\section{Candidate document retrieval}~\label{simon}
The basic concept of candidate document retrieval is to use a web search
-engine to find suitable documents. In PAN 12 competition we used ChatNoir search
+engine to find suitable documents. In PAN 2012 competition we used ChatNoir search
engine~\cite{chatnoir} which indexes the entire English part of the ClueWeb09 corpus.
We can now reduce the problem to constructing appropriate queries for ChatNoir search engine.
The goal is to retrieve similar documents, which contains the same passages as the source document
The strongest emphasis has been places on minimizing the number of executed queries
since they may be charged or time limited or number limited in the real world.
-In the PAN 12 test corpus, there were 32 text documents all contained at least one plagiarism case.
-The documents were approximately 30 KB of size, the smallest were 18 KB and the largest were 44 KB.
-We have used methods which are applicable in general to any type of text input with no apriori information about the input document.
-
-All queries from a given document were firstly preconstructed and they were afterwards sequentially executed
-according to their priority. After each executed query the intermediate results of plagiarism
-were calculated using the multi feature document comparison techniques described in section~\ref{yenya}.
-This process leads to continuing lowering the number of executed queries, for more details see the following passages.
+%All queries from a given document were firstly preconstructed and they were afterwards sequentially executed
+%according to their priority. After each executed query the intermediate results of plagiarism
+%were calculated using the multi feature document comparison techniques described in section~\ref{yenya}.
+%This process leads to continuing lowering the number of executed queries, for more details see the following passages.
We can divide constructed queries into three main groups according to their origin:
Thus we firstly filter out stop words according to general English stop words list.
Secondly we used reference corpus word list as an additional filter.
We used words in their simple verbatim occurrence allied to a
-statistical estimate of likelihood. As a result of this a word to become a key
+statistical estimate of likelihood. As a result of this, a word to become a key
must not exists in the stop list and must occur in the suspicious document relatively much more frequently
than in the reference corpus. We used a corpus created from common English web sources, which contains
more then 4 billion tokens~\cite{ententen}.
-The statistical measure TF-IDF was applied in order to decide whether the processed word shall become a keyword.
+The statistical measure TF-IDF~\cite{Introduction_to_information_retrieval}
+was applied in order to decide whether the processed word shall become a keyword.
The TF-IDF combines term frequency (TF) and inverse document frequency (IDF) to produce composite weight of a given term.
Taking into account the average length of the suspicious document and minimizing the number of used queries we set TF-IDF weight threshold to 0.02,
which resulted in about 10 to 20 terms identified as keywords per document.
largest change of AWFC compared to AWFC of their neighbouring previous window were selected as candidate for query extraction.
-The query from the representative sentence were composed by 6 words from the sentence beginning, also omitting stop words.
+The query from the representative sentence was composed by 6 words from the sentence beginning, also omitting stop words.
We chose 6 words to cover higher portion of selected sentence and not to exceed
the phrasal ChatNoir query limit.
This type of query is actually looking for documents containing the very same or at least the very similar sentence,
therefore we choose more than 5 word queries in order to narrow the search to more specific
-results. Unfortunately in the time of PAN 12 competition the ChatNoir search engine did not
+results. Unfortunately in the time of PAN 2012 competition the ChatNoir search engine did not
support phrasal search. Since querying a single sentence is phrasal search, one might have
expected even better results from those queries.
%Due to the fact, that ChatNoir does not currently support phrasal search,
The intrinsic plagiarism based queries are document positionable, meaning that we can store
their position within the document. It is due to the fact, that
-those queries are created from more or less subsequent words. As a result of that they are also executed conditionally, for more
+those queries are created from more or less subsequent words. As a result, they are also executed conditionally, for more
information see section~\ref{process}.
The position within the document were set to the
start position of the representative sentence.
They characterize only a specific portion of the suspicious document on the contrary to keyword based queries.
\subsection{Headers based queries}
-The last type of queries are queries constructed from local headers of document.
-The header usually characterize briefly and accurately following section, thus it
-can be used as a search query. We considered headers only from 2 to 6 words in length.
+The last type of queries were constructed from local headers of the document.
+A header usually characterize briefly and accurately the following section, thus it
+can be used as a search query. In real scenarios the headers should
+be located using some document metadata, in that way we can distinguish for example more header levels.
+In the case of PAN 2012 competition we have detected headers according to several
+basic rules: As a headers based query we considered any line which has an empty line above and bellow the actual line
+and is from 2 to 6 words in length also omitting stop words.
+Those basic rules applied on most headers from given format of test corpus text files.
+
+Note that length of header based queries are variable, thus they can be both specific
+and general according to their length. They are also document positionable.
+We calculated their position as start position of the following text.
+
+\subsection{Combining and executing queries}~\label{process}
+
+All queries from a given document were firstly preconstructed and they were afterwards sequentially executed
+according to their priority.
+The first all keyword based queries, the second
+intrinsic plagiarism based and the last headers based.
+During the process we could omit some of the positionable queries, purposing to
+lower total number of executed queries.
+The condition for their omitting is described further in this section.
+
+
+After each executed query the intermediate results of plagiarism
+were calculated using the multi feature document comparison techniques described in section~\ref{yenya}.
+The queries processing is outlined as Algorithm~\ref{alg:queries}, where for simplicity the snippet calculation and a web source download is omitted.
+After executing a query the intermediate results were always
+calculated as follows:
+
+For every query result in result page based
+on a snippet to input suspicious document similarities, we decided whether to actually
+download the resulting URL or not. Since the snippet contains portions of Web document to each
+given keyword, we calculated pre-similarity as apportionment of every word match between the snipped and the suspicious document.
+If there were at least 80\% match, we downloaded the web source for thorough investigation.
+
+For each downloaded web document we calculated similarities with the input suspicious document
+as a document pair described in section~\ref{yenya}.
+All similarities were stored in form of intervals from within
+the input suspicious document. In other words for every similar part between the downloaded
+document and the suspicious document we stored the beginning position and the ending position of that part in
+the suspicious document.
+
+As a result of that, we can during processing further queries, which haven't been executed yet, omit those
+of which document position intersects with any of already found similarities intervals.
+%This procedure leads us to lowering the total number of executed queries.
+All downloaded documents, which have at least one similarity interval, are declared
+as possible source of the plagiarism.
+
+\renewcommand{\algorithmicrequire}{\textbf{Input:}}
+\renewcommand{\algorithmicensure}{\textbf{Output:}}
-\subsection{Input document processing}~\label{process}
-\subsection{Queries comparison}
+\algsetup{indent=2em}
+\begin{algorithm}[h!]
+\caption{Queries execution}\label{alg:queries}
+\begin{algorithmic}[1]
+\REQUIRE suspicious document $d$, queries $Q_{KW}, Q_{Intrinsic}, Q_{Headers}$
+%keyword, intrinsic and headers based queries $Q_{KW}, Q_{Intrinsic}, Q_{Headers}$,
+\ENSURE plagiarism source candidate web documents $W_{plag}$
+\medskip
+
+\STATE $I \leftarrow \emptyset; I \in \mathbb{N}^2$
+\STATE $W \leftarrow \emptyset$
+
+\FORALL{$q \in (Q_{KW} \cup Q_{Intrinsic} \cup Q_{Headers}) $}
+ \IF{$position(q)$ do not intersects with any interval $ \vec{i} \in I$}
+ \STATE $W \leftarrow execute(q)$
+ \FORALL{$w \in W$}
+ \STATE $I_{sub} \leftarrow get\_similarities(w, d)$ %\COMMENT{output is set of intervals similarities $[d_{start},d_{end}]$}
+ \STATE $I \leftarrow I \cup I_{sub}$
+ \IF{$I_{sub} \neq \emptyset$}
+ \STATE $W_{plag} \leftarrow W_{plag} \cup \{w\}$
+ \ENDIF
+ \ENDFOR
+ \ENDIF
+\ENDFOR
+\RETURN $W_{plag}$
+
+\end{algorithmic}
+\end{algorithm}
+
+\subsection{Queries comparison}~\label{comparison}
+During the test phase there were extracted 133 keywords based queries, 165 intrinsic plagiarism
+based queries and 331 headers based queries in total. Table~\ref{querycount} compares
+results according to query types.
+\begin{center}
+\begin{table}[h]
+\begin{center}
+\begin{tabular}{|c|c|c|c|}
+\hline
+{\bf Query type} & {\bf Extracted}& {\bf Omitted} & {\bf Similarities portion }\\ \hline \hline
+KW & 4.16 & N/A & 72.5 \% \\ \hline
+Intrinsic & 5.16 & 2.35 & 24.3 \% \\ \hline
+Headers & 10.34 & 4.75 & 3.2 \% \\ \hline
+\end{tabular}\\
+\end{center}
+\vspace{0.3 cm}
+\caption{\footnotesize Queries type comparison.}
+\label{querycount}
+\end{table}
+\end{center}
+The second and the third column
+represents the mean of the query count and the omitted query count per document. The fourth
+column shows total portion of similarities found, taking into account the number of similarities regardless of interval sizes.
+ We can see that nearly half of the prepared queries were
+omitted due to the fact, that there had been found a similarity covering their document position.
+We can also see that there were detected about 5 cases
+of potential plagiarism on average, by means of used AWFC intrinsic plagiarism detection method.
+Table~\ref{querycount} also shows keyword based queries as the most successful and
+headers based queries as the least successful. Despite the fact, that they were greatest
+in number they ended with only more than a 3\% of total similarities found. Nevertheless, please
+note that the headers based queries were executed as the last, thus they were used only for
+finding undiscovered potential similarities. In order to really compere the query type performance, we
+would need to execute and evaluate them separately.
+
+To conclude this section we can say, that all types of queries were more or less successful. The headers based
+were executed last and in the process they were the least successful. The interesting
+ finding is the fact, that we can even greatly lower the number of executed queries.
+By omitting all of headers based queries we could lover the total number of executed queries by 45 \% with only
+3.2 \% of recall lost.
+%\begin{center}
+\begin{table}[h]
+\begin{center}
+{ \scriptsize
+\begin{tabular}{c c c c c c c c c c c c }
+\hline
+& \multicolumn{2}{c}{\bf Total workload} & \multicolumn{2}{c}{\bf Time to 1st Result}&\multirow{2}{*}{\parbox{0.6cm}{\bf No \\ result}} &
+\multicolumn{2}{c}{\bf Reported Srcs.} & \multicolumn{2}{c}{\bf Downloaded Srcs.} &
+ \multicolumn{2}{c}{\bf Retrieved Srcs.} \\
+{\bf Team}&{\bf Queries}&{\bf Downloads}&{\bf Queries}&{\bf Downloads}
+& & {\bf Prec.}&{\bf Recall}&{\bf Prec.}&{\bf Recall}&{\bf Prec.}&{\bf Recall}\\ \hline \hline
+
+\parbox{2,3cm}{Gillam et al. \\ University of Surrey, UK} & 63.44 & 527.41 & 4.47 &
+ 25.88 & {\bf 1} & 0.6266 & 0.2493 & 0.0182 & {\bf 0.5567} & 0.0182 & {\bf 0.5567} \\ \hline
+
+\parbox{2,3cm}{ Jayapal \\ University of Sheffield, UK }& 67.06 & 173.47 & 8.78 & 13.50 &
+ 1 & {\bf 0.6582} & {\bf 0.2775} & 0.0709 & 0.4342 & {\bf 0.0698} & 0.4342 \\ \hline
+
+ \parbox{2,3cm}{Kong Leilei \\ Heilongjiang Institute of Technology,\\ China} & 551.06 &
+ 326.66 & 80.59 & 27.47 & 2 & 0.5720 & 0.2351 & 0.0178 & 0.3742 & 0.0141 & 0.3788 \\ \hline
+
+\parbox{2,3cm}{Palkovskii et al. \\ Zhytomyr State University, Ukraine} & 63.13 &
+1026.72 & 27.28 & 318.94 & 6 & 0.4349 & 0.1203 & 0.0025 & 0.2133 & 0.0024 & 0.2133 \\ \hline
+
+\parbox{2,3cm}{\bf our approach}& {\bf 12.56} & {\bf 95.41} & {\bf 1.53} & {\bf 6.28} & 2 & 0.5177 & 0.2087 & {\bf 0.0813} &
+0.3513 & 0.0094 & 0.4519 \\ \hline
+\end{tabular}\\}
+\end{center}
+\vspace{0.3 cm}
+\caption{\footnotesize PAN 2012 candidate document retrieval results.}
+\label{candidateDocsResults}
+\end{table}
+%\end{center}
+
+Table~\ref{candidateDocsResults} shows results of PAN 2012 candidate document retrieval
+task as averages over the all 32 documents from the test corpus. Our approach led
+to obtain decent retrieval performance with very little total workload and time to 1st result.
+Also the 80 \% word match treshold in Web snippet appear to be suitable, since we
+also achieved the highest precision among downloaded sources.
+
+
+
+
https://www.fi.muni.cz/~kas/git / pan12-paper.git / blobdiff