of fields. Proc. of ICDAR. 2003, 1: 384-388.Google Scholar
- Takasu A: Bibliographic attribute extraction from erroneous references based on a statistical model. Proc. of JCDL. 2003, 49-60.Google Scholar
- Okada T, Takasu A, Adachi J: Bibliographic component extraction using support vector machines and Hidden Markov Models. Proc. of ECDL. 2004, 501-512.Google Scholar
- Zou J, Le DX, Thoma GR: Locating and parsing bibliographical references in the HTML medical journal articles. International Journal on Document Analysis and Recognition. 2010, 13 (2): 107-119. 10.1007/s10032-009-0105-9.PubMed CentralView ArticlePubMedGoogle Scholar
- Cortez E, da Silva AS, Goncalves MA, Mesquita F, de Moura ES: A flexible approach for extracting metadata from bibliographic citations. Journal of the American Society for Information Science and Technology. 2009, 60 (6): 1144-1158. 10.1002/asi.21049.View ArticleGoogle Scholar
- Councill IG, Giles CL, Kan KY: ParsCit: an open-source CRF reference string parsing package. Proc. of the Language Resources and Evaluation Conference(LREC08). 2008, [http://wing.comp.nus.edu.sg/parsCit/]Google Scholar
- FreeCite. [http://freecite.library.brown.edu/welcome]
- Tsochantaridis I, Hofmann T, Joachims T, Altun Y: Support vector machine learning for interdependent and structured output spaces. Int’l Conf. on Machine Learning(ICML). 2004, 104-112.Google Scholar
- Joachims T, Finley T, Yu CN: Cutting-plane training of structural SVMs. Machine Learning Journal. 2009, 77 (1): 27-59. 10.1007/s10994-009-5108-8.View ArticleGoogle Scholar
- Herbst E, Joachims T: SVMHMM: sequence tagging with structural support vector machine. 2008, [http://www.cs.cornell.edu/People/tj/svm_light/svm_hmm.html]Google Scholar
- Chang CC, Lin CJ: LIBSVM: a library for support vector machines. 2001, Software available at http://www.csie.ntu.edu.tw/~cjlin/libsvmGoogle Scholar
- Lafferty J, McCallum A, Pereira F: Conditional random fields: probabilistic models for segmenting and labeling sequence data. Proc. of ICML. 2010, 282-289.Google Scholar
- McCallum AK: MALLET: a machine learning for language toolkit. 2002, [http://mallet.cs.umass.edu/index.php]Google Scholar
This article is published under license to BioMed Central Ltd. This article is in the public domain.