The source retrieval is a subtask in a plagiarism detection process during\r
which only a relatively small subset of documents are retrieved from the\r
large corpus. Those candidate documents are usually further compared in detail with the\r
-suspicious document. In the PAN 2013 source retrieval subtask the main goal was to\r
+suspicious document. In PAN 2013 source retrieval subtask the main goal was to\r
identify web pages which have been used as a source of plagiarism for test corpus creation.\r
\r
The test corpus contained 58 documents each discussing only one theme.\r
\r
%In the real-world scenario the corpus is the whole Web and the search engine can be a contemporary commercial search engine which scales to the size of the Web.\r
\r
-The same methodology -- utilizing a search engine is use for the source retrieval. \r
+The same methodology -- utilizing a search engine; is used for source retrieval. \r
This approach is based on the fact that we do not\r
possess enough resources to download and effectively process the whole corpus.\r
In the case of PAN 2013 competition we utilized the ChatNoir~\cite{chatnoir} \r
and in decision what to actually download and what to report as a plagiarism case from the search results.\r
\r
These first two stages are shown in Figure~\ref{fig:source_retr_process} as Querying and Selecting. Selected results \r
-from the search engine are forthwith textually aligned with the suspicious document (see section~\ref{text_alignment} for more details).\r
+from the search engine are then textually aligned with the suspicious document (see section~\ref{text_alignment} for more details).\r
%This is the last decision phase -- what to report.\r
If there is any continuous passage of reused text detected, the result document is reported\r
and the continuous passages in the suspicious document are marked as {\it discovered} and no further processing\r
i) keywords based queries, ii) intrinsic plagiarism\r
based queries, and iii) paragraph based queries. Three main properties distinguish each type of query: i) Positional; ii) Phrasal; iii) Deterministic.\r
Positional queries carry extra information about a textual interval in the suspicious document which the query represents.\r
-A phrasal query aims for retrieval of documents containing the same small piece of a text. They are usually created from closely coupled words. \r
+A phrasal query aims for retrieval of documents containing the same small piece of text. They are usually created from closely coupled words. \r
Deterministic queries for specific suspicious document are always the same no matter how many times we run the software. \r
%On the contrary the software can create in two runs potentially different nondeterministic queries.\r
\r
Their purpose is to retrieve documents concerning the same theme.\r
%Two documents discussing the same theme usually share a set of overlapping keywords. Also the combination of keywords in query matters. \r
As a method for automated keywords extraction, we used a frequency based approach described in~\cite{suchomel_kas_12}.\r
-The method combines term frequency analysis with TF-IDF score~\cite{Introduction_to_information_retrieval}. As a reference\r
-corpus we used English web corpus~\cite{ententen} crawled by SpiderLink~\cite{SpiderLink} in 2012 which contains 4.65 billion tokens. \r
+The method combines term frequency analysis with TF-IDF score.\r
+As a reference corpus we used English web corpus~\cite{ententen} crawled by SpiderLink~\cite{SpiderLink} in 2012 which contains 4.65 billion tokens. \r
\r
-Each keywords based query was constructed from five top ranked keywords consecutively. Each keyword was\r
-used only in one query. Too long keywords based queries would be overspecific and it would have resulted\r
-in a low recall. On the other hand having constructed too short queries (one or two tokens) would have resulted\r
-in a low precision and also possibly low recall since they would be too general.\r
+Each keywords based query was constructed from five top ranked keywords consecutively.\r
+Each keyword was used only in one query. \r
+%Too long keywords based queries would be overspecific and it would have resulted in a low recall.\r
+%On the other hand having constructed too short queries (one or two tokens) would have resulted in a low precision and also possibly low recall since they would be too general.\r
In order to direct the search more at the highest ranked keywords we also extracted their \r
-most frequent two and three term long collocations. These were combined also into queries of 5 words.\r
+most frequent two and three term long collocations.\r
+These were combined also into queries of 5 words.\r
Resulting the 4 top ranked keywords alone can appear in two different queries, one from the keywords\r
alone and one from the collocations.\r
%Collocation describes its keyword better than the keyword alone. \r
\subsubsection{Intrinsic Plagiarism Based Queries.}\r
The second type of queries purpose to retrieve pages which contain text detected\r
as different, in a manner of writing style, from other parts of the suspicious document.\r
-Such a change may point out plagiarized passage which is intrinsically bound up with the text. \r
-We implemented vocabulary richness method which computes average word frequency class value for \r
-a given text part. The method is described in~\cite{awfc}.\r
+%Such a change may point out plagiarized passage which is intrinsically bound up with the text. \r
+%We implemented vocabulary richness method which computes average word frequency class value for \r
+%a given text part. The method is described in~\cite{awfc}.\r
+For this purpose we implemented vocabulary richness method~\cite{awfc} together with\r
+sliding windows concept for text chunking as described in~\cite{suchomel_kas_12}.\r
%The problem is that generally methods based on the vocabulary statistics work better for longer texts.\r
%According to authors this method scales well for shorter texts than other text style detection methods. \r
-The usage of this method is in our case limited by relatively short texts.\r
-It is also difficult to determine\r
-what parts of text to compare. Therefore we used sliding window concept for text chunking with the \r
-same settings as described in~\cite{suchomel_kas_12}.\r
+%The usage of this method is in our case limited by relatively short texts.\r
+%It is also difficult to determine\r
+%what parts of text to compare. Therefore we used sliding window concept for text chunking with the \r
+%same settings as described in~\cite{suchomel_kas_12}.\r
\r
A representative sentence longer than 6 words was randomly selected among those that apply from the suspicious part of the document.\r
The query was created from the representative sentence leaving out stop words.\r
-The intrinsic plagiarism based queries are positional. They carry the position of the representative sentence in the document.\r
+The intrinsic plagiarism based queries are positional. They carry the position of the representative sentence.% in the document.\r
They are phrasal, since they represent a search for a specific sentence. And they are\r
nondeterministic, because the representative sentence is selected randomly. \r
\r
\subsection{Search Control}\r
For each suspicious document we prepared all three types of queries during the first phase at once.\r
Queries were executed stepwise. \r
-After processing each query the results were evaluated %(see the following subsection~\ref{resSelection} for more details)\r
-and from \r
-all textual similarities between each result and the suspicious document the suspicious document intervals of those similarities\r
+After processing each query the results were evaluated. %(see the following subsection~\ref{resSelection} for more details)\r
+From all textual similarities between each result and the suspicious document, the document intervals of those similarities\r
were marked as {\it discovered}. \r
+\r
%At first there were processed the keywords based queries.\r
Firstly, there were always all of the keywords based queries executed. \r
%After having all the keywords based queries processed,\r
Secondly the intrinsic plagiarism based queries were executed according to \r
their creation sequence. \r
-Since they carry its position, not all of them were carried out.\r
+%Since they carry its position, not all of them were carried out.\r
During the execution, if any of the query position intersected with any of the {\it discovered} interval, the\r
query was dropped out. Analogically, the last there were paragraph based queries processed. \r
\r
This search control results in two major properties. Firstly, the source retrieval effort was increased \r
in parts of the suspicious document, where there has not yet been found any textual similarity.\r
-It was increased especially by the paragraph based queries. And secondly, after detection a similarity for a certain part of the text,\r
+%It was increased especially by the paragraph based queries.\r
+And secondly, after detection a similarity for a certain part of the text,\r
no more intentionally retrieval attempts for that part were effectuated. Meaning that all\r
discovered search engine results were evaluated, but there were executed no more queries regarding that passage.\r
\r
\r
\subsection{Result Selection}~\label{resSelection}\r
The second main decisive area about source retrieval task is to decide which from the search engine results to download.\r
-This process is represented in Figure~\ref{fig:source_retr_process} as 'Selecting'. \r
-Nowadays in the real-world a download is cheap and quick operation.\r
+This process is represented in Figure~\ref{fig:source_retr_process} as Selecting. \r
+%Nowadays in the real-world a download is cheap and quick operation.\r
%There can be some disk space considerations if there is a need to store original downloaded documents.\r
- The main cost probably represents a document post processing. \r
-Mainly on the Internet there is a wide range of file formats, which for text alignment must be\r
-converted into plaintext. This can be time and computational-consuming.\r
-For example from many pdf documents the plain text is hardly extractable, thus one need to use optical character recognition methods.\r
+% The main cost probably represents document post processing. \r
+%Mainly on the Internet there is a wide range of file formats, which for text alignment must be\r
+%converted into plaintext. This can be time and computational-consuming.\r
+%For example from many pdf documents the plain text is hardly extractable, thus one need to use optical character recognition methods.\r
\r
The ChatNoir offers snippets for discovered documents. The snippet generation is considered costless\r
operation. The snippet purpose is to have a quick glance at a small extract of resulting page.\r
Our approach had the second best ratio of recall to the number of used queries, which\r
tells about the query efficacy. The approach with the best ratio used few queries (4.9 queries per document which\r
was 0.4 of the amount we used), but also obtained the lowest recall (0.65 of our recall).\r
-The approach with highest recall (and also lowest precision) achieved 2.8 times higher recall with 3.9 times more queries compared to ours.\r
+The approach with the highest recall (and also lowest precision) achieved 2.8 times higher recall with 3.9 times more queries compared to ours.\r
\r
Our approach achieved also low precision, which means we reported many more results and they\r
were not considered as correct hits. On the other hand each reported result contained some\r
https://www.fi.muni.cz/~kas/git / pan13-paper.git / blobdiff