From: Jan "Yenya" Kasprzak Date: Fri, 10 Aug 2012 13:36:49 +0000 (+0200) Subject: yenya-detailed.tex: odeslana verze x-abstractu X-Git-Url: https://www.fi.muni.cz/~kas/git//home/kas/public_html/git/?a=commitdiff_plain;ds=sidebyside;h=fe99c4d71399c862a7a6db9f539849d94e2a6dcd;p=pan12-paper.git yenya-detailed.tex: odeslana verze x-abstractu Puvodne tam byla nejaka starsi verze. --- diff --git a/yenya-detailed.tex b/yenya-detailed.tex index 3615ab9..6e4ef03 100644 --- a/yenya-detailed.tex +++ b/yenya-detailed.tex @@ -1,49 +1,48 @@ \section{Detailed Document Comparison} -\subsection{General Approach} - -The approach Masaryk University team has used in PAN 2012 Plagiarism -detection---detailed comparison sub-task is based on the same approach -that we have used in PAN 2010 \cite{Kasprzak2010}. This time, we have -used a similar approach, enhanced by several means - -The algorithm evaluates the document pair in several stages: - -\begin{itemize} -\item intrinsic plagiarism detection -\item language detection of the source document -\begin{itemize} -\item cross-lingual plagiarism detection, if the source document is not in English -\end{itemize} -\item detecting intervals with common features -\item post-processing phase, mainly serves for merging the nearby common intervals -\end{itemize} - -\subsection{Intrinsic plagiarism detection} - -Our approach is based on character $n$-gram profiles of the interval of -the fixed size (in terms of $n$-grams), and their differences to the -profile of the whole document \cite{pan09stamatatos}. We have further -enhanced the approach with using gaussian smoothing of the style-change -function \cite{Kasprzak2010}. - -For PAN 2012, we have experimented with using 1-, 2-, and 3-grams instead -of only 3-grams, and using the different measure of the difference between -the n-gram profiles. We have used an approach similar to \cite{ngram}, -where we have compute the profile as an ordered set of 400 most-frequent -$n$-grams in a given text (the whole document or a partial window). Apart -from ordering the set we have ignored the actual number of occurrences -of a given $n$-gram altogether, and used the value inveresly -proportional to the $n$-gram order in the profile, in accordance with -the Zipf's law \cite{zipf1935psycho}. -This approach has provided more stable style-change function than -than the one proposed in \cite{pan09stamatatos}. Because of pair-wise -nature of the detailed comparison sub-task, we couldn't use the results -of the intrinsic detection immediately, so we wanted to use them -as hints to the external detection. +\subsection{General Approach} -\subsection{Cross-lingual detection} +Our approach in PAN 2012 Plagiarism detection---Detailed comparison sub-task +is loosely based on the approach we have used in PAN 2010 \cite{Kasprzak2010}. + +%The algorithm evaluates the document pair in several stages: +% +%\begin{itemize} +%\item intrinsic plagiarism detection +%\item language detection of the source document +%\begin{itemize} +%\item cross-lingual plagiarism detection, if the source document is not in English +%\end{itemize} +%\item detecting intervals with common features +%\item post-processing phase, mainly serves for merging the nearby common intervals +%\end{itemize} + +%\subsection{Intrinsic plagiarism detection} +% +%Our approach is based on character $n$-gram profiles of the interval of +%the fixed size (in terms of $n$-grams), and their differences to the +%profile of the whole document \cite{pan09stamatatos}. We have further +%enhanced the approach with using gaussian smoothing of the style-change +%function \cite{Kasprzak2010}. +% +%For PAN 2012, we have experimented with using 1-, 2-, and 3-grams instead +%of only 3-grams, and using the different measure of the difference between +%the n-gram profiles. We have used an approach similar to \cite{ngram}, +%where we have compute the profile as an ordered set of 400 most-frequent +%$n$-grams in a given text (the whole document or a partial window). Apart +%from ordering the set, we have ignored the actual number of occurrences +%of a given $n$-gram altogether, and used the value inveresly +%proportional to the $n$-gram order in the profile, in accordance with +%the Zipf's law \cite{zipf1935psycho}. +% +%This approach has provided more stable style-change function than +%than the one proposed in \cite{pan09stamatatos}. Because of pair-wise +%nature of the detailed comparison sub-task, we couldn't use the results +%of the intrinsic detection immediately, therefore we wanted to use them +%as hints to the external detection. + +\subsection{Cross-lingual Plagiarism Detection} %For language detection, we used the $n$-gram based categorization \cite{ngram}. %We have computed the language profiles from the source documents of the @@ -54,97 +53,130 @@ as hints to the external detection. %an ad-hoc fix, where for documents having their profile too distant from all of %English, German, and Spanish profiles, we have declared them to be in English. -For cross-lingual plagiarism detection, our aim was to use the public -interface to Google translate if possible, and use the resulting document -as the source for standard intra-lingual detector. -Should the translation service not be available, we wanted -to use the fall-back strategy of translating isolated words only, -with the additional exact matching of longer words (we have used words with -5 characters or longer). -We have supposed these longer words can be names or specialized terms, -present in both languages. - -We have used dictionaries from several sources, like -{\tt dicts.info\footnote{\url{http://www.dicts.info/}}}, -{\tt omegawiki\footnote{\url{http://www.omegawiki.org/}}}, -and {\tt wiktionary\footnote{\url{http://en.wiktionary.org/}}}. The source -and translated document were aligned on a line-by-line basis. +%For cross-lingual plagiarism detection, our aim was to use the public +%interface to Google translate if possible, and use the resulting document +%as the source for standard intra-lingual detector. +%Should the translation service not be available, we wanted +%to use the fall-back strategy of translating isolated words only, +%with the additional exact matching of longer words (we have used words with +%5 characters or longer). +%We have supposed that these longer words can be names or specialized terms, +%present in both languages. + +%We have used dictionaries from several sources, like +%{\it dicts.info}\footnote{\url{http://www.dicts.info/}}, +%{\it omegawiki}\footnote{\url{http://www.omegawiki.org/}}, +%and {\it wiktionary}\footnote{\url{http://en.wiktionary.org/}}. The source +%and translated document were aligned on a line-by-line basis. In the final form of the detailed comparison sub-task, the results of machine translation of the source documents were provided to the detector programs by the surrounding environment, so we have discarded the language detection and machine translation from our submission altogether, and used only line-by-line alignment of the source and translated document for calculating -the offsets of text features in the source document. +the offsets of text features in the source document. We have then treated +the translated documents the same way as the source documents in English. \subsection{Multi-feature Plagiarism Detection} Our pair-wise plagiarism detection is based on finding common passages -of text, present both in the source and suspicious document. We call them -{\it features}. In PAN 2010, we have used sorted word 5-grams, formed from +of text, present both in the source and in the suspicious document. We call them +{\it common features}. In PAN 2010, we have used sorted word 5-grams, formed from words of three or more characters, as features to compare. Recently, other means of plagiarism detection have been explored: -Stop-word $n$-gram detection is one of them +stopword $n$-gram detection is one of them \cite{stamatatos2011plagiarism}. We propose the plagiarism detection system based on detecting common -features of various type, like word $n$-grams, stopword $n$-grams, -translated words or word bigrams, exact common longer words from document -pairs having each document in a different language, etc. The system +features of various types, for example word $n$-grams, stopword $n$-grams, +translated single words, translated word bigrams, +exact common longer words from document pairs having each document +in a different language, etc. The system has to be to the great extent independent of the specialities of various feature types. It cannot, for example, use the order of given features as a measure of distance between the features, as for example, several word 5-grams can be fully contained inside one stopword 8-gram. -We thus define {\it common feature} of two documents (susp and src) -as the following tuple: -$$\langle +We therefore propose to describe the {\it common feature} of two documents +(susp and src) with the following tuple: +$\langle \hbox{offset}_{\hbox{susp}}, \hbox{length}_{\hbox{susp}}, \hbox{offset}_{\hbox{src}}, -\hbox{length}_{\hbox{src}} \rangle$$ - -In our final submission, we have used only the following two types +\hbox{length}_{\hbox{src}} \rangle$. This way, the common feature is +described purely in terms of character offsets, belonging to the feature +in both documents. In our final submission, we have used the following two types of common features: \begin{itemize} \item word 5-grams, from words of three or more characters, sorted, lowercased -\item stop-word 8-grams, from 50 most-frequent English words (including +\item stopword 8-grams, from 50 most-frequent English words (including the possessive suffix 's), unsorted, lowercased, with 8-grams formed only from the seven most-frequent words ({\it the, of, a, in, to, 's}) removed \end{itemize} -We have gathered all the common features for a given document pair, and formed -{\it valid intervals} from them, as described in \cite{Kasprzak2009a} -(a similar approach is used also in \cite{stamatatos2011plagiarism}). +We have gathered all the common features of both types for a given document +pair, and formed {\it valid intervals} from them, as described +in \cite{Kasprzak2009a}. A similar approach is used also in +\cite{stamatatos2011plagiarism}. The algorithm is modified for multi-feature detection to use character offsets only instead of feature order numbers. We have used valid intervals consisting of at least 5 common features, with the maximum allowed gap inside the interval (characters not belonging to any common feature of a given valid interval) set to 3,500 characters. -We have also experimented with modifying the allowed gap size using the -intrinsic plagiarism detection: to allow only shorter gap if the common -features around the gap belong to different passages, detected as plagiarized -in the suspicious document by the intrinsic detector, and allow larger gap, -if both the surrounding common features belong to the same passage, -detected by the intrinsic detector. This approach, however, did not show -any improvement against allowed gap of a static size, so it was omitted -from the final submission. +%We have also experimented with modifying the allowed gap size using the +%intrinsic plagiarism detection: to allow only shorter gap if the common +%features around the gap belong to different passages, detected as plagiarized +%in the suspicious document by the intrinsic detector, and allow larger gap, +%if both the surrounding common features belong to the same passage, +%detected by the intrinsic detector. This approach, however, did not show +%any improvement against allowed gap of a static size, so it was omitted +%from the final submission. \subsection{Postprocessing} +In the postprocessing phase, we took the resulting valid intervals, +and made attempt to further improve the results. We have firstly +removed overlaps: if both overlapping intervals were +shorter than 300 characters, we have removed both of them. Otherwise, we +kept the longer detection (longer in terms of length in the suspicious document). + +We have then joined the adjacent valid intervals into one detection, +if at least one of the following criteria has been met: +\begin{itemize} +\item the gap between the intervals contained at least 4 common features, +and it contained at least one feature per 10,000 +characters\footnote{we have computed the length of the gap as the number +of characters between the detections in the source document, plus the +number of charaters between the detections in the suspicious document.}, or +\item the gap was smaller than 30,000 characters and the size of the adjacent +valid intervals was at least twice as big as the gap between them, or +\item the gap was smaller than 30,000 characters and the number of common +features per character in the adjacent interval was not more than three times +bigger than number of features per character in the possible joined interval. +\end{itemize} + +These parameters were fine-tuned to achieve the best results on the training corpus. With these parameters, our algorithm got the total plagdet score of 0.73 on the training corpus. \subsection{Further discussion} +As in our PAN 2010 submission, we tried to make use of the intrinsic plagiarism +detection, but despite making further improvements to the intrinsic plagiarism detector, we have again failed to reach any significant improvement +when using it as a hint for the external plagiarism detection. + In the full paper, we will also discuss the following topics: \begin{itemize} -\item language detection +\item language detection and cross-language common features +\item intrinsic plagiarism detection \item suitability of plagdet score\cite{potthastframework} for performance measurement \item feasibility of our approach in large-scale systems -\item other possible features to use, especially for cross-lingual detection \item discussion of parameter settings \end{itemize} +\nocite{pan09stamatatos} +%\nocite{ngram} + +